Distribution Planning Guidebook for the Modern Grid

3002011007

10253878
10253878
Distribution Planning Guidebook for the Modern Grid

3002011007
Technical Update, April 2018

EPRI Project Manager

L. Rogers

ELECTRIC POWER RESEARCH INSTITUTE

3420 Hillview Avenue, Palo Alto, California 94304-1338  PO Box 10412, Palo Alto, California 94303-0813  USA
10253878800.313.3774  650.855.2121  askepri@epri.com  www.epri.com
DISCLAIMER OF WARRANTIES AND LIMITATION OF LIABILITIES
THIS DOCUMENT WAS PREPARED BY THE ORGANIZATION(S) NAMED BELOW AS AN ACCOUNT OF
WORK SPONSORED OR COSPONSORED BY THE ELECTRIC POWER RESEARCH INSTITUTE, INC. (EPRI).
NEITHER EPRI, ANY MEMBER OF EPRI, ANY COSPONSOR, THE ORGANIZATION(S) BELOW, NOR ANY
PERSON ACTING ON BEHALF OF ANY OF THEM:
(A) MAKES ANY WARRANTY OR REPRESENTATION WHATSOEVER, EXPRESS OR IMPLIED, (I) WITH
RESPECT TO THE USE OF ANY INFORMATION, APPARATUS, METHOD, PROCESS, OR SIMILAR ITEM
DISCLOSED IN THIS DOCUMENT, INCLUDING MERCHANTABILITY AND FITNESS FOR A PARTICULAR
PURPOSE, OR (II) THAT SUCH USE DOES NOT INFRINGE ON OR INTERFERE WITH PRIVATELY OWNED
RIGHTS, INCLUDING ANY PARTY'S INTELLECTUAL PROPERTY, OR (III) THAT THIS DOCUMENT IS
SUITABLE TO ANY PARTICULAR USER'S CIRCUMSTANCE; OR
(B) ASSUMES RESPONSIBILITY FOR ANY DAMAGES OR OTHER LIABILITY WHATSOEVER (INCLUDING
ANY CONSEQUENTIAL DAMAGES, EVEN IF EPRI OR ANY EPRI REPRESENTATIVE HAS BEEN ADVISED
OF THE POSSIBILITY OF SUCH DAMAGES) RESULTING FROM YOUR SELECTION OR USE OF THIS
DOCUMENT OR ANY INFORMATION, APPARATUS, METHOD, PROCESS, OR SIMILAR ITEM DISCLOSED IN
THIS DOCUMENT.
REFERENCE HEREIN TO ANY SPECIFIC COMMERCIAL PRODUCT, PROCESS, OR SERVICE BY ITS
TRADE NAME, TRADEMARK, MANUFACTURER, OR OTHERWISE, DOES NOT NECESSARILY
CONSTITUTE OR IMPLY ITS ENDORSEMENT, RECOMMENDATION, OR FAVORING BY EPRI.
THE ELECTRIC POWER RESEARCH INSTITUTE (EPRI) PREPARED THIS REPORT.

This is an EPRI Technical Update report. A Technical Update report is intended as an informal report of
continuing research, a meeting, or a topical study. It is not a final EPRI technical report.

NOTE
For further information about EPRI, call the EPRI Customer Assistance Center at 800.313.3774 or
e-mail askepri@epri.com.
Electric Power Research Institute, EPRI, and TOGETHER…SHAPING THE FUTURE OF
ELECTRICITY are registered service marks of the Electric Power Research Institute, Inc.
Copyright © 2018 Electric Power Research Institute, Inc. All rights reserved.

10253878
 ACKNOWLEDGMENTS
The Electric Power Research Institute (EPRI) prepared this report.
Principal Investigators
R. Dugan
J. Taylor
J. Smith
M. Rylander
This report describes research sponsored by EPRI.

This publication is a corporate document that should be cited in the literature in the following
manner:
Distribution Planning Guidebook for the Modern Grid. EPRI, Palo Alto, CA: 2018.
3002011007.

10253878 iii
10253878
 ABSTRACT
The power industry is going through a significant transformation at all levels. In 2010, EPRI
Program 180 published the first edition of the EPRI Underground Distribution Systems
Reference Book (Bronze Book) (report 1019937, with supplemental content in 3002006895,
2016). This book prompted the beginning work on a similar reference capturing the state of the
art in both conventional and advanced distribution planning techniques. EPRI’s planned
comprehensive Reference Book on Distribution Planning (another EPRI “color” book) will fill a
needed gap in practical reference material for distribution planners as they transition from
traditional planning methods to techniques demanded for the modern integrated grid.
This technical update contains the third installment of chapters for the planned reference book.
Nine chapters on current topics of interest have been selected to continue the process of writing
the reference book, which began in 2015, including the following:
1. Distribution System Basics for Planning and Analysis
2. Distribution Power Flow Methods
3. Distribution Planning with Distributed Energy Resources (DER)
4. Reliability Analysis for Distribution Planning
5. Models of Circuit Elements
6. Optimal Recloser/Distribution Automation (DA) Switch Siting
7. Energy Storage Modeling for Distribution Planning
8. Planning for Harmonics
9. Dynamics Simulation for Distribution
These chapters will provide timely reference material on nine topics of current interest. The
material is expected to be particularly useful to EPRI member utilities dealing with issues related
to the modern integrated grid such as renewable DER and state mandates for implementing
energy storage.
Keywords
Distribution planning
Renewable energy
Wind energy
Solar energy
Environmental impacts
EPRI color book

10253878 v
10253878
 CONTENTS
ABSTRACT ..................................................................................................................................V
1 DISTRIBUTION SYSTEM BASICS FOR PLANNING AND ANALYSIS ................................ 1-1
Introduction .......................................................................................................................... 1-1
On Power and Energy .................................................................................................... 1-2
Tools for Distribution Planners ............................................................................................. 1-3
Gaps in Tools ................................................................................................................. 1-4
Distribution System Topologies............................................................................................ 1-4
Why Are Most Distribution Systems Radial?...................................................................... 1-10
A Brief Introduction to Distribution Planning for the Modern Integrated Grid ..................... 1-16
Peak Capacity Planning ............................................................................................... 1-16
Cost Minimization Planning .......................................................................................... 1-16
Risk Analysis ................................................................................................................ 1-17
Market Assessment Methods ....................................................................................... 1-17
Fuzzy Sets ................................................................................................................... 1-17
Other Miscellaneous Planning Methods ....................................................................... 1-18
Typical Distribution Planning Procedure ............................................................................ 1-18
Observations on Traditional Planning Methods with Respect to Planning Systems
with DER and Other Smart Grid Features .................................................................... 1-21
Cost Minimization Planning ................................................................................................ 1-21
Load Forecasting and Loadshapes .............................................................................. 1-22
Cost Minimization Planning: Overall Process............................................................... 1-27
Reliability Costs ............................................................................................................ 1-28
Planning Limits ................................................................................................................... 1-31
Voltage ......................................................................................................................... 1-31
Transformer Rating Limits ............................................................................................ 1-34
On Losses in Planning ....................................................................................................... 1-35
2 DISTRIBUTION POWER FLOW METHODS ......................................................................... 2-1
Introduction .......................................................................................................................... 2-1
The Traditional Power Flow Problem Definition ................................................................... 2-3
Gauss-Seidel Solution Method ....................................................................................... 2-4
Newton-Raphson Solution Method ................................................................................ 2-5
Fast Decoupled Newton-Raphson Solution Method ...................................................... 2-6
The Distribution System Power Flow ................................................................................... 2-6
Radial Circuit Power Flow Calculations ............................................................................... 2-7
Forward-Backward Sweep ............................................................................................. 2-8
Per Units or Actual Values? ......................................................................................... 2-11
Quantities Required for Distribution Power Flow Models ................................................... 2-13
Line Impedance Models ............................................................................................... 2-13
Transformer Data Required ......................................................................................... 2-15

10253878 vii
 Transmission System Source Equivalent....................................................................2-16
Shunt Capacitors ........................................................................................................2-17
Voltage Regulators or Substation OLTC .....................................................................2-19
OpenDSS Power Flow Solution Method ...........................................................................2-22
The Math …................................................................................................................2-25
Y prim Examples ...........................................................................................................2-26
The OpenDSS Network Model ...................................................................................2-27
3 DISTRIBUTION PLANNING FOR DER ................................................................................3-1
Planning Process ...............................................................................................................3-1
Planning Limits for DER .....................................................................................................3-5
Feeder Hosting Capacity ....................................................................................................3-6
Input .............................................................................................................................3-8
Methodology .................................................................................................................3-8
Output ..........................................................................................................................3-9
Mitigation Analysis ......................................................................................................3-11
Energy Analysis ..........................................................................................................3-11
Thermal Capacity Analysis .........................................................................................3-11
Volt-var Control ..........................................................................................................3-11
Capacity Provided by DER .........................................................................................3-12
The “15% Hosting Limit”? ...........................................................................................3-12
Screening Proposed DER Installations for Operational Conflicts ......................................3-13
Reactive Power Control....................................................................................................3-13
Advanced PV Inverter Control ....................................................................................3-14
Capabilities of Planning Tools for DER Planning ..............................................................3-15
Sequential-time Power Flow .......................................................................................3-15
Dynamics Simulation Capability..................................................................................3-18
Unbalanced, Multi-phase Line Models ........................................................................3-19
Transformer Modeling ................................................................................................3-20
Protective Relaying Simulation ...................................................................................3-20
Control Simulation ......................................................................................................3-21
Evaluating Impact of DER on Capacity .............................................................................3-21
Stochastic Analysis ..........................................................................................................3-27
The Analysis ...............................................................................................................3-27
Simulation ..................................................................................................................3-27
Processing Results .....................................................................................................3-27
4 RELIABILITY ANALYSIS FOR DISTRIBUTION PLANNING ...............................................4-1
Reliability Indices ...............................................................................................................4-1
Sustained Interruption Indices ......................................................................................4-2
Indices for Momentary Interruptions .............................................................................4-3
Major Event Days (MED) ..............................................................................................4-4
Power Quality Indices.........................................................................................................4-4

10253878 viii
 Predictive Reliability Analysis .............................................................................................4-5
Analytical Methods in Radial Systems ..........................................................................4-7
Reliability Analysis by Simulation .......................................................................................4-7
Sequential Monte Carlo ................................................................................................4-8
Stochastic Fault Simulation ..........................................................................................4-8
Example Reliability Study By Monte Carlo Analysis............................................................4-9
5 MODELS OF CIRCUIT ELEMENTS .....................................................................................5-1
Introduction ........................................................................................................................5-1
Transformer Models for Distribution Analysis .....................................................................5-1
Estimating Impedances to the Phantom Winding ..........................................................5-5
Line Models ........................................................................................................................5-6
Overhead Lines ............................................................................................................5-6
Concentric Neutral Cables ............................................................................................5-8
Tape Shielded Cable ........................................................................................................5-10
Load Models ....................................................................................................................5-11
Storage Models ................................................................................................................5-11
PV System Models ...........................................................................................................5-16
Regulator Models .............................................................................................................5-21
Capacitor Models .............................................................................................................5-22
Capacitor Controls ......................................................................................................5-23
6 OPTIMAL RECLOSER/DA SWITCH SITING .......................................................................6-1
Objective Functions ............................................................................................................6-2
Assumption of Automated Restoration ...............................................................................6-2
Initialization of Switch Placement Algorithm .......................................................................6-3
Placement of One DA Switch Algorithm .............................................................................6-4
Placement of Two DA Switches at the Same Time Algorithm .............................................6-4
Required Input Data ...........................................................................................................6-5
Algorithm Configurations & Settings ...................................................................................6-5
Output Reports / Charts .....................................................................................................6-5
Visualizing Reliability Benefits ............................................................................................6-8
Example 1 –Analysis of a Planning Area ..........................................................................6-10
Calibration of Fault Rates ...........................................................................................6-10
Locating 1 Switch at a time sequentially ...........................................................................6-12
Locating 2nd Switch sequentially .......................................................................................6-14
Locating 2 Switches at the same time ..............................................................................6-16
Planning Area Analysis ....................................................................................................6-17
Add One Recloser to Existing Configuration ...............................................................6-17
Add Reclosers Two-at-a-Time to Existing Configuration .............................................6-18
Greenfield Analysis.....................................................................................................6-19
7 ENERGY STORAGE MODELING FOR DISTRIBUTION PLANNING ..................................7-1

10253878 ix
 Background ........................................................................................................................7-1
Applications of Storage on Distribution Systems ................................................................7-2
Distribution Planning Issues Introduced By Energy Storage ...............................................7-2
Simulation Modes ...............................................................................................................7-3
Static Analyses...................................................................................................................7-5
Capacity Evaluations ..........................................................................................................7-6
Compensating for Renewable Generation ..........................................................................7-9
Dynamics Simulations ......................................................................................................7-11
On the Complexity of Dynamics Simulation Models with Respect to Distribution
Planning .....................................................................................................................7-13
Electromagnetic Transients Simulation.............................................................................7-14
Vendor-Supplied Model Interfaces for Different Simulation Modes ...................................7-14
Conclusions About Storage Modeling ...............................................................................7-15
8 PLANNING FOR HARMONICS ............................................................................................8-1
Introduction ........................................................................................................................8-1
Harmonics Fundamentals ..................................................................................................8-2
Sources of Current Distortion .............................................................................................8-3
Rms and Power Values ................................................................................................8-7
Determining Capacity with Distorted Currents ....................................................................8-8
Resonance .........................................................................................................................8-9
Series Resonance ............................................................................................................8-12
Parallel Resonance ..........................................................................................................8-13
Situations with Both Series and Parallel Resonance ........................................................8-14
Resonance and Harmonic Distortion Problems ................................................................8-15
Sharpness of Resonance on the Power System...............................................................8-16
Harmonic Current Flow in Distribution Systems................................................................8-17
Modeling Nonlinear Loads When Resonance is Present ..................................................8-19
Norton Equivalent Load Model for Harmonic Analysis ................................................8-20
A More Detailed Load Model ......................................................................................8-21
Modeling Transformer Impedance Variation with Frequency ......................................8-22
Combined Effect of Load and Transformer Modeling........................................................8-22
Avoiding Resonance ........................................................................................................8-24
Harmonic Studies .............................................................................................................8-26
Harmonic Study Procedure ..............................................................................................8-27
Principles for Controlling Harmonics.................................................................................8-28
Modifying System Frequency Response ..........................................................................8-28
Filters ...............................................................................................................................8-29
9 DYNAMICS ...........................................................................................................................9-1
Dynamics Mode vs. Power Flow Modes .............................................................................9-1
Inverter-Based DER ...........................................................................................................9-3
Binary Simulation Model Interface ......................................................................................9-5

10253878 x
 Standard models vs. vendor-specific models ................................................................9-5
APIs for Binary Models .......................................................................................................9-6
EMTP-RV DLL function definitions ...............................................................................9-6
OpenDSS Storage Model DynaDLL Interface ...............................................................9-9
Example: The OpenDSS Dynamics Mode ........................................................................9-11
The Basic Algorithm ...................................................................................................9-11
Going into Dynamics Mode.........................................................................................9-11
Integrating the State Variables ...................................................................................9-12
Predictor and Corrector Steps ....................................................................................9-12
Derivative Calculation (Generator model): ..................................................................9-13
Integration ..................................................................................................................9-13
Circuit Solution in Dynamics Mode .............................................................................9-13
Computing Terminal Currents in Dynamics Mode (Generator) ...................................9-14
Integration for a Power Electronics Model Implementation .........................................9-15
10 REFERENCES AND BIBLIOGRAPHIES .........................................................................10-1
References (1-4) ..............................................................................................................10-1
References (5-8) ..............................................................................................................10-2
Bibliography .....................................................................................................................10-6
Distribution Planning...................................................................................................10-6
Power Flow ................................................................................................................10-9
Modeling ...................................................................................................................10-12

10253878 xi
10253878
 LIST OF FIGURES
Figure 1-1 Schematic diagram of typical North American 4-wire multi-grounded neutral
distribution system .............................................................................................................1-4
Figure 1-2 Schematic diagram of a common European-style distribution system .....................1-6
Figure 1-3 One-line diagram of a North American distribution system, all MV ..........................1-7
Figure 1-4 One-line diagram of a European-style MV and LV distribution system ....................1-8
Figure 1-5 Large urban low-voltage (LV) network in a major city ..............................................1-9
Figure 1-6 One-line diagram of a typical urban LV network ......................................................1-9
Figure 1-7 Fuse time-current characteristic dictates utility fault protection coordination on
radial distribution systems ................................................................................................1-11
Figure 1-8 In series overcurrent protection, fault current is expected from only source and
only one device is required to interrupt the fault current ...................................................1-12
Figure 1-9 On transmission systems, with multiple sources, two devices must operate to
clear the fault....................................................................................................................1-13
Figure 1-10 Typical reclosing sequences used in North American systems ...........................1-14
Figure 1-11 Clearing faults on the primary feeders to urban LV networks requires operation
of multiple devices ............................................................................................................1-15
Figure 1-12 Representing a load as a fuzzy set .....................................................................1-18
Figure 1-13 Which slope is chosen for a temperature-normalized forecast of substation
load? ................................................................................................................................1-22
Figure 1-14 The maximum daily demand may lag the maximum daily temperature ...............1-23
Figure 1-15 Load and temperature vs time for two substations ..............................................1-24
Figure 1-16 Daily peak load vs. max temperature ..................................................................1-24
Figure 1-17 Performing a load forecast with weather-adjusted linear regression ....................1-25
Figure 1-18 Yearly loadshape for the distribution planning area .............................................1-26
Figure 1-19 Overall process for Cost Minimization Planning ..................................................1-27
Figure 1-20 Cost functions for different growth scenarios .......................................................1-28
Figure 1-21 Illustrating the computation of energy exceeding normal (EEN) and unserved
energy (UE) for a daily load simulation for two instances of load growth ..........................1-29
Figure 1-22 Comparison of cost minimization planning to peak capacity planning assumed
cost functions ...................................................................................................................1-30
Figure 1-23 Example EEN characteristic for a summer peaking system ................................1-31
Figure 1-24 Normal minimum load voltage profile -- no DER..................................................1-32
Figure 1-25 Minimum-load feeder voltage profile with large amount of DER ..........................1-33
Figure 1-26 Adjusting the last regulator to provide more headroom for DER output ...............1-33
Figure 1-27 Substation transformer temperatures for a typical daily loadshape......................1-34
Figure 2-1 Common unbalances in the distribution system that require more than a positive-
sequence model .................................................................................................................2-2
Figure 2-2 System model for conventional power flow problem definition.................................2-3
Figure 2-3 Distribution power flow problem ..............................................................................2-7
Figure 2-4 Current injection model for representing loads and sources ....................................2-8
Figure 2-5 Illustrating the common forward-backward sweep method for solving radial
circuits ................................................................................................................................2-8
Figure 2-6 What is the voltage base for the LV side that would allow removing the explicit
transformer model? ..........................................................................................................2-11
Figure 2-7. Balanced, symmetrical impedance matrix constructed from symmetrical
component values Z 1 and Z 0 ............................................................................................2-13
Figure 2-8 4-wire line model constructed from symmetrical component impedances .............2-15
Figure 2-9 Bulk power system source modeled as a 3-phase short-circuit equivalent ............2-17

10253878 xiii
Figure 2-10 Capacitor control monitoring load side voltages and currents..............................2-19
Figure 2-11 Bank of three 1-phase voltage regulators deployed as feeder regulators
(Courtesy of Eaton, used by permission) ..........................................................................2-20
Figure 2-12 OpenDSS bus and node definitions ....................................................................2-23
Figure 2-13 Loads and other power conversion elements are typically modeled as being in
shunt with the power system and have one or more conductors that are connected to
nodes at buses .................................................................................................................2-24
Figure 2-14 Primitive Y matrix for a simple resistor ................................................................2-26
Figure 2-15 Primitive Y matrix for a pi-section line model .......................................................2-27
Figure 2-16 OpenDSS network model ....................................................................................2-28
Figure 2-17 Basic concept for modeling nonlinear power conversion elements like loads ......2-28
Figure 2-18 OpenDSS solution process illustrated .................................................................2-29
Figure 2-19 The standard OpenDSS load models switch to a linear model when voltage is
outside normal limits.........................................................................................................2-30
Figure 3-1 Typical daily load profiles -- 3-day period ................................................................3-2
Figure 3-2 Voltage Profile at Peak Load ...................................................................................3-3
Figure 3-3 Voltage Profile at 40% of Peak Load .......................................................................3-4
Figure 3-4 Distribution Analysis Components for Integrated Grid Assessment .........................3-7
Figure 3-5 Hosting Capacity Determination Factors .................................................................3-8
Figure 3-6 Streamlined Hosting Capacity Methodology ............................................................3-9
Figure 3-7 Cumulative Distribution of Non-optimally Sited Centralized DER Hosting Capacity
for 185 12-kV Distribution Feeders ...................................................................................3-10
Figure 3-8 Maximum and Minimum Centralized DER Hosting Capacity for 185 12kV
Distribution Feeders .........................................................................................................3-10
Figure 3-9 Impact of Interconnection Transformer on Apparent Power Factor .......................3-14
Figure 3-10. Volt-var Control Characteristic Proposed for Inverters.......................................3-15
Figure 3-11 Typical Distribution Feeder Annual Load Profile ..................................................3-16
Figure 3-12 Energy Exceeding Normal Planning Limits..........................................................3-17
Figure 3-13 Voltage Fluctuations Due to Cloud Transients ....................................................3-18
Figure 3-14 Open-Conductor Simulation Result, Delta-Y Transformer ...................................3-19
Figure 3-15 Residential Service Transformer, North America.................................................3-20
Figure 3-16 OpenDSS Solution Loop with Control Modeling ..................................................3-21
Figure 3-17 Amount of Additional Load Served Depends on Location....................................3-22
Figure 3-18 Example Impact of Solar PV Generation on Peak Substation Demand ...............3-23
Figure 3-19 Example Impact of Solar PV Generation on Peak Substation Demand ...............3-23
Figure 3-20 Tabulating Energy Exceeding Normal (EEN) Rating ...........................................3-24
Figure 3-21 EEN Plot for Summer Peaking System ...............................................................3-25
Figure 3-22 Example Incremental Capacity Calculation for a 2 MW CHP Unit Using EEN .....3-26
Figure 3-23 Example Incremental Capacity Calculation for 20 MW Solar PV Using EEN .......3-26
Figure 3-24 Maximum Primary and Secondary Voltage Deviations for Each PV Deployment
Scenario ...........................................................................................................................3-28
Figure 3-25 Total Number of PV Deployment Scenarios Exceeding 1% Deviation Threshold 3-28
Figure 3-26 PV Deployment Characteristics that Cause > 1% Primary Voltage Deviation ......3-29
Figure 4-1 Circuit Diagram for Case Study ...............................................................................4-9
Figure 4-2 Distribution of Voltages Observed at 120-V Loads for 1000 Random SLG Faults .4-10
Figure 4-3 Expected Distrribution of Voltages at a 480V Bus .................................................4-11
Figure 4-4 Expected Distribution of Voltages at a 208 V Bus .................................................4-12
Figure 5-1 Transformation of short circuit impedances into the primitive Y matrix for a
transformer.........................................................................................................................5-2
Figure 5-2 Three-phase Wye-Delta transformer schematic ......................................................5-3
Figure 5-3 The 3 short circuit measurements required to model a wye-delta-wye transformer .5-4

10253878 xiv
Figure 5-4 Three-legged core transformers offer a low impedance to zero sequence currents
because the flux path has a large gap outside the steel. ....................................................5-4
Figure 5-5 The effect of a 3-legged core is like having an extra “phantom” delta winding .........5-5
Figure 5-6 A 4-winding transformer requires 6 short circuit measurements (matrix is
symmetrical; lower triangle shown) .....................................................................................5-5
Figure 5-7 Typical construction of a 3-winding core form power transformer ............................5-6
Figure 5-8 Overhead line geometry for a horizontal crossarm construction for a 4-wired
multi-grounded neutral system [64] ....................................................................................5-6
Figure 5-9 Neutral-to-earth voltage (NEV) test feeder requiring modeling of pole downlead
impedances and grounding resistances [63] ......................................................................5-8
Figure 5-10 Structure of a concentric neutral cable: a) cutaway view b) simplified cross
section (not in proportion) ...................................................................................................5-9
Figure 5-11 Structure of a tape shielded cable: a) cutaway view b) simplified cross section
(not in proportion) .............................................................................................................5-10
Figure 5-12 Generic storage model used in the EPRI OpenDSS program .............................5-13
Figure 5-13 Storage controller model from EPRI OpenDSS program .....................................5-13
Figure 5-14 EPRI OpenDSS generic PVSystem model. .........................................................5-17
Figure 5-15 Solar PV simulation at a 1-hour step ...................................................................5-17
Figure 5-16 Actual 1-second solar PV data with variations caused by cloud transients. .........5-18
Figure 5-17 A possible impact on feeder voltages of the solar PV generation in Figure 5-16. 5-18
Figure 5-18 Typical i-V curves for values of irradiance ...........................................................5-20
Figure 5-19 Typical power vs temperature curve....................................................................5-20
Figure 5-20 Typical inverter efficiency curve. .........................................................................5-20
Figure 6-1 Overview of the Test Bed for Developing the Optimal Siting Algorithm ...................6-1
Figure 6-2 For Assumed Automatic Restoration Downline Section are Assumed to be
Restored by Automatically Closing Normally-Open Ties .....................................................6-3
Figure 6-3 IEEE 8500-Node Test Feeder with Two Recloser Sited. Restoration Assumed. .....6-6
Figure 6-4 Value of Customer Interruptions Plot for Adding First Device ..................................6-7
Figure 6-5 Value of Customer Interruptions Plot for Adding 2nd Device ...................................6-8
Figure 6-6 Illustrating Number of Customer Interruptions Prior to Adding Recloser/DA Switch.6-9
Figure 6-7 Illustrating Number of Customer interruptions After Adding Optionally-Sited
Recloser/DA Switch............................................................................................................6-9
Figure 6-8 18-Feeder Planning Area with Existing Recloser Locations Shown .......................6-10
Figure 6-9 Plot of Accumulated Number of Customer Interruptions for All Feeders ................6-12
Figure 6-10 A Sample Feeder showing fused laterals ............................................................6-13
Figure 6-11 Graph of Customer Interruptions Saved vs. Distance from Substation ................6-13
Figure 6-12 Graphical Display of Optimal Recloser Location..................................................6-14
Figure 6-13 Locating the 2nd Switch - Graph of Customer Interruptions Saved .....................6-15
Figure 6-14 Graphical Display of Second Optimal Recloser Location.....................................6-15
Figure 6-15 Customer Interruptions Saved - 2 DA Switches at a Time ...................................6-16
Figure 6-16 Placement of Two DA Switches (2 at a Time left) vs. (1 at a Time right) .............6-17
Figure 7-1 Basic Concept of the EPRI OpenDSS STORAGE Model ........................................7-5
Figure 7-2 Using storage for daily peak shaving.......................................................................7-7
Figure 7-3 OpenDSS Storage Controller Concept ....................................................................7-9
Figure 7-4 Using storage to shift PV generation .......................................................................7-9
Figure 7-5 Smoothing variations in PV generation .................................................................7-10
Figure 7-6 Power output smoothing operation ........................................................................7-11
Figure 7-7 6-second DESS simulation ...................................................................................7-12
Figure 7-8 Layout of an ultracapacitor-based Distributed Energy Storage System (EDF) .......7-13
Figure 8-1 Voltage Distortion is Due to Distorted Current Passing Through the System
Impedance .........................................................................................................................8-3

10253878 xv
Figure 8-2 Current waveform from a variable-frequency drive and its corresponding
harmonic spectrum. ............................................................................................................8-5
Figure 8-3 Schematic of a 3-phase pulse-width modulated variable-frequency drive. [18]........8-5
Figure 8-4 Replace the harmonic-producing device with a current source in the model. ..........8-6
Figure 8-5 The Components Of Power For Non-Sinusoidal Currents .......................................8-9
Figure 8-6 Simulation of current in a Capacitor Bank in an Industrial Power System that is in
11th-Harmonic Resonance ...............................................................................................8-10
Figure 8-7 Simulation of a System Going into 5th-Harmonic Resonance when a Capacitor
Bank is Energized; Voltage across Capacitor and Current in the Capacitor. .....................8-11
Figure 8-8 Series Resonant Circuit ........................................................................................8-12
Figure 8-9 Parallel Resonant Circuit.......................................................................................8-13
Figure 8-10 A Common Power System Configuration that Can Appear Either Series
Resonance or Parallel Resonance Depending on Perspective .........................................8-14
Figure 8-11 Illustrating the effect of increasing the apparent resistance (decreasing the Q) of
a resonant circuit. .............................................................................................................8-17
Figure 8-12 Magnification factor: Amps through transformer per amp injected. ......................8-17
Figure 8-13 Harmonic currents from sources of harmonic distortion tend to flow toward the
utility source. ....................................................................................................................8-18
Figure 8-14 Parallel resonance at the substation magnifies the current injected into the
power system ...................................................................................................................8-18
Figure 8-15 The normal flow of harmonic currents are altered by series resonance from
power factor correction capacitors. ...................................................................................8-19
Figure 8-16 Replacing Simple Current Source Models with Thevenin or Norton Equivalents
to Get Better Answers for Simulations at Resonant Frequencies......................................8-19
Figure 8-17 Simple Norton Equivalent Model of a Load for Harmonics Analysis ....................8-20
Figure 8-18 Load model with both series and parallel branches .............................................8-21
Figure 8-19 Comparing the impact of different load and transformer modeling assumptions
on resonance ...................................................................................................................8-23
Figure 8-20 One-Line Diagram of the OpenDSS REACTOR Object.......................................8-24
Figure 8-21 A simple circuit that can be analyzed by manual calculations..............................8-25
Figure 8-22 Positive-Sequence frequency scan of all possible capacitor configurations with
the EPRI Grid-IQ Harmonics Evaluation Module. .............................................................8-26
Figure 8-23 Common harmonic filter topologies .....................................................................8-29
Figure 8-24 Converting an existing capacitor bank to a single-tuned filter ..............................8-30
Figure 8-25 C Filter configuration. ..........................................................................................8-31
Figure 8-26 C-Filter Characteristic .........................................................................................8-31
Figure 8-27 Broadband Filter Schematic ................................................................................8-32
Figure 8-28 A Broadband Filter Characteristic........................................................................8-32
Figure 9-1 Generator model for power flow analysis ................................................................9-1
Figure 9-2 Simple single-mass generator model for dynamics analysis ...................................9-2
Figure 9-3 Part of the block diagram of an ultracapacitor-based distributed energy storage
system (EDF) [73] ..............................................................................................................9-4
Figure 9-4 Default model of electrical part of generator object in dynamics mode ..................9-14

10253878 xvi
 LIST OF TABLES
Table 1-1 Comparison of typical North American and European-style distribution systems ......1-8
Table 1-2 Typical distribution capacity solutions considered (ascending order of cost) .........1-20
Table 2-1 Comparison of transmission system analysis to distribution system analysis ...........2-2
Table 3-1 Summary of Typical Planning Limits for Distribution-Connected DER ......................3-5
Table 4-1 Annual SARFIx Results ..........................................................................................4-12
Table 5-1 Typical model properties required for modeling storage elements for planning
analysis ............................................................................................................................5-14
Table 5-2 Typical data for specifying a PV system for planning analysis ................................5-19
Table 5-3 Typical data required for simulation of a regulator or OLTC control ........................5-21
Table 5-4 Example regulator impedances (%) on through-kVA base (Source: Siemens-Allis
Regulator Engineering Manual, 1978) ..............................................................................5-22
Table 5-5 Data typically required to define a capacitor bank for analysis ...............................5-23
Table 5-6 Typical capacitor control settings used in QSTS simulations for planning studies ..5-24
Table 6-1 System Reliability Indices Computed for 2nd Recloser Configuration.......................6-8
Table 6-2 Actual Permanent Fault Data Provided for 9 of the 18 Feeders in the Planning
Area .................................................................................................................................6-11
Table 6-3 Reliability Indices Computed for the Feeders - Adding One DA Switch ..................6-18
Table 6-4 Reliability Indices Computed for the Feeders - Adding 2 DA Switches
Simultaneously .................................................................................................................6-18
Table 6-5 Reliability Indices Computed for the Feeders – Greenfield Analysis .......................6-19
Table 9-1 The eight key functions in the OpenDSS storage model DynaDLL API ..................9-10

10253878 xvii
10253878
1
DISTRIBUTION SYSTEM BASICS FOR PLANNING AND
ANALYSIS
Introduction
This chapter describes the basic characteristics of electric power distribution systems that
influence how planning and analysis is done. This is important to understand because the
distribution system is changing from strictly delivering power from a bulk power system to
numerous consumers to a system that exchanges power between various consumers and
producers on the distribution system and transmission system. Planning techniques for evaluating
the options for the modern grid are presented and compared with traditional planning techniques.
Distribution planners are generally focused on providing sufficient power delivery capacity to
meet the maximum (peak) power demand of consumer load. This is to be provided as
economically as possible with adequate reliability to satisfy the customers. Electric power
utilities are overseen by regulatory agencies that provide guidance for what constitutes
reasonable cost and acceptable reliability.
The capacity of the distribution system in terms of technical design issues has both voltage and
current-carrying aspects. The consumer expects that the power will be provided at a standard
voltage that is maintained within acceptable bounds of magnitude and frequency, and is
acceptably stable. The voltage should be compatible with the load equipment ratings the
consumer is able to purchase and connect to the power system. What is commonly referred to as
the power quality is mostly related to the quality of the voltage supplied.
The current-carrying capabilities are basically related to the thermal characteristics of the
conductors in transformers and power lines and cables. When current passes through wires there
are losses that create heat that must be dissipated. Thermal issues limit many power distribution
applications. This is an energy issue because the temperature of the conductor depends on how
long a particular amount of current has been flowing through the conductor. Thus, technologies
applied in the modern grid that disrupt the traditional shape of the load current over time will
also disrupt the traditional planning guidelines for thermal capacity. This also has a correlation to
the impedances of the power delivery system, which ties the current-carrying capacity with the
voltage regulation issues.
The economics and reliability issues factor strongly into the structure of the distribution system.
Meshed networks are generally more reliable than radial circuits, but are significantly more
expensive to build and operate. Underground cable systems are generally more reliable than
overhead lines more exposed to the weather, but can be an order of magnitude more costly to
build. A major challenge to the distribution planner is designing a system that meets the
expectations of the local consumers for reliable power at a cost they are willing to pay.
Most power distribution systems have been designed and optimized to deliver power in one
direction – from the bulk power, or transmission, system to the consumer. The voltage regulation
is designed with this assumption as well as the protection system for clearing faults. The changes

10253878 1-1
brought on by modern grid developments alter the assumption of power flow in one direction
and alter the daily load shape. This can change how the system is designed. Here are some
examples:
• The traditional voltage regulation design is based on the assumption that the power flow will
cause a voltage drop as one moves from the substation toward the load. Power sources on the
distribution system may cause voltage rise. This is one of the key limiting factors in how
much solar PV generation can be accommodated on a distribution feeder.
• Planning standards such as capacity limits based on thermal limits of transformers, overhead
lines, and underground cables have been established based on decades of experience with
how load varies over a day. Distributed sources on the distribution system can significantly
alter the daily shape requiring a revised way of computing the current-carrying capacities.
• Modern distribution grids containing distributed energy resources (DER) and such things as
microgrids may require revised protection schemes for preventing damage from overcurrent
and overvoltages.
On Power and Energy
In April 2008, the IEEE Power Engineering Society (PES) changed its name to Power and
Energy Society (still PES) to emphasize the difference between power and energy. These two
terms are often used interchangeably in non-scientific media and in casual conversation, even by
engineers. People who are not trained in science often do not understand the difference and are at
a disadvantage in understanding many of the issues related to the modern grid.
A typical example that has an impact on distribution planning is a newspaper article on a new
solar PV installation that is claimed to produce enough electricity to serve a certain number of
homes. What is meant is that over a day’s time it can produce enough energy to serve the total
amount of energy that the homes consume during a day, but the issue that the solar PV
generation cannot produce any power at night when the sun is not shining is not mentioned. A
similar issue arises with other forms of renewable energy with the difference being the time of
day when the resource in unavailable. Wind generation is typically diurnal, producing strongly
twice a day with dead periods in between.
The distribution planner’s main interest has traditionally been to compute the power (in kW or
MW) because the distribution system must have the capacity to deliver the required amount of
power drawn by the load. The main computer tool used by the planner is called a power flow
program. With the advent of storage to compensate for the variable production of renewable
generation, planners are now becoming more aware of the need to consider energy. The
consumer taps into the available power stream as needed over time to convert the instantaneous
power into energy. The major portion of the cost of electricity to the consumer is the amount of
energy (kWh or MWh) consumed. Some consumers also are billed for a demand charge, which
reflects the cost to provide the capacity for the consumer’s peak kW demand. Some localities
have begun to tack on a capacity-related charge for rooftop solar PV generation in recognition of
the cost of providing power delivery capacity to the site.

10253878 1-2
Tools for Distribution Planners
The principal computer tool used by distribution planners is a power flow program. In
commercial packages it may be supported by a graphical user interface and a database containing
a model of circuit, load data, etc. There are two fundamental classes of power flow tools based
on the structure of the distribution system they were designed for:
1. Radial circuit only, and
2. Meshed networks.
Power flow solvers for meshed networks can also generally solve radial circuits, but not always.
Some of the network tools do not work well for systems with low X/R ratio, which would
include most distribution systems.
Power flow models for distribution system analysis have typically been full 3-phase unbalanced
models since about 1990. In contrast, power flow applications for transmission systems typically
assume a balanced system and employ a single-phase positive-sequence equivalent of the 3-
phase system. The transmission system power flow programs also work in the per-unit system,
eliminating the detailed transformer model. Most distribution system analysis power flow
program work in actual volts, ampere, and ohms with transformers modeled explicitly.
The second principal planning tool is a short circuit current calculation application. It may be
embedded with the power flow tool or implemented as a separate program that may or may not
work from the same circuit database. The results of the short circuit program are used to
determine settings for the overcurrent protection devices on the system: fuses, reclosers, and
breaker relays. Overcurrent protection has been the main technology employed by utilities to
protect against the damage caused by short-circuit faults. An underlying assumption in
overcurrent protection is that fault currents flow only in one direction. This increasing presence
of DER on distribution systems is altering that assumption.
Other common computer applications used in distribution planning include:
• Overcurrent protection coordination programs based on time-current characteristic (TCC)
curves
• Line constants (for computing line and cable impedances)
• Load forecasting
• Quasi-static time-series (QSTS) simulation. Also called long-term dynamics.
• Motor starting
• Arc flash
• Harmonics analysis
• Optimal recloser placement
• Optimal voltage regulator placement
It is common now for the main power flow application to be scriptable, or customizable, by the
users. It is common to provide a Python-language interface for the scripting. Some tools also
provide a Component Object Model (COM) interface to a user-written program for custom
applications.

10253878 1-3
Gaps in Tools
While tools to support distribution planning have made significant advancements in the last
decade – largely due to grid modernization efforts – there remain several gaps:
• Analysis of storage.
• Modeling inverter controllers for power flow and short circuit contribution.
• Short-term dynamics, transient stability, etc.
• Modeling advanced distribution management systems (ADMS).
• Distribution state estimation (DSE).
• Modeling microgrids.
• Modeling edge-of-grid voltage controllers.
While each of these topics can be addressed by various general-purpose simulations tools such as
MATLAB and EMTP-RV, they are not addressed in the tools that distribution planners typically
have at their disposal or have time to use. The simulation tools are either too costly to be applied
in sufficient numbers to be practical in a planning group or require modeling skills not present in
typical planning groups. Many of the research efforts at EPRI and other organizations are
currently aimed at filling these gaps.

Distribution System Topologies

Distribution system topologies vary around the world. The typical North American 4-wire multi-
grounded neutral system depicted in Figure 1-1 can be one of the more complicated system
topologies to model.

Figure 1-1
Schematic diagram of typical North American 4-wire multi-grounded neutral distribution system

The substation transformer is typically connected delta/wye-grounded as shown. Following the

ANSI standards for mixed-winding transformers, the vector group would be Dy1 (low voltage
lags the high voltage by 30 degrees). In utility applications the neutral on the wye side is either
solidly grounded, or is low-impedance grounded if it is necessary to limit short-circuit currents.
There are also some areas of North America where 3-winding wye-delta-wye transformers are
used in HV/MV distribution substations. The delta winding may be either buried or connected to
station load and/or shunt capacitors or reactors. These are usually employed where the
transmission grid is weaker and needs the grounding transformer effect on the HV side to keep
unfaulted-phase overvoltages in check so that the transmission-side arresters survive during

10253878 1-4
faults and single-phasing conditions. Some utilities employ the 3-winding transformers simply
because they believe such transformers are more robust or are needed to provide polarizing for
directional overcurrent relaying.
Residential loads are served from single-phase transformers with a center-tapped LV winding
that supplies both 120- and 240-V loads. Some refer to this as a “split-phase” connection. The
primary of the transformer is usually connected line-to-neutral, although the line-to-line
connection is also commonly used. This transformer is difficult to model in detail in some
distribution planning tools. Fortunately, a detailed model is not necessary in most planning
analyses. It is necessary when for evaluating the impact of certain types of DER connected to the
LV side.
There are numerous capacitor banks applied on North American distribution systems for power
factor correction and voltage regulation. Most are connected grounded-wye as shown but a
significant percentage are ungrounded-wye to nominally block the flow of triplen harmonic
currents. Capacitors may be fixed (always on), switched daily with an automatic switch, or
switched seasonally with a manual switch. Capacitors are also applied on 1- and 2-phase feeders.
Industrial and commercial 3-phase loads are served by one or more 3-phase distribution
transformers. The winding connections of these transformer vary, in contrast to what is found
elsewhere in the world. There is an old saying among distribution engineers that “if you can
imagine it, someone has done it.” The most common connections are grounded-wye/grounded-
wye (Y n yn ) and delta/grounded-wye (Dy1).
This is another area where the modern grid comes into play. Several large generation
installations are connected grounded-wye/delta (Ynd1) to help control overvoltages during
islanding and fault events. This requires careful engineering of the system because this
transformer connection becomes an active participant in all ground faults that occur on the
system. This connection can also create thermal problems for the transformer itself. Some
reactance in the neutral on the Y side may be required to enable the survival of the transformer
for unbalanced situations that normally occur on the typical North American feeder.
The grounded-wye/grounded-wye connection is the most common today, particularly when
underground cables are used to supply the system. This is to reduce the chances of having
damaging ferroresonance occur when the transformer is disconnected by either pulling cable
elbows or manually opening the fused cutout switches on the riser pole to the overhead system.
A transformer with a delta primary would be very susceptible to ferroresonance in this case.
Tapchanging voltage regulators are not shown, but are common on longer distribution feeders.
They are typically applied in banks of 1-phase units when placed on the feeders. Three-phase
OLTC are common in the substation transformer.
This kind of system is employed in most of North America with key exceptions. Unigrounded or
Delta 3-wire MV systems are also common on the West Coast of the US for 15-kV class systems
and below. The typical HV/MV substation transformer is Dy1 like the 4-wire multi-grounded
neutral systems, but the neutral is grounded only in the substation. No neutral wire is carried
along with overhead lines, but is carried with underground cable systems out of necessity to
apply uniform dielectric stress on the cable insulation. MV/LV distribution transformer are not
connected to neutral on the MV (primary) side in this kind of system. Single-phase residential

10253878 1-5
service transformers are connected line-to-line on the primary; 3-phase transformers are
connected in delta on the primary.

Figure 1-2
Schematic diagram of a common European-style distribution system

Figure 1-2 depicts the schematic diagram for distribution systems influenced by European
designs. The HV/MV substation transformer is shown as delta/wye, but there are also many
wye/delta transformers with a zig-zag transformer on the delta side to provide grounding.
The winding connection of delta/wye transformers in European style systems are commonly
classified in vector group Dy11. The number “11” denotes that the y side (LV side) leads the D
side by 30 degrees, the opposite of the ANSI standard connection. Many European systems
employ high-impedance neutral grounding to minimize the impact of single-line-to-ground
(SLG) faults on the consumers served off the LV system. Some size the neutral reactor to
resonate with the zero-sequence capacitance of the feeder. This is called a Petersen Coil. With
perfect resonant grounding temporary SLG faults will self-extinguish and no breaker operation
occurs. Some Petersen coil devices will automatically adjust the tuning as the topology of the
MV feeder changes due to switching or failure.
This is in contrast to the North American 4-wire multi-grounded neutral system that requires the
fault current to be interrupted so that temporary faults can be cleared.
The MV/LV distribution transformers are nearly all Dy11 connections. The LV service voltage is
nominally 400 V three-phase (230 V to neutral). Single-phase loads are rated 230 V and are
connected line-to-neutral. The LV system is a 3-phase, 4-wire, multi-grounded neutral system.
The neutral point at the distribution transformer is solidly grounded. It is operated similarly to
the North American MV system, although the length is obviously shorter because the voltage is
lower.
The MV/LV transformer size ranges from 225 kVA to about 600 kVA for residential LV feeders.
There may be 100 homes on the feeder, many of which will have 3-phase service in some
countries. In the North American system, the distribution transformers are typically 25 to 50

10253878 1-6
kVA and serve up to 4 or 5 houses each. The secondary service drops are limited to much shorter
lengths due to the low 120-V service voltage for lighting circuits and smaller loads.
Many larger-size 3-phase DER devices may be connected directly to the 400-V systems in areas
that employ the European-style design. In North America, rooftop PV systems are generally
connected to the residential service at 240 V, although some employ small 120-V inverters.
Larger-sized 3-phase DER designed for 460 V connection will generally require a separate
distribution transformer unless it is connected at a commercial facility that already has service at
this voltage.
Figure 1-3 and Figure 1-4 compare the topologies of the two types of distribution systems. In the
North American system the MV is extensive and comes to within 100 ft (30 m) of the load due to
the voltage limitations. To add a new load, the MV lines must generally be extended and a new
MV/LV pole transformer or padmounted transformer added to provide electrical service. The
LV system in this kind of system is small and the topology is not complex.

= Normally-closed switch or line recloser = Normally-open tie switch

Figure 1-3
One-line diagram of a North American distribution system, all MV

A European style MV system has a much cleaner look because few loads, if any, are served
directly from it. The LV system is quite extensive, going house-to-house throughout the
neighborhoods and down the streets. To add a new load, in most cases one can simply add more
wire to the 400-V system. It is not necessary to add a new MV/LV transformer very often.
International agencies that provide funding for electrification of developing nations often prefer
the European-style system because it requires less capital to build. There are few distribution
transformers and less of the more expensive MV equipment to purchase and install. LV wire is
relatively inexpensive and takes less skill to install. Actually, the latter is one of the drawbacks of
this kind of system in many areas of the world. It can be installed and extended easily by
stringing LV wire, but illegal connections can be made by the same process. One reason theft is a
smaller problem in the North American system is the specialized equipment required to connect
illegal taps safely.
Capacitor banks are found less frequently on the MV system in European-style systems. Some
utilities apply MV capacitors only in the substation, connected ungrounded wye. Power factor
correction, if done at all, is done on the LV system. Likewise, voltage regulator banks like those

10253878 1-7
found in American systems are seldom found on European style MV systems. Tapchangers for
voltage regulation are applied at the 400-V level in the MV/LV transformer. [24] This kind of
transformer was introduced to compensate for the voltage variations on LV systems caused by
high penetration solar PV generation.

MV

LV

Figure 1-4
One-line diagram of a European-style MV and LV distribution system

Table 1-1
Comparison of typical North American and European-style distribution systems

North American Style System European Style System

MV, or Primary, system is extensive, complex
LV, or Secondary, system is short
4-8 homes per distribution transformer, 25-50
kVA in size. 120/240 V “split phase” service
Extended by adding distribution transformer and
MV lines
MV system has simpler, more consistent structure
400-V LV system is extensive
Could have100 homes on one MV/LV 3-phase Dy11
transformer, 225 kVA and higher. Some residences will have
400/230 V 3-phase service
Extended by adding 400 V lines; fewer transformers

Another kind of distribution system is employed in the downtown areas of large cities. It is
referred to generically as an “urban LV network” system. Figure 1-5 and Figure 1-6 give two
views of this kind of distribution network. It consists of a grid of LV lines supplied by a number
of MV/LV network vault transformers that are supplied by 4, or more, MV feeders.

10253878 1-8
 138 kV Transmission Supply

26.4 kV Distribution

FEEDERS

LOAD

LOW-VOLTAGE GRID NETWORK

(MESHED)

Figure 1-5
Large urban low-voltage (LV) network in a major city

SUBSTATION
PRIMARY FEEDERS

TRANSMISSION
SYSTEM

FEEDER BREAKER
OR RECLOSER
LOW-VOLTAGE NETWORK
NETWORK
PROTECTOR

Figure 1-6
One-line diagram of a typical urban LV network

10253878 1-9
Urban LV network systems are used in downtown areas of large cities requiring extraordinary
reliability. They are found in many large cities, such as New York, Seattle, and Chicago in the
US. The networks in these cities differ slightly, but the general concept is the same. These
networks can be quite large with the number of 3-phase buses exceeding 10,000 and the total
load being up to 300 MW. Smaller networks of similar design may serve a campus of
government offices or a university. Many of the networks are designed to continue serving the
load without interruption after suffering two failures, or more.
The network transformers are all generally connected delta/wye-grounded. The 3-phase LV
network is commonly 120/208 V in the US. There are also 277/480 V networks. Common
network transformer sizes are 500 kVA and 1000 kVA with overload capacities of
approximately 1.5 times the nameplate rating depending on the ability to get the heat out of the
underground vault. Fault currents on the LV networks can exceed 30 kA.
Modern grid initiatives usually involve some kind of DER or advanced control. However, it can
be more difficult to coordinate the operation of the DER with the operation and protection of the
LV network. It is also difficult to achieve local benefit from DER installed on the LV network
because the power output dissipates into the myriad junctions in the network. However, there
could be benefits upstream to the transmission system by reducing the demand in time of
constrained capacity and high load demand. So this presents planning challenges for which
research efforts are active and continuing.
The power flow tools the planners use to study urban LV networks must be capable of modeling
large, highly meshed networks. Fortunately, these 3-phase systems are usually sufficiently
balanced that positive-sequence equivalents can be made. Planners in utilities that operate these
networks often have customized algorithms in the software for their system to estimate the
loading. For example, there are many unmetered LV lines that may, or may not, be in service at
any given time. The only knowns are the flows at the network transformers. Empirical
adjustments are made to the estimates of the loading to achieve a flow distribution in the network
that approximates the measurements at the network transformers.
The reliability of an urban LV network is on the order of 100 times better than the common
radial distribution systems that serve most other areas. Urban LV networks are much more costly
to build and use devices not found on other kinds of distribution systems such as network
transformers and network protectors. The cost is justifiable due to the high value of the load in a
densely-populated area and critical customers such as a major financial district with its
computers - and air conditioning to keep the computers and their operators cool and comfortable.
Most MV distribution systems in the world are configured radially from the HV/MV substation.
Transmission systems (HV) are mesh networks for reliability and for connecting to multiple
generating facilities. However, the meshing is not as fine as many urban LV networks.

Why Are Most Distribution Systems Radial?

The basic answer to this question can be summed up in one word: Cost. All distribution systems
will suffer the occasional short circuit fault, so all distribution systems must have adequate
protection against damage from these faults. The radial configuration can be protected with
relatively simple series overcurrent elements, which is generally the lowest cost protection
option.

10253878 1-10
The basic element in series overcurrent protection schemes is a fuse. Thus, coordination of the
overcurrent protection devices in series along the distribution feeder from the substation to the
load is based on the shape of the fuse time-current characteristic (Figure 1-7).
A fuse is a one-shot fault current interrupter consisting of a metallic element that melts when the
current reaches a value where the heat generated by the fuse element is greater than the heat that
can be radiated and conducted from the fuse structure. Once the fuse element melts an arc forms
and continues until the arc extinguishes. The basic expulsion fuse extinguishes the arc by
pressure and cooling from outgassing of ablative material lining the fuse tube. The fault current
is interrupted at a natural current zero after the arc has been sufficiently lengthened and cooled.
The arc products are expelled out the open end of the fuse tube, often with a loud report similar
to gunfire – hence, the name “expulsion” fuse.
The lower curve in Figure 1-7 describes the minimum current at which the fuse element melts.
The higher curve represents the time in which the fuse clears the fault current.
A current-limiting fuse operates by stretching the arc along a silver element spiraled through
sand. This forces the current to zero within a quarter cycle. Operation is usually silent unless
moisture has infiltrated the fuse tube and causes the tube to fail.
Because of cost, most fuses on the distribution system are expulsion fuses. As shown in the
figure, the fuse operates faster for higher current. This is called an “inverse time-current
characteristic” (TCC). Upstream mechanical overcurrent protection devices such as breakers and
reclosers must also have an inverse TCC to coordinate with fuses. Relay manufacturers supply a
variety of inverse TCC options with names like standard inverse, very inverse, long time inverse,
extremely inverse, and ultra inverse. Each is intended for a specific application in the series
overcurrent protection scheme in a radial circuit.

1000

100

Time

10

0.10

0.010
5 50 500 5000 50000
CURRENT

Figure 1-7
Fuse time-current characteristic dictates utility fault protection coordination on radial distribution
systems

10253878 1-11
In the usual overcurrent protection scheme in a radial distribution system there will typically be
at least three or four overcurrent-based fault interrupting devices in series. Since short-circuit
current is expected to flow from only one direction, coordination is achieved by setting the
devices closer to the substation to act more slowly than the ones downstream. Only one of the
devices in series needs to operate to interrupt the fault current (Figure 1-8). In this example the
fault location is between the line recloser and the lateral fuse. Both the line recloser and the
feeder breaker will see the fault current but only the recloser needs to operate. The recloser TCC
is set faster than the breaker’s. If the recloser fails to clear the fault, the breaker will operate in a
backup role after a short delay. The fuse does not see any fault current because there is no power
source on the lateral. This is one protection conflict with radial circuits in the modern grid
context in which DER capable of contributing fault current exists in many places and could alter
the coordination of overcurrent protection devices.
Thus, the protection of a radial system can be accomplished by this relatively simple and low
cost scheme. There are many other details to an overcurrent protective relaying scheme that are
beyond the scope of this document, but these are the key concepts. One exception to the general
rule that devices upstream must operate slower is the so-called “fuse-saving” coordination. It is
expensive to send a line crew to change a fuse. Therefore, many utilities employ “fast” or
“instantaneous” tripping for one or two breaker operations in an attempt to clear the fault before
the fuse is damaged. This is generally called fuse-saving coordination. Although the fuse is
nearest to the fault, and would be expected to be the main interrupter if the fault is permanent,
the immediately-upstream mechanical interrupter is set to operate faster than the fuse melt curve.
It is a significant technological challenge to design a mechanical interrupting device and relay
system that can detect a fault, open the contacts, and interrupt the fault current faster than the
fuse can melt. If the fault remains after one or two operations, the fuse is then allowed to blow by
intentionally delaying the next breaker operation.

Figure 1-8
In series overcurrent protection, fault current is expected from only source and only one device is
required to interrupt the fault current

Thus, most distribution systems are configured radially because the protection system is less
expensive to build and simpler to operate than a meshed network system. Radial systems can
provide acceptable reliability in most areas. However, this naturally creates some conflicts with
designs for grid modernization efforts to apply storage and renewable generation on the
distribution system. With DER providing multiple sources of fault current, the current flows

10253878 1-12
cannot be assumed to flow in one direction. This will disrupt the overcurrent coordination
schemes utilities have employed for decades and is the source of many of the interconnection
conflicts for distributed sources on radial distribution systems.
In contrast, the transmission system is designed to accommodate multiple generation sources
(Figure 1-9). When a short circuit fault occurs, at least two interrupting devices – one at each end
of the faulted line – must operate to clear the fault. The meshed network generally allows this
line to be removed without interrupting power to the loads, which are served from the buses.
This offers higher reliability, but is also considerably more expensive than the typical radial
circuit protection scheme. Transmission system protection does not rely as much on overcurrent
relaying. Instead, it is more common to use impedance relaying to detect a fault on a line. This
requires both current and voltage signals, which increases the cost of the relaying scheme.
Impedance relaying is being used more frequently on radial distribution systems with large DER,
but it is not yet a universal practice. Impedance relaying functions are included on many modern
microprocessor-based relays that are nominally deployed for overcurrent protection.

Figure 1-9
On transmission systems, with multiple sources, two devices must operate to clear the fault

Utility fault-clearing actions often require operations beyond simply interrupting the fault current
to clear the fault current event. Automatic reclosing on radial systems is often employed to
restore power to as many customers as possible in a short time for faults that are temporary, or
transient. Temporary faults make up the majority of short-circuit faults on distribution systems
that are predominantly overhead lines. These fault are mostly due to lightning strikes and tree
contact and will clear once the current to the fault arc is interrupted. DER could not only
interfere with the fault current interruption but also subsequent actions of the overcurrent
protection scheme. Therefore, a radial system containing DER must revert to a radial
configuration for the fault clearing actions to proceed. This is the basis for the disconnection
requirements in IEEE Std 1547®-2003. More recent updates to this standard have proposed low-
voltage ride-through (LVRT) requirements for systems with such large amounts of DER that its
loss could cause system instability.

10253878 1-13
Figure 1-10
Typical reclosing sequences used in North American systems

Figure 1-10 illustrates a requirement for DER protection coordination with typical North
American radial distribution feeder automatic reclosing sequences. The first opening operation of
the breaker or recloser often occurs quite quickly, in 3-6 cycles, or faster, especially when fuse
saving is employed. The first reclose operation occurs after a delay interval of typically 0.5 to
5.0 s. An interval of 0.5 s or less is often referred to as an “instantaneous” reclose. The reclose
interval is intended to allow sufficient time for the fault arc products to dissipate, or clear, so that
the subsequent reclose operation will be successful. It is not intended to be sufficient time for
DER to disconnect.
All connected DER downstream of the operating breaker should disconnect as early as possible
within this interval to:
1. Remove sources of current that could maintain the fault arc, and
2. Avoid damage to the DER upon reclosing.
This can be difficult to achieve when using the instantaneous reclose philosophy. IEEE Std
1547-2003 requires the disconnection in 10 cycles (0.166 s), but this is quite long in the world
for fault interruption and clearing. While instantaneous reclose intervals are nominally 0.5 s, they
can be as short as 0.2 s. Thus, many utilities have extended the first reclose interval to at least
2.0 s and, often, to 5.0 s to allow time for the DER to detect the island and disconnect. As the
industry moves forward with such issues as LVRT for DER deemed necessary to sustain system
stability, the series overcurrent protection issue will continue to be a source of conflict and
compromise.
Another Integrated Grid concept that requires a different form of protection is the microgrid. In
this context, a microgrid refers to a small portion of the distribution system being separated from
the rest of the system and supplied by one or more independent, smaller power sources. The
normal settings on the breakers, reclosers, and fuses require higher current levels to operate than
can often be delivered by the microgrid sources. This requires new protection schemes, possibly
involving impedance and voltage relaying to supplement overcurrent relaying. As a default
behavior, the sources on a microgrid are designed to simply cease to energize the system in case

10253878 1-14
of a fault being detected while the microgrid is isolated from the main distribution grid. This an
area of continuing active research and standards efforts such as the IEEE Std. 2030 work.
Urban LV networks have different fault interruption and clearing schemes than radial
distribution systems. As shown in Figure 1-11, several interrupting devices may have to operate
to isolate the fault. Since these systems are generally predominantly underground cable systems,
reclosing is not employed. When the fault on the feeder as shown occurs there is strong fault
current contribution through the feeder breaker or recloser at the head of the faulted feeder. In
addition, there is a significant contribution to the fault back through the network transformers
connected to the faulted feeder. Keep in mind that the meshed LV network connects all the
primary feeders together via the network transformers. The unfaulted feeders will supply current
through the LV network to the faulted feeder.

Figure 1-11
Clearing faults on the primary feeders to urban LV networks requires operation of multiple
devices

There is a network protector on the LV side of each network transformer. It is a heavy duty
circuit breaker capable of interrupting several thousands of amperes of fault current. It is relayed
by a reverse power relay with the assumption that the only time active power will flow
backwards through the network transformer is when there is a fault on the primary feeder
supplying the transformers. The network protector opens to stop the flow of fault current back
into the primary system. Various schemes are employed to automatically close the network
protector back in after normal voltage appears for a certain amount of time.
In its simplest form, the reverse power relay is set to trip on a small amount of reverse current
with no intentional delay. This creates operational issues for applying DER on LV networks.
Even a very short period in which the total DER power output exceeds the network load power
momentarily could result in tripping several, if not all, network protectors. Relays for network
protectors have been developed that give more options for reverse power levels and time delays,
but there remain thousands of legacy network protectors with the “hair trigger” reverse power
relay installed on urban LV networks.

10253878 1-15
A Brief Introduction to Distribution Planning for the Modern Integrated Grid
There are new methods being proposed to address the distribution planning problem for the
modern distribution grid that has both loads and sources. Not all have to do with DER, but many
do. The planning problem becomes more than simply ensuring that there is sufficient wires
delivery capacity to serve the peak load. The discussion here represents a sampling of a few
general classes of these methods.
Peak Capacity Planning
This is the traditional distribution planning method in which the planner simply evaluates the
ability of the system to supply the forecasted peak system demand over a selected planning
horizon – typically, 5-10 years for distribution planning.
Planning cases may be done with models representing the most-limiting emergency condition.
This assumes the failure of one key system element in what is believed to be the worst case
contingency (N-1 contingency planning). An attempt is made to serve the peak demand within
acceptable voltage and current limits with this contingency. The limits in the emergency
condition are generally relaxed from the normal system condition.
Some utilities, especially those with large urban LV networks commonly design for N-2
contingency conditions. That is, two key components can fail and the peak load demand can still
be served.
One of the reasons this method remains popular despite its shortcomings for planning for the
modern integrated grid environment is limitations of computer tools for planning. Those tools are
designed to best support studies of one loading condition at a time – usually the peak load or
peak load plus a margin.
Cost Minimization Planning
This is the primary advanced planning method that will be emphasized in this chapter. The
method presented here was developed jointly by Energy and Environmental Economics, Inc.,
San Francisco, CA, Electrotek Concepts, Inc., Knoxville, TN, and EPRI in the mid-1990s. The
research team began to consider methods for including DER in distribution planning that were
more palatable to distribution planners, but still effective in capturing the benefits of DER. The
method extends the traditional peak capacity planning method by simulating load variation over
time. This captures many of the key impacts of variable resources on the modern integrated grid.
One feature of this method is that it can accommodate diverse planning alternatives, such as
DER, DSM, and DA along with conventional distribution planning alternatives such as new
feeders and substations. It develops operating cost functions by assigning marginal costs to
losses, reliability, etc. By comparing costs, the following become more clear:
• When it is economical to make an investment?
• Which investment alternative is likely to be more economical?
It is a value-based planning method. Uncertainty is typically handled determining sensitivity of
solution to load growth.

10253878 1-16
Risk Analysis
Risk analysis is one of the chief alternative approaches to Least Cost Planning, but also can be a
complementary approach. The main idea is to minimize the regret felt by the decision maker
after verifying that the decision was non-optimal given that the future has in fact occurred.
The method involves creating a “Regret function” for the i-th alternative for the k-th future in the
form

Regretik = R ( fik − fk )
opt

Where R is typically either linear or parabolic.

The function "f ik " is cost of the i-th alternative for the k-th future while “f k opt” is the cost
of the optimal solution for the k-th future.
An advocate for this method is V. Miranda, who claims that this planning method outperforms
probabilistic planning methods. [25]
Merrill [26] lists three main characteristics of risk analysis in planning:
1. Robustness: How likely are we to not regret a decision?
2. Regret: If it doesn't work out how sorry will we be?
3. Exposure: When will a decision be regrettable?
Popovic [27] proposes using risk management for restoration.
Market Assessment Methods
Market assessment methods involve a method to simulate the power market. They assume the
cost of power will reflect the need for and value of energy.
The methods basically combine
• Production costing
• Optimal power flow (OPF)
The main drawback to this method is that the simulations are lengthy and expensive. They are
also designed primarily for larger, transmission-connected DER. So their applicability to
distribution systems is suspect.
New, less time-consuming methods based on Risk Analysis are being developed.
Fuzzy Sets
A popular topic in the mid-1990’s was the application of Fuzzy Sets to engineering problems. El-
Hawary [28] summarized much of the activity in his book in 1998.
Proposals have been made for doing distribution planning with fuzzy methods and fuzzy power
flows [29][30]. A fuzzy power flow represents a load as a range of values with a possibility, or
plausibility, factor between 0 and 1 for each load point. A simple trapezoidal fuzzy set is shown

10253878 1-17
in Figure 1-12. This is perhaps the most common way fuzzy sets are represented although
smooth functions many be used as well.
The main idea is that the method naturally incorporates the uncertainty in the power flow
solution. The solution is also a fuzzy set. The result can be defuzzified, or made into a “crisp”
value by a variety of ways including the maximum power, average power, and centroid of the
set.
While there are attractive features of this method, there are few signs that it has caught on in
distribution planning. The common power flow tools do not support fuzzy sets. Peak capacity
planning would use the rightmost point in Figure 1-12 despite the fact that it is not very likely to
occur.
Fuzzy set planning techniques are often combined with simulated annealing and AI techniques.

Figure 1-12
Representing a load as a fuzzy set

Other Miscellaneous Planning Methods

A literature search on “distribution planning” will show many different approaches to optimal
planning. Many of the "planning" methods are more engineering than planning, being focused on
determining how to get something working rather than determining whether or not it is worth it.
These methods can be lumped into the following categories:
• Genetic algorithms[31]
• AI and Expert Systems
• "Branch exchange" methods - optimal feeder routing[32][33]
• Multivariable optimization
• Knowledge-based approach
• Stochastic methods [39]
Each has particular benefits, but have not become common except perhaps in academic research.
Refer to the Bibliography for papers and books on other planning methods.

Typical Distribution Planning Procedure

Distribution Planning has traditionally meant determining a least cost means of supplying power
(capacity) to the customers being served for a selected planning horizon. Planners do a load
forecast for a selected planning horizon. The horizon is typically 5 – 10 years for existing
distribution systems where the growth is moderate, but sometimes a much longer horizon is
chosen for new areas that have been opened for development. This can depend on how certain

10253878 1-18
the projection of new loads is. Utilities generally wish to limit the risk of investing in too much
capacity that is never needed.
Once the forecast has been agreed upon, a few alternatives for serving the load are considered.
Each alternative is analyzed with respect to feasibility, reliability, power quality, and cost. Power
flow analysis is the chief tool employed by the planners to determine the sizes and locations of
transformers, lines, cables, capacitors, etc. From decades of experience and company standards, a
limited number of sizes and types of devices have been selected for planners to construct
possible solutions to the capacity problem.
The least cost solutions considering the time value of money are selected. Often two alternatives
will be presented to management, which will select one to be entered as a candidate for inclusion
in the next capital budget.
The basic outline of a typical planning process is:
1. Define the distribution planning area
2. Perform the load forecast for the distribution planning area
3. Evaluate Normal Operation of the distribution systems
a) If normal operating constraints (i.e., planning limits for voltage and current) are violated
at or before the end of the planning horizon, determine least cost plan to correct.
4. Evaluate Emergency Operation (generally single contingencies, although some utilities plan
for 2 simultaneous failures of key components)
a) If emergency operation constraints are violated, revise least cost plan.
b) Repeat until the constraints are met.
Normal and emergency states are commonly defined as follows:
Normal state
1. Feeder switches in "normal" configuration. This is the preferred operating configuration,
which could change seasonally.
2. No failed components.
3. Current and voltage limits have considerable engineering margin and exceeding them does
not necessarily mean that load must be dropped.
a) The operating margin allows load to be picked up from other circuits in case of a failure
on those circuits
b) The normal limits are used to trigger planning studies understanding it may be a few
years before anything will need to be done.
Emergency state
1. This is the reconfiguration of a normal circuit after a failure, such as the loss of one of two
substation transformer or having to back up a feeder to another yielding either a thermal
overload or low service voltage.
2. Current and voltage limits have little or no engineering margin. These are generally critical
limits that can be tolerated for only a short time such as 1 hour for cables or one day for
transformers. Loads must be curtailed when these limits are exceeded.

10253878 1-19
Emergency criteria usually dictate when it is necessary to invest in a capacity upgrade. Using the
assumed load growth, determine when during the planning horizon a critical failure will result in
unserved load. The upgrade must be in place by that time.
Table 1-2
Typical distribution capacity solutions considered (ascending order of cost)

Option Cost/kVA or kW Typical Capacity Increment

Capacitors $2 - $30 / kvar 1-2 MW (declines with each additional capacitor)
Voltage regulators $4 $20 / kVA 1-5 MW
Feeders $20 - 150 / kW 10-20 MW
Substations $35 - 200 / kW 20-90 MW
Generators $500 - 5000 / kW 0.1 - 50 MW

Table 1-2 shows typical capacity upgrades that distribution planners employ with cost ranges and
the typical incremental capacity gained. The two least expensive option are capacitors and
voltage regulators and these are commonly employed by planners in North America.
Interestingly, they are not utilized as often on the MV level by operators of European-style
distribution systems.
Capacitors provide additional capacity by reducing the current in the feeders and helping to
support the voltage. The incremental capacity declines after the first few banks are added to the
system so there is a limit to what can be achieved with capacitors alone. A capacitor bank is one
of the few capacity upgrades that can pay for itself in loss savings if it can be operated most of
the year. A yearly load profile simulation may have to be performed to determine if the savings
can actually be achieved. However, there is generally more value in being able to serve more
load reliably.
Voltage regulators applied to the line allow greater use of the feeder asset investment by
providing good voltage regulation without investing in a new substation for loads that are distant
from existing substation. They can be applied as long as there is current-carrying capacity
available in the wires. There are typically no more than three regulators in series on long
distribution feeders. It would likely be time to build a new substation if more are needed. Feeders
with several regulators in series can suffer from load-rejection overvoltage issues when load
disconnects after a fault on the systems. Special regulators with a fast runback option are
supplied for that application.
One final thing to note on this table is that generators are considerably more expensive than any
of the wires options. Thus, they are typically not the first options for utility planners unless
serving remote areas where there are limited options for power delivery by wire. Of course,
today customers are installing generation themselves to take advantage of various subsidies and
net metering laws. They are being reimbursed for providing energy but whether they are able to
produce power at the right time to satisfy the planners capacity problem is another issue.

10253878 1-20
Observations on Traditional Planning Methods with Respect to Planning Systems
with DER and Other Smart Grid Features
1. Constraint violations are evaluated at peak loading. The system is at peak loading only a few
hours per year and averages 40-60% of peak. (See Figure 1-18.) DER often must be
evaluated at minimum loading, which is 30-40% of peak. Planners need to simulate over a
significant period of time to see the true impact of planning decisions when the options being
considered disrupt the daily load shape.
2. Emergency constraints are often violated before normal configuration constraints. Typically,
only single contingencies are considered although a few utilities consider two contingencies.
The more contingencies considered, the higher the reliability and the more the cost to build
and operate the system because more redundancy must be provided.
3. Generally, planners evaluate only a few options for which there are existing heuristics based
on decades of historical performance. The one with the least cost over the time span of the
planning horizon is chosen. The heuristics, such as those for thermal overload ratings, are
commonly based on an assumed, historical daily loadshape, which in the modern grid may
now be altered by DER such as rooftop solar generation and such loads as EVs.
4. This method assumes that it is automatically economic to invest in new capacity when the
peak "capacity" of the system is judged to be exceeded. It is often difficult to boil a complex
distribution system down to one capacity number. Peak capacity planning methods do not
give much insight into
a) Value of the problem,
b) When it would be most economical to invest, and
c) How to rank options.
This is particularly the case when there is little experience in implementing the option, as is
usually the case with new DER and other Smart Grid planning options.

Cost Minimization Planning

The modern “integrated grid” requires a fundamental change in the planning process. Planners
accustomed to wires-based solutions are now faced with non-wires options such as DER,
demand response, storage, and intelligent controls in line devices and distribution management
systems. In this environment, distribution planners often find themselves reacting and scrambling
rather than planning. The key question is: How can you compare these diverse alternatives on a
common planning framework?
The Cost Minimization Planning method was developed to provide one means of answering that
question. In the example presented here, it is implemented in a manner that is a relatively minor
extension of the traditional peak capacity planning method described above. But it does require
an upgrade to the typical distribution planning power flow tools. [37]
Like the traditional method, it starts with a load forecast over a planning horizon and power flow
analysis is the key tool used to evaluate options. Instead of looking at only at a peak load case,
the method captures the time value of the capacity problem by simulating the load over a
significant amount of time, such as one year. This is now called quasi-static time-series, or
QSTS, simulation and is available in several distribution planning tools. This also allows the
planner to capture the coincidence of the load and DER output. For example, if the DER is solar

10253878 1-21
PV and the peak load occurs in the evening, the PV generation is given very little capacity credit.
Likewise, the effects of other disruptive technologies such as electric vehicles can be captured.
The same criteria for normal and emergency ratings used in traditional peak capacity planning
are applied. As the QSTS simulation progresses, a tabulation of the amount of load served in
violation of the ratings is made. This essentially becomes a risk function that can be applied to
compare alternatives on the same basis. Thus, this method is a combination of traditional
methods and a risk analysis method.
Before we look at the process, we will visit the subject of load forecasting with a common
example.
Load Forecasting and Loadshapes
Load forecasting is often as much of an art as a science. [36] Many things go into a forecast
including meeting with city planners and developers to keep on top of possible additional loads
so that there are no large surprises. In areas with a regular historical pattern of load consumption
and only minor amounts of new load additions, planners frequently project the future load based
on a simple linear regression from the historical peak load. A number of things can throw this
projection off. One example is the dramatic increase in the rooftop solar generation in some
areas. Because the solar PV compensates for some of the load growth, it is difficult to know what
the true load demand is. Another example is that the linear load projection may be thrown off by
the temperature sensitivity of the load. For example, a summer-peaking load may not reach its
peak every year, having several years of cooler temperatures in between peaks. In fact, it may
appear that the load is declining, making it difficult to convince management that a major
investment such as a new substation transformer is needed. An actual case will be used to
illustrate some interesting aspects of temperature-normalized load forecasting. [35]

Figure 1-13
Which slope is chosen for a temperature-normalized forecast of substation load?

10253878 1-22
Figure 1-13 shows a plot of thousands of load MVA and temperature points collected over
several years. This is one way to estimate the sensitivity of the load to temperature so that the
historical load can be corrected for temperature. The question is: Which of the three slopes give
the correct sensitivity? There is a natural tendency to choose the average slope in the middle,
which is approximately 2 MVA/deg F. This is a technique sometimes employed on large
transmission systems. If one chooses the slope fitted to the samples with the hottest temperatures
– the rightmost slope – the slope is approximated 4 MVA/deg F. The other line, fitted to the
highest load points is 0.25 MVA/deg F. So there is a significant difference between the three
methods.
It turned out that none of these slopes matched the observed load growth characteristic. It was
observed that the peak load for this distribution system historically occurred on the hottest day.
This is different than what is observed on many transmission systems where it may take 3 very
hot days in a row to see the peak load demand. There is less diversity on a single distribution
substation and in this case, it only took one hot day to produce a peak load.
Additional insight into what is behind the temperature dependencies was gained by plotting only
the MVA-temperature points for the peak day. Figure 1-14 shows the trajectory of the load
during the day and gives a clue as to why it can be difficult to determine the proper slope.

90

80
One Hour
70 Max Daily Demand

60

50
MVA

40

30
Max Daily Temperature
20

10

0
50 55 60 65 70 75 80 85 90 95 100
T, degrees F

Figure 1-14
The maximum daily demand may lag the maximum daily temperature

The peak load lasts for 3-4 hours and lags the peak temperature by 2-3 hours. So there is not a
direct correlation between load and temperature. Figure 1-15 shows the daily load and
temperature as function of time. The temperature peaks a little past 12:00 pm and then gradually
declines into the evening. As is common with many mixed residential/light commercial areas,
there is an evening peak between 5:00 and 8:00 pm when residents leave their workplace and go
to their homes, stores and restaurants on their way. Due to the high temperatures, the air
conditioning demand increased and the total load spiked up as the load diversity was lost. Thus,
the peak load day coincided with the peak temperature day.

10253878 1-23
Thus, it is difficult to determine which slope to use from the scatter plot of the hourly load-
temperature data. Once the coincidence of the peak substation demand with the hottest day was
understood, the hourly load-temperature data were converted to a different set of points matching
the peak load and temperature within 24-hr periods. The result is shown in Figure 1-16. This
yields a clearer picture of the sensitivity of peak load to temperature. These points were found to
be quite accurately fitted with a 4th or 5th order polynomial curve using a spreadsheet. Then the
slope at peak demand and temperature was accurately determined by taking the first derivative of
the polynomial equation. As shown, the slope at the highest load point was 0.8 MVA/deg F.

Figure 1-15
Load and temperature vs time for two substations

Figure 1-16
Daily peak load vs. max temperature

10253878 1-24
Another noteworthy issue that appears in Figure 1-16 is the sparse scattering of points below the
dense cluster of the main peak daily load-temperature points. These were found to represent days
in which part of the substation load was temporarily transferred to another station during the time
of peak load. These points are inconsequential to the issue at hand where the important thing is
what happens to the peak demand on the hottest days. Therefore, one further step is to filter out
these outlier points. The technique used was to discard outliers more than a specified number of
standard deviations (usually between 1 and 2) from the mean at each temperature.
Figure 1-17 shows the result of applying the temperature correction to the forecast of the peak
load for each year. Historical data going back over 50 years indicated that a temperature of 100F
measured at a selected permanent weather station was reached approximately once per decade
and this was chosen as the planning temperature. The temperature of 100F was reached in 2001
and the load was equal to the firm rating of the substation. Planners wanted to add capacity
immediately, but the load declined each of the next three years. It was unclear if the reason was
cooler summers or load declining due to economic conditions. Correcting the actual load for
temperature suggested that the load was continuing to grow at a steady pace. This was proven
correct in 2006 when the temperature once again reached 100F.

Substation A Weather Normalized Load Forecast

Normalized to y = 1.2117x - 2347.4
90 T=100 F

80

70

60
Actual Load Firm Rating
76.6 MVA
50
MW

40

30

20
Assumes Incremental Load = 0.8 MVA/degree
10

0
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Year

Figure 1-17
Performing a load forecast with weather-adjusted linear regression

The main takeaway from this example is that accurate load forecasting requires a clear
understanding of the load characteristic of the planning area.
This exercise also underscores the importance of utility planners coming up with means to
account for the masking of load growth by DER. For example, if a fault were to occur on a hot
day with much air conditioning load, the DER will disconnect for a few minutes leaving parts of
the system overloaded during that time. For solar PV on residential feeders, planners can likely
get a good idea of the actual peak load by observing the evening peak at sunset.

10253878 1-25
Another aspect of the load modeling that is important to the Cost Minimization planning
technique is the load shape. Figure 1-18 shows a typical yearly loadshape for a summer peaking
distribution system that has only a minor winter peak. This is an 8760-hour loadshape for the
entire substation. Some interesting features of this loadshape relevant to distribution planning
are:
• There are only a few hours at peak load and not many at 90% or greater.
• Since utilities have an obligation to serve the load, the system is built to supply the peak,
which means that it is under-utilized most of the year.
• The minimum load is about 30% of the peak load and the average load for the year is
between 40-50%.
• DER output during the minimum load periods, which is most of the year, is more likely to
cause voltage regulation problems than during peak load.
• There was a hot week in the month of May, much earlier than expected. This is at a time
when many utilities are performing maintenance and feeders are in suboptimal
configurations.

Figure 1-18
Yearly loadshape for the distribution planning area

Many utilities have now deployed 100% AMI metering for their customer and could supply this
kind of curve for each customer. So this kind of data will be even more available to utility
distribution planners in the future. The readers should keep this in mind when considering power
flow tools. EPRI is participating in research projects to considerably speed up the annual load
simulation.

10253878 1-26
Cost Minimization Planning: Overall Process
Start with
Do-Nothing Case
Operating Load,
System Model Simulations Growth
Charac-
teristics
System Energy
Data T&D Expansion Losses
Investment Plans DG Energy, Hours
Energy above capacity
PQ Indices
etc.
Decision
Options: Information
Substations
Feeders
Cost
DG
Costing Data
DSM
DA
$$
Figure 1-19
Overall process for Cost Minimization Planning

The Cost Minimization Planning Process starts with the "Do Nothing" case. The basic idea is to
keep doing nothing until it is no longer economic to do so. Then you invest in something that
adds capacity and reduces costs. The planner strives to determine the lowest cost option over the
planning horizon.
The planner performs yearly load simulations over the planning horizon using load forecasts.
These simulations are used to compute the costs of operating the system.
The planner chooses from a menu of alternatives to address the planning problem constraints
including such things as DER, DSM, and DA as well as traditional solutions such as new feeders
and substations. The present value of each alternative is known, but we don't yet know how
effective each will be.
The planner strives to choose an option with lower cost than the cost of operating the system
with the constraints. Each feasible alternative is evaluated over the planning horizon and the
costs are computed. The alternatives are compared on the basis of the selected cost functions and
the alternative with the lowest overall cost is chosen. This process allows the consideration of
many diverse options planners may encounter in the modern grid and the comparison on a
common cost basis.
The process is repeated for various load growth scenarios to determine how that will affect the
solution.
As illustrated in Figure 1-20, the selected planning cost functions are computed over the
planning horizon as the load, measured in kWh, grows. When the load grows to a point where
the capacity is exceeded, the cost functions begin to increase rapidly, generally because the
reliability of the system is at risk. When the cost reaches the present value cost of a planning

10253878 1-27
option it should be economic to invest in that object. Of course, it could be more economic to
wait until the load grows a bit more and invest in a larger capacity option. One should check all
the options.

High Growth Scenario

Capacity Exceeded Low-Growth Scenario

Investment Option 1
$,
Present
Value Investment Option 2

kWh

Figure 1-20
Cost functions for different growth scenarios

For a high growth scenario, there is usually little time difference between the low-cost
investment and the high-cost investment. For a low-growth scenario it can be economic to invest
in a lower-cost DER solution for a few years before investing much more in feeders and a new
substation.
The cost functions typically include a measure of reliability, cost of losses, emissions and other
environmental costs, power quality penalties and other costs pertinent to the planning area.
Reliability Costs
While there are operating costs such as losses and revenue from energy sales that can be directly
computed from the power flow model of the system, the key driver in the planning investments
is nearly always reliability. Reliability has a much higher value than most of the other elements
of the cost equation. Otherwise, how would one ever justify investing in expensive options such
as multi-million dollar substations? Some of the value of reliability is based on the mandate by
regulators to serve the load and some is based on the desire of the utility to provide service
acceptable to their customers.
Loss savings will not justify such large investments, although they can contribute incrementally
to support an investment decision. There is more on this subject in a subsequent section.
One way to assign a value to reliability compatible with cost minimization planning is to
compute an estimate of the energy at risk of being unserved when there are component failures
affecting the delivery of power. A means of computing the energy is illustrated in Figure 1-21.

10253878 1-28
Figure 1-21
Illustrating the computation of energy exceeding normal (EEN) and unserved energy (UE) for a
daily load simulation for two instances of load growth

Each power delivery element of the system is assigned two ratings: Normal and Emergency. The
Emergency limit is the maximum power delivery limit for system elements. When the power
demand exceeds this limit, some load must be curtailed to prevent conductor burndown,
transformer failures, etc. The disconnected load becomes unserved. The Normal limit is also
called the planning limit. The system still operates within engineering limits as long as there are
no failed components. However, a failure of some key component will result in overloads for
which load shedding must occur. The Normal rating is more subjective than the Emergency
rating, which is the limit at which something is assumed to fail. Some utilities simply establish
the Normal rating as a percentage of the Emergency rating based on experience and trigger
planning studies when the rating is exceeded. This is typically 50-80% of the Emergency rating.
Another way to establish the Normal rating is to determine the maximum power that can be
delivered without exceeding an Emergency rating given a critical failure.
Two daily load shapes are shown representing the load growing in successive years over a
planning horizon. The shaded areas under the curves represent the energy served above the two
limits for the two load levels. EEN represents the energy served above the Normal limits and UE
is the energy served above the Emergency limits. The main idea of this concept is to compute the
annual EEN or UE and assign a cost value to it. When the cost rises to the level of an investment
option, that option is judged to be economic.
The value of unserved energy is basically the cost one would be willing to pay for energy in the
event it becomes unavailable. While the electricity retail price may be in the range of $0.10 -
$0.20/ kWh, EEN and UE would typically be assigned a cost at the rate $5/kWh, or more. One
way to determine this cost is to observe what energy consumers are willing to pay for backup
sources to cover power outages. The cost will obviously vary depending on the type of customer.
When the system begins to be constrained, the cost function constructed from EEN or UE turns
up sharply.

10253878 1-29
It is informative to compare this method with the traditional peak demand planning technique.
Figure 1-22 compares the cost functions for the high-growth and low-growth scenarios illustrated
previously with the assumed cost function for the peak demand planning method. In peak
demand planning, exceeding a capacity limit is assumed to be sufficient justification to invest in
something to correct the capacity deficiency. This essentially means the cost function goes
vertical at the loading when the capacity constraint is exceeded. Of course, this infinite slope
does not discriminate well between the costs of the possible alternatives. It intersects both
investment options as the same time and least cost planning will generally, but not always, favor
the larger investment in more capacity.

Peak Planning
High Growth Scenario
Assumed Cost
Low-Growth Scenario

Investment Option 1
$,
Present
Investment Option 2
Value

Capacity Exceeded

kWh

Figure 1-22
Comparison of cost minimization planning to peak capacity planning assumed cost functions

Figure 1-23 shows a 3-D plot of an annual simulation of EEN for a summer peaking system
computed with the EPRI OpenDSS program to illustrate the information the method provides.
For most of the year, EEN=0, meaning the system under study can suffer a critical N-1
contingency and still serve all the load. When the failure occurs there is sufficient capacity in the
alternate feed to supply all the load with exceeding the Emergency limit of any element. As the
EEN rises in the heat of the day in mid-summer, this is no longer true. A failure at those times
will likely result in some load being shed to keep the system viable.
Thus, the total EEN summed over this curve is a measure of the risk of experiencing unserved
energy. Comparing this number to the same measure for other plans will help the planner discern
between options. [38]
The planner can quickly grasp the type of capacity problem being faced and the extent of the
problem. This figure also tells the planner that the potential overload problem is confined to a
few hours of the year in the Summer. It also gives some idea of the magnitude of the problem: a
solution must be able to provide approximately 1200 MWh of energy to the planning area at 3:00
PM for the month of August. This analysis does not say where the resource should be located
within the planning area or how many locations should be employed; that is a separate analysis.
Often, problems with the relatively narrow peaking characteristic shown can be solved
economically with incremental solutions provided the load is not growing rapidly. If the base and
height of the EEN shape were greater, a traditional substation or feeder wire delivery-based
solution is usually indicated. As with many planning issues, it all depends on the technologies
available. This particularly problem could possibly be addressed by solar PV generation
appropriately distributed over the planning area.

10253878 1-30
Figure 1-23
Example EEN characteristic for a summer peaking system

Planning Limits
Voltage
Distribution systems have traditionally been designed to prevent excessive voltage drop from the
HV/MV substation to the feeder extremities. DER now requires planners to consider voltage rise
as well.
Acceptable minimum planning voltage is generally 95% to 97.5% of nominal. This allows some
margin for the additional voltage drop that occurs within the facility up to the point of connection
to the load. The ANSI C84.1 Range A voltage standard is -8% to +5% of nominal.
Without DER there are few cases where the maximum voltage exceeds +5%. Excessive reactive
power production from power factor correction capacitor can cause sufficient voltage rise to
exceed the +5% limit, but this is relatively rare. The proliferation of DER such as rooftop solar
PV generation has made steady-state voltage rise a much more common occurrence.
Figure 1-24 and Figure 1-25 illustrate the problem faced by planners designing voltage
regulation for a feeder with a large amount of DER. This example is based on the IEEE 123-Bus
Test Feeder and uses the OpenDSS program. The overvoltage problem usually requires analysis
of the minimum load demand condition.

10253878 1-31
 Voltage rise caused
by capacitor banks
compensating for
reactive power

Figure 1-24
Normal minimum load voltage profile -- no DER

In Figure 1-24, without DER, the voltage rise past the last voltage regulator is caused by the
power factor correction capacitors that remain on the system at low load. One might argue that
these capacitors are not needed at minimum load and should be switched off and that would be a
good argument. Nevertheless, this conditions exists on many feeders where banks without
switches are installed.
The last regulator on the feeder is set to regulate about +4%, which provides sufficient margin to
support the voltage at peak loading. This is common. Note that the substation regulators have
their Line Drop Compensators activated so that the voltage profile for the first zone of the feeder
is relatively flat at minimum load.
The regulation problem begins to appear after a large amount of DER is installed near the end of
one of the feeder extremities (Figure 1-25). Buses at the ends of the feeder are pushed up over
the +5% limit. This problem is perhaps most likely to occur on a Saturday or Sunday afternoon
when the load demand is at minimum and there is high irradiance from the sun.

10253878 1-32
 Overvoltage at minimum
load caused by excessive
DER output

Figure 1-25
Minimum-load feeder voltage profile with large amount of DER

Figure 1-26
Adjusting the last regulator to provide more headroom for DER output

10253878 1-33
The overvoltage is exacerbated by the high setting of the downline voltage regulator. One
common solution is to adjust the settings on the last regulator on the feeder to regulate to 122 V
instead of 124 V. The result of doing this is shown in Figure 1-26.
This illustrates one of the distribution planning philosophy changes that must frequently be done
to accommodate large amounts of DER on the distribution system. Instead of designing only to
prevent low voltages due to voltage drop, the voltage profile is targeted more toward the middle
of the operating range to also allow for voltage rise.
Transformer Rating Limits
Transformer current rating limits are based on the thermal limits of the transformer’s insulation.
It is mainly a matter of dissipating the heat in the transformer that arises from the losses within
the transformer as it serves load. A typical distribution substation transformer has a thermal time
constant of about 2 h, so the transformer can absorb a significant amount of overload above the
nameplate rating before the temperature increases to the point that damage is done.
Transformer thermal rating limits are generally set based on an assumed daily loadshape. EPRI
has provided a computer program, PTLOAD [40], for computing transformer ratings by this
procedure. It follows IEEE Std. C57.91-1995 to determine transformer ratings. Figure 1-27
illustrates the process of computing the winding temperatures as a function of time.
A transformer is built with many organic materials as well as copper and steel. The insulation is
typically some form of cellulosic material, mineral oil, plastics, and other material. These
materials are assumed to degrade rapidly for temperatures above 115 C. Figure 1-27 shows the
hot spot temperature exceeding 115 C for more than 4 hours.
Following IEEE guidelines, the maximum normal rating is established for the peak loading the
yields 0.0369% loss of life per day. This typically allows the transformer to be loaded to
approximately 110% of maximum nameplate rating for a typical daily load pattern. In an
emergency where the transformer has to compensate for the failure of another transformer,
planners might allow perhaps 1% loss of life per day. This permits the transformer to be
overloaded by typically 140 – 150% for a typical daily load shape.
p )
150 Loss of Life

Top O il
Degrees C

100

50 Hot Spot

0
0 5 10 15 20 25 Ambient

Hours
Figure 1-27
Substation transformer temperatures for a typical daily loadshape

10253878 1-34
On Losses in Planning
It is very popular among researchers in various subject areas related to distribution planning to
make “optimal” planning decisions based on minimizing losses. Losses are proportional to the
square of the current so it would seem to make sense that a solution that reduces losses might
also free up capacity to serve more load and defer major investments. The minimum loss solution
may very well point in the right direction for the planning decision, but can a planner base a
justification for a capacity investment on the value of the loss savings?
Loss minimization is attractive to many engineers and engineering students because one can
easily write an equation and calculate them. So a power flow program may be readily written to
perform the task. It is not so easy to formulate a computer program that captures the full set of
things that must be considered in a planning decision including the intangibles that are embedded
in minds of experienced planners.
One issue that is overlooked by many is that if the rate at which the cost of losses is increasing is
less than the cost of money, it is almost always more economic to DO NOTHING. Looking at
the net present value of an investment stream into the future, it would appear that the cost of
losses is actually decreasing without doing anything. So depending on loss reduction can be
misleading.
One of the few capacity investments that pays for itself based on the cost of loss savings is the
application of power factor correction capacitors. Capacitors have an installed cost of $10-
30/kvar, depending on options, and the first bank or two added to a feeder can pay for itself with
loss savings in a few years. The return on investment declines with each additional capacitor
bank added because there are diminishing percentage improvements in the line currents.
However, the capacitors may still have substantial value for voltage improvement, but that value
is not as easy to compute as losses.
To attempt to justify DER based on loss improvement is generally futile; DER costs generally
exceed $500/kW. It might be possible to determine a good site for the DER using loss savings,
but the value of the DER is related to other things.
The bulk of the value of DER is in providing capacity and reliability.
Another error that is commonly made when evaluating losses is to compute the losses only at the
peak loading condition. It is difficult to claim “optimal” siting or sizing of the resource when the
system spends less than 10 h per year at peak load. Proposed optimal solutions should be
confirmed by simulating over a significant time period (week, month or year) to prove that the
proposed solution will work as expected over the whole year.
For example, during the EPRI Green Circuits program [41] it was learned that no-load losses
dominated in approximately 50% of the feeders simulated. No-load losses are primarily
excitation losses in distribution transformers and vary approximately by the square of the
voltage. Computing only the peak loading losses could lead to the appearance that there are great
savings in reconductoring a feeder section to save line losses (I2R losses). This is a very
expensive solution that would require substantial cost justification. It was found that the optimal
solution considering the entire yearly load profile was to reduce the no-load losses by lowering

10253878 1-35
the voltage 2-3% by simply adjusting the voltage regulator and substation OLTC settings. In
some cases a voltage regulator might be needed to prevent low voltages, but that should be less
costly than reconductoring a section of feeder to reduce losses.

10253878 1-36
2
DISTRIBUTION POWER FLOW METHODS
Introduction
If you were to ask most engineers in the electric power industry to list the different types of
power flow programs, you will probably get these three answers:
1. Gauss-Seidel
2. Newton-Raphson
3. Fast Decoupled
These are actually three different kinds of general mathematical methods for solving sets of
nonlinear equations. “Fast Decoupled” algorithms take advantage of certain peculiarities of the
power system, but the other two methods are used to solve systems of nonlinear equations in
many different sciences. So they are not technically power flow methods, although most power
engineers associate the names with certain power flow solution methods.
The basic power flow equations solved by these three methods are nearly the same. Descriptions
may be found in most basic power system texts such as the classic Elements of Power System
Analysis by William Stevenson (McGraw-Hill) and many newer books. The methods presented
in these power system texts – and taught to most power engineering students – were developed
for power flow calculations for transmission (HV) systems. Some key characteristics of these
methods with respect to distribution system analysis are:
• Generally developed for balanced, positive-sequence models only
• Positive-sequence models are adequate for capacity planning of 3-phase systems, but not for
other issues on unbalanced distribution systems
• Formulation is heavily influenced by the so-called P-V generation bus which is not needed in
most distribution analysis
• DER and other Integrated Grid issues demand more detailed models
• They are not necessarily the best methods to use for distribution (MV and LV) systems.
There are other numerical methods that might be more appropriate – especially if dynamics
and harmonics analysis are important.
Recognition of the shortcomings of the classic power flow methods to handle such things as the
interconnection of DER has led to much research since the early 1990’s into new ways to write
the equations describing the power system to accommodate the needs of the modern grid. Some
of that research is described in this chapter.
Figure 2-1 illustrates a few of the unbalances on a distribution system that cannot be adequately
addressed by a balanced positive-sequence equivalent. The distribution feeder shown is a 4-wire
multi-grounded neutral system in which the neutral is active in the power flow and short circuit
behavior. In the US there may be 3-phase, 2-phase (sometimes called “V-phase”), and 1-phase
feeders as well as multiple feeders on the same structure sharing one neutral conductor. Once the

10253878 2-1
current gets on the neutral, the delta-wye transformer and its neutral reactor get involved. This
diagram also indicates loads connected both line-to-neutral and line-to-line in the same system.
There are usually multiple capacitor banks on a feeder. The diagram shows a common condition
in which one phase of the capacitor bank has a blown fuse. This can stymie distribution state
estimation algorithms. Finally, there is an open-delta regulator bank at the end of the feeder,
which presents a significant challenge to even the most capable distribution system modeling
software.

Blown Fuse

Figure 2-1
Common unbalances in the distribution system that require more than a positive-sequence model

Table 2-1 is a side-by-side comparison of the issues relevant to transmission system and
distribution system modeling for planning analysis
Table 2-1
Comparison of transmission system analysis to distribution system analysis

Transmission System Distribution System

Model is balanced (Positive Sequence)
Scale is MW, Mvar, MVA
Quantities (impedances, voltage, and current)
expressed in per unit
100 MVA Base is common
Transformers eliminated by per unit system
Model is unbalanced; phase domain models
Scale is kW, kvar, and kVA
Voltage in kV or V; Current in amperes;
Impedances in ohms
10 MVA, if any base.
Transformers explicitly modeled

This demonstrates some of the circuit topology issues that arise when attempting to model
modern grid issues, but there are many others. Many of the advances in distribution power flow
methods since the early 1990’s were driven by the need to model distributed generation. As the
industry moves forward toward 2020, new problems such as including storage and modeling
microgrids will drive the advancements in distribution system analysis tools.
In the rest of this chapter, the traditional transmission-oriented power flow problem will be
presented. Then a traditional radial-circuit forward-backward sweep (FBS) method for solving
distribution system power flow is presented and the chapter concludes with a description of the
open-source EPRI OpenDSS program, which was formulated specifically to handle the demands
of modeling the modern distribution grid.

10253878 2-2
The Traditional Power Flow Problem Definition
We will first address the classic power flow definition for balanced transmission system analysis
and then address methods that have been developed for power flow analysis of unbalanced
distribution systems.
Swing Bus

|V| /φ

Power Delivery Network

I = [Y] V

Generator Bus
Voltage Magnitude,
P, |V|
Angle Known
Known

Load Bus
S = P + jQ Known

Figure 2-2
System model for conventional power flow problem definition

The power flow problem definition as presented in most basic power system analysis textbooks
generally defines three different kinds of buses.
1. Swing bus (or Slack bus)
a. At this bus, the voltage magnitude (expressed in pu) and angle are specified. That is, it is
a voltage source. The power in this source swings, or takes up the slack, to compensate
for deficits or excess power in load buses and generator buses.
b. One generator in the system is chosen to be the swing, or slack, bus and is dispatched to
keep the power balanced between the loads and the other generators.
2. Load bus
a. The active and reactive power consumed by the load is specified for a load bus.
b. The equation is S=P +jQ = VI* where ‘*’ denotes complex conjugate. The variables, S,
V, and I are complex numbers representing phasors in the steady state, or frequency
domain, as is common in power system analysis.
3. Generator bus
a. At a generator bus, the active power, P, and voltage magnitude, |V|, are specified.
i. P = Re{VI*}
ii. |V| = constant, expressed in pu
b. P and V are determined by some sort of optimal dispatch or unit commitment computer
program that determines the best dispatch of the generators. These functions may be built
into the modern power flow program for transmission system operation.
The power delivery network is usually assumed to be a balanced three-phase system and is
represented employing Symmetrical Components to create a single-phase equivalent in the
positive sequence. It is generally described by an admittance matrix (variously called Y bus or
simply Y).

10253878 2-3
With this model, the power flow problem is generally stated in terms of the mismatch in the
complex power, S, at each bus. The mismatch, M, at the j-th bus is expressed as
M j = -V j I j * + S j
Where
S j = specified power (active and reactive) at a Load bus
I j = j-th element of the system I vector from the network nodal admittance equation
I = YV
where the boldface type denotes a vector or matrix of complex numbers.
The object of the power flow solution technique is to minimize the mismatch at each bus. We
will look at three most common methods for solving the power flow problem as mentioned at the
beginning of this chapter. The description here is based on a concise description of these
techniques as presented by Heydt. [42]
Gauss-Seidel Solution Method
This is a straightforward technique of successive approximations. Many of the early power flow
programs used this method because it is simple to program.
The basic formulation can be written
𝑠 ∗
𝑣𝑘 = ��𝑣𝑘 � − ∑ 𝑦𝑗𝑗 𝑣𝑗 � /𝑦𝑘𝑘 (𝑗 ≠ 𝑘)
𝑘

The idea is to make an initial guess at the voltages and then iterate on this equation until the
voltages converge and the mismatch criteria are met.
A straight Gauss method would use two separate voltage vectors for the left- and right-hand
sides of this equation. The voltages used on the right-hand side would not be updated until the
iteration has cycled through all the buses. However, that has been found to have very slow
convergence characteristics. A Gauss-Seidel method starts with the slack bus where the voltage
is known and uses each successive bus voltage immediately as an improved estimate of the final
voltage. A program for this method would keep only one vector of voltages so that it always uses
the latest estimate in the equation for the next bus as the iteration cycles through the buses in
sequence.
There have been many modifications and enhancements to this method by many different people
over the years, adding acceleration factors, and the like. The Gauss-Seidel method is popular for
students learning about power flow solutions and writing their first power flow program. It is
easy to program and does not require the handling of large matrices. However, in practice today,
the Gauss-Seidel method for solving the power flow problem has been supplanted by Newton-
Raphson methods that offer improved convergence characteristics. There is a parallel to current
events in distribution system analysis in which older, traditional methods that do not require
handling of matrices and are relatively simple to program are giving way to methods that rely
heavily on large sparse matrices.

10253878 2-4
Newton-Raphson Solution Method
A Newton-Raphson method is an extension of the basic Newton’s method for solving a nonlinear
equation to a system of nonlinear equations.
If we have a simple equation we are attempting to solve that can be formulated as a function
equal to zero, such as
g(x n ) = 0
the Newton iteration to solve for x can be written:

xn + 1 = xn − J − 1 g ( xn)
 ∂g 
J = 
 ∂xn 
J is referred to as the Jacobian. It represents the rate of change of g with respect to x at the n-th
iteration and the slope should always be such that it points in the direction such that the function
g will be closer to zero.
For power flow equations, the common implementations split the mismatch equations into the
real and imaginary parts (P + jQ) and force the mismatch for each part to zero. This system of
equation can be written in vector form as

 Fp(δ , | V |)
G= =0
 Fq(δ , | V |)
where the unknowns are the voltage angle and magnitude vectors.

δ 
X = 
|V |
This set of equations is written for each bus. Therefore, the order of the system of equations to
solve is 2 times the number of buses, or 2N.
The jacobian for this system is then

 ∂Fp ∂Fp 
 ∂δ ∂V 
J = 
 ∂Fq ∂Fq 
 ∂δ ∂V 

For a system of power flow equations, each block of the jacobian is a NxN matrix.
For the full Newton iteration, the jacobian is updated at each iteration. For very large systems,
this can be time-consuming. Fortunately, for most cases, this method converges quickly in a few
iterations and converges to a solution quite well. In comparison, the Gauss-Seidel method may
take dozens of iterations and more frequently fails to find a solution. Various schemes have been
employed to improve the efficiency of the Newton-Raphson algorithm including the Fast

10253878 2-5
Decoupled Newton-Raphson method in the next section. Another approach is to not update the
jacobian at every iteration, which saves time with the matrix factorization. Some call this a
“dishonest” Newton method. Although the jacobian may not be precise, the derivatives tend to
point in the correct direction for a solution. It may simply take a few more iterations to get there.
This approach is perhaps more appropriate for distribution systems than for transmission
systems.
Fast Decoupled Newton-Raphson Solution Method
Stott and Alsac [43] are credited with developing this method. They observed that one could
eliminate the two off-diagonal blocks of the jacobian and the solution would still converge, This
leaves a jacobian of the following form, which obviously decouples the equations for the bus
voltage phase angle and the magnitude.

 ∂Fp 
 ∂δ 0
J = 
0 ∂Fq 
 ∂V 

This simplification takes advantage of the fact that many power systems – transmission grids in
particular – are reactance dominant. That is the X/R ratios are relatively high in general. In these
kinds of system, the relationship between P and V is weak as is the relationship between Q and
the phase angle, δ . This leads to the common assumption that the phase angle between
generator buses is what drives the active power, P, around the network while the reactive power,
Q, helps regulate the voltage. This is what is represented in the two remaining blocks of the
decoupled jacobian.
This simplification also takes advantage of the fact that Newton’s method will frequently
converge as long as the derivatives in the jacobian are pointed in the right direction even if the
jacobian is not precisely correct.
This method yields a dramatic reduction in the number of calculations that are required
compared to the full Newton-Raphson method. Therefore, it is quite popular today for
applications where computational speed is important and the decoupled method works.
This method does not work as well on many distribution systems. The resistance of the
distribution network is more prominent, making the off-diagonal blocks that were discarded
more influential in the solution. Therefore, the method may not converge. It needs a system with
a dominant reactance. This may be the case for overhead primary (MV) distribution lines but
underground cable lines are often have X/R < 1. It is also becoming more common to model
secondary (LV) distribution systems that are resistance dominate. These must be included in the
model to study the impact of the growing number of rooftop solar PV installations. So the Fast
Decoupled Power Flow is general not an option for distribution systems.

The Distribution System Power Flow

The power flow problem and methods of solution were largely developed for single-phase
positive-sequence equivalents of 3-phase transmission systems assumed to be balanced. Early
distribution power flows also used some of the same simplifying assumptions, but this had

10253878 2-6
changed by at least 1991 when Kersting published the first of the IEEE Radial Test Feeders.
These were full 3-phase, unbalanced power flow models. This, coupled with the expanding needs
to model distributed generation, forced the distribution system analysis software suppliers into
full 3-phase (or multiphase) unbalanced circuit modeling.

Radial Circuit Power Flow Calculations

Figure 2-3 is a one-line diagram of one feeder of a typical radial distribution system along with
key elements for power flow analysis
Substation
Transformer

Load Bus
Load Bus Constant Z
P + jQ
Swing Bus

|V|,θ
Load Bus
P, Constant X

Figure 2-3
Distribution power flow problem

The typical characteristics of a radial distribution circuit are

• Only one source
• Substation transformer may have a tap changer for voltage regulation
• Many different kinds of loads, e.g.:
- P + jQ (traditional power flow model)
- Constant impedance (Z)
- Motor: Constant P, Q varies quadratically
- Mixed: Linear P, quadratic Q
- Some Rectifiers: Constant P, constant |I|
- Many others …
• Unbalances
- 1-phase, 2-phase, and 3-phase loads
- Open-wye, Open-delta transformers and regulators
- 1-phase and 3-phase capacitor banks
- Etc.

10253878 2-7
Figure 2-4 shows the one-line sketch of a radial distribution system that has served as the
template for distribution power flow. There is a voltage source in which the voltage and angle
are specified, similarly to the slack bus in a transmission power flow. It is followed by a
substation transformer modeled by its leakage impedances and winding connections and may
have an on-load tap changer (OLTC). Loads are commonly treated as current injections of at
least 3 basic types: Constant power (P+jQ), Constant Z, Constant current, or some other type of
load model.
|V| /θ

I = (S/V)* I = V/Z I = ____

[Constant P, Q] [Constant Z] [Other Type]

Figure 2-4
Current injection model for representing loads and sources

This system is commonly solved by a forward-backward sweep (a ladder network technique) but
could also be solved by one of several formulations employing sparse matrices.
Forward-Backward Sweep
The Forward-Backward Sweep (FBS) solution method for radial circuits is comprised of two
steps: the Forward sweep and the Backward sweep as illustrated in Figure 2-5.
FORWARD
SWEEP

COMPUTE
VOLTAGE
DROPS

BACKWARD
SWEEP

ACCUMULATE
CURRENTS

Figure 2-5
Illustrating the common forward-backward sweep method for solving radial circuits

10253878 2-8
There are variety of forward-backward sweep formulations that will solve the radial circuit
power flow problem. We will describe a simple method here based on summing currents, but
there are formulations that sum the powers that achieve the same result. Kersting [44] is an oft-
cited reference source for multi-phase unbalanced forward-backward sweep methods.
The forward sweep starts with the known voltage at the source bus and computes the voltage
drops across each downline branch considering the current in that branch.
Vk = Vj − ZjkIjk

Where,
Bus k is immediately downline from Bus j.
Z jk = impedance of the branch, or line, between Buses j and k.
I jk = current in the branch, or line, between Buses j and k .
To do the sweep efficiently, programmers generally construct some sort of list that points to each
of the branches in the proper sequence. It may be a linked list or simply an ordered array in
which each element of the array points to a branch that is guaranteed to be downstream from its
parent branch. The program follows this list from top to bottom for the forward sweep and in the
opposite direction for the backward sweep to come later.
In the backward sweep, from the feeder extremities back to the main source, the current into each
load is computed based on the latest estimate of the bus voltages. Then the current is
accumulated from the loads back to the source by following the circuit tree structure in the
opposite direction from the forward sweep. There is only one path from each load bus to the
source in the backward sweep. There frequently are multiple forward paths from a bus.
Like the Gauss-Seidel power flow method, the FBS method is easy to program and no large
matrices are used. Each iteration is generally fast although many iterations may be required. The
method is not necessarily quick to converge although many iterations can be done in a short
time. Some sparse matrix methods, such as the one in EPRI’s OpenDSS program, can compete
with FBS methods, especially for very large circuits.
In words, the Forward-Backward Sweep algorithm is:
1. Assume all load currents are zero and all branch currents are zero.
2. Do the forward Sweep, computing the voltage drops from the source to the ends of the radial
circuit.
3. Compute new estimates of the load currents based on the improved estimate of the voltages
and do the backward Sweep to the source.
4. Repeat steps 2 and 3 until the stopping criterion is met (power flow solution converged).
This method will work for multiphase as well as single-phase models – simply let the V and I
represent n phase vectors and Z an nxn impedance matrix. Kersting [44] shows the details of
three-phase sweep methods. Branches are modeled with small matrices, but there are no large
matrices. Kersting also demonstrates how to handle a variety of transformer connections with the
sweep methods, which is tricky. Special handling is required when wye and delta windings are

10253878 2-9
mixed. This is a detail outside the scope of the present document; students and programmers
implementing this sweep method are referred to the reference.
One advantage of this method is that it naturally represents the effect of the time delay in voltage
regulators. The tap adjustment is made during the forward Sweep in the order the regulators are
encountered in the branches. Thus, the ones closest to the substation will change first, which is
the way the time delays are set in most line regulator applications. (With the advent of DER,
some utilities use a different approach.) This is significant when using sweep methods for quasi-
static time-series (QSTS) simulation for many modern grid analysis applications.
Load models can be just about any function that will return a consistent and realistic estimate of
load current as a function of voltage:
I = f(V).
It is common for software for distribution system analysis to provide three basic load models:
1. Constant impedance (I = V/Z)
2. Constant power (I = (S/V)*)
3. Constant current (I = S*/Vbase /( V/|V|)*)
This sweep method converges relatively easily on practical systems that are not overloaded. It
will typically converge in 3 to 10 iterations depending on the initial guess at starting voltage. If
the initial voltage estimate is close to the final solution, it may only take 1 or 2 iterations. This is
advantageous for QSTS simulations and such analyses as optimal capacitor or generator siting.
The method is related to fixed-point iteration methods for which the first derivative of the
function for computing the current, f’(V), should be less than 1. For distribution system models,
this basically translates into the equivalent shunt impedances for all loads must be significantly
greater than the short circuit impedances at all buses. This is generally not a problem when
circuits are not overloaded and have good voltage regulation. The equivalent shunt impedance is
generally more than 20 times the series impedance of the lines up to each load bus.
Note that this is not the case for harmonics analysis due to resonances, so sweep methods are not
good candidates for modeling harmonic current flow. This has a bearing on the methods used in
EPRI’s OpenDSS program, which was adapted from a harmonics flow solver, and will be
described later in this document.
When using the traditional constant power model, this method will generally fail to converge
when the voltage sags to approximately 70%. However, this should not be construed to mean
that the voltage on the radial distribution circuit will collapse at this level. When the voltage
drops below 80%, load characteristics change significantly with motor contactors dropping out,
etc. Voltages can sag much lower than 70% and recover in actual systems. One solution for the
analyst is to linearize the load model when the voltage goes outside a predefined normal band of
perhaps +/- 10%. Sweep methods will generally handle this by converting the load to constant
impedance at low voltages. Although the solution may not be precise, the iterative solution will
not “blow up” numerically, causing power flow solutions to crash.

10253878 2-10
A simple voltage magnitude convergence test is generally sufficient to determine solution
convergence. The solution can usually be considered converged if the voltage change between
iterations is less than 0.01% or 0.0001 pu. This is a common convergence tolerance for
distribution system power flow.
Per Units or Actual Values?
Sweep methods evolved from manual voltage drop calculations performed before the widespread
use of computers. Actual values in volts, amperes, and ohms were used in these calculations and
this has continued to influence the way modern distribution system analysis computer programs
work. Most commercial computer analysis programs developed by distribution engineers retain
this approach.
Computer programs for transmission system power flows use a positive-sequence circuit model
and the per unit system. Some early distribution power flow programs were adaptations of
transmission power flows and also used this kind of model. However, it is difficult to adapt that
approach to the unbalanced power systems encountered is distribution system analysis.
The per unit system was developed to avoid explicit modeling of transformer winding
configurations and ratios. Initially this to simplify calculations on transmission systems
containing multiple voltage levels. In combination with the positive-sequence model, this led to a
manageable single-phase equivalent of power system without multiple voltage levels and explicit
transformer models.
Unfortunately, several instances arise in distribution system modeling where it is necessary to
model the transformer explicitly in detail. With computerized power flow analysis it is no longer
necessary to make things easy for manual computation. Computers do not seem to mind making
the tedious detailed calculations. One instance is shown in Figure 2-6: the common “split-phase”
or “center-tapped” residential service transformer used throughout North America and also found
in other parts of the world. It is actually constructed as a 3-winding transformer as indicated. It
serves Loads A and B at 120 V and Load C at 240 V. Loads A and B would represent “lighting”
loads but could also include refrigerators, computers, and most loads less than 2000 W. Load C
would represent loads like clothes dryers and hot water heaters, which typically draw as much as
5000- 6000 W. Air conditioners and solar PV inverters are also typically connected to the 240 V
lines.

Figure 2-6
What is the voltage base for the LV side that would allow removing the explicit transformer
model?

10253878 2-11
As the figure caption asks, what voltage base could be used to allow a per-unit model of the
transformer. There are two choices: 120 V and 240 V. It turns out that whichever one is chosen a
“kluge,” or inelegant patch, is required to handle the other voltage base. Even then the winding
connection cannot be eliminated as might be done for 3-winding transformer in transmission
system models. The two secondary windings are in series while in transmission grid models of 3-
winding transformers, the windings are each in shunt with the grid.
It is not always necessary to model the center-tapped distribution service transformer in detail,
but when it is necessary, the per unit system cannot be used. Of course, symmetrical components
are not valid for 1-phase circuits.
Other examples requiring detailed transformer models that are difficult, or impossible, to model
with per units include:
1. Open-Wye/Open-Delta or Open-Delta/Open-Delta banks
2. Delta or Open-Delta banks with a center-tapped transformer in one leg of the Delta.
3. Transformer banks with units having unequal ratings.
4. Faults involving two voltage levels such as a fault between the distribution lines and the
subtransmission overbuild on the same pole structure.
5. Zig-zag transformers that also serve load.
The voltage bases in the per-unit model are dictated by voltage rating and turns ratio of the
transformers in the model. When a radial distribution system containing tap-changing
transformers is modeled in a per unit system, the voltage bases should ideally be adjusted for
each tap change. If a forward-backward sweep method is employed an algorithm must be
devised for sweeping forward to adjust the voltage bases for each line, load, and capacitor before
making voltage drop calculations. This can be difficult to program and time-consuming to
execute. None of this is necessary when using actual values.
One objection formerly raised to using actual values instead of per units was actual value models
led to numerical inaccuracies. The claim is that using the per unit system naturally scaled the
impedances so that admittance and impedance matrices were less likely to be ill-conditioned.
That idea has its basis in the earlier days of computing when many engineers only had access to
computers with 32-bit floating point arithmetic. This concern is no longer valid for two reasons:
1. Common desktop computers use IEEE 80-bit floating point arithmetic, giving more than
enough precision the model the power system all the way from EHV to LV in one model,
2. Modern sparse matrix solvers automatically perform scaling to minimize numerical issues in
distribution system analysis.
EPRI’s OpenDSS program exploits the 80-bit math capability often found in standard computers
and saves all results as 64-bit floating-point (double precision) values by default. This is now a
common approach in modern distribution system analysis programs.
In summary, the per-unit system is not needed for distribution system analysis – EPRI’s
OpenDSS program does not use it and it can perform all the standard frequency-domain (phasor-
based) analyses commonly performed in distribution planning. Some distribution system analysis
problems, such as a 3-phase transformer bank with unequally-sized units, are actually easier to

10253878 2-12
represent in ohmic values and actual turns ratios than with per-unit models. Distribution system
models for harmonics studies are generally easier to construct in actual values than in per-unit
models. This is due to all the adjustments that have to be made for frequency and the detail
necessary in line, transformer, and load models. Of course, EMT solvers have used actual values
since the 1960’s.

Quantities Required for Distribution Power Flow Models

Line Impedance Models
Possible approaches for modeling the impedances of power lines and cables for distribution
power flow include:
1. Positive-sequence values only.
2. Positive- and zero-sequence impedance values.
3. Full phase-domain, n-phase impedance matrices.
Positive-Sequence Only
The positive-sequence only approach is the same as the approach for line models in transmission
system power flow programs where the 3-phase system is assumed to be balanced. In
distribution usage the values tend to be in actual ohms and siemens in contrast to per-unit
quantities used almost exclusively in transmission system analysis programs. The main problem
with this approach is that most distribution systems, at least in North America, are not balanced
and are not entirely 3-phase lines. European-style systems have more balanced MV, or primary,
distribution feeders and it is possible to get reasonable results with positive-sequence-only
models. However, the 400-V LV system is often very unbalanced and, since more LV analysis is
now being done, a more detailed approach is required.
Positive- and Zero-Sequence Symmetrical Components
The second choice is the most common for many distribution system analysis tools. Lines and
other elements of the circuit are described by positive- and zero-sequence values, e.g., Z 1 and Z 0 .
For lines and many other elements, the positive- and negative-sequence values, Z 1 and Z 2, are
equal, so it is only necessary to specify Z 1 . A balanced impedance matrix can be constructed
from Z 1 and Z 0 in the form shown in Figure 2-7.

 Zs Zm Zm 
Z =  Zm Zs Zm
 Zm Zm Zs 
Figure 2-7
Balanced, symmetrical impedance matrix constructed from symmetrical component values
Z 1 and Z 0

10253878 2-13
Some handy formulae for converting between (Z 1, Z 0 ) and (Z s, Z m ) are
If given Z 1, Z 0 :
Zs = [2Z1 + Z 0] / 3
Zm = [Z 0 − Z1] / 3

If given Z s, Z m from a balanced, symmetrical Z matrix:

Z 1 = Zs − Zm
Z 0 = Zs + 2 Zm

Note that although this method includes the zero-sequence impedances, it does not represent
unbalance in the lines. Only the balanced line impedance matrix can be represented. This implies
a continuously-transposed line, or at least a line in which the three phases are in a triangular
configuration. If the line is constructed untransposed on a horizontal crossarm, as many are, there
will be phenomena that can be simulated only by using the full unbalanced impedance matrix.
One example of this is the “hot-phase/cold-phase” effect of the horizontal crossarm that has been
observed on some systems where one of the outside phases conducts more power than the
opposite outside phase. As the power ramps up on this system, the voltage on one outside phase
has a tendency to increase while the opposite outside phase tends to decrease. Thus, voltage
regulators on the outside phases often have more tap changes than the one on the middle phase.
Simulations on a balanced matrix model will not show this effect.
One reason this method of describing the lines is preferred by distribution engineers is that the Z 1
and Z 0 values can be plugged directly into the short circuit formulae. For example, for a single-
line-to-ground fault on a 3-phase system, the fault current formula is:

3𝑉1
𝐼𝑓𝑓𝑓𝑓𝑓 =
2𝑍1 + 𝑍0
Performing a standard short circuit study on a radial circuit simply means accumulating the
sequence impedances from the source to each bus and plugging them into the formulae. This has
led to the practice in some computer tools of assigning Z 1 and Z 0 values to 1-phase and 2-phase
line segments so that they, too, may be plugged directly into the formulae for the 1-phase and 2-
phase faults that appear on those segments. This is unfortunate because the symmetrical-
component impedances of 1-phase and 2-phase lines are undefined and there is no standard way
to compute them. It is common to find two different computer programs handling these
impedances differently in incompatible ways.
Equivalent 4-wire Line Model from Z1 and Z0 values
Readers may find it necessary at times to construct a 4-wire line model including the neutral
conductor when Z 1 and Z 0 values are the only data available. Figure 2-8 illustrates how this
model is constructed. The three phase conductors are modeled exclusively by Z 1 and the 4th
conductor is comprised of both Z 1 and Z 0 . This model was extensively employed on analog
models such as found in transient network analyzers (TNA). If one were to write a loop equation
through one of the phases and back through the neutral/ground conductor, the value for Z s would

10253878 2-14
be found. The impedance in the neutral, is identical to Z m . This impedance contains not only the
effect of the neutral conductor but the earth return path as well, as is documented in many power
system analysis textbooks.
Z1

Phases

(Z0 - Z1)/3
Neutral +
Ground

Figure 2-8
4-wire line model constructed from symmetrical component impedances

Full Phase-Domain Impedance Matrix Models

Most commercial distribution system analysis programs now support full 1-, 2- and 3-phase
unbalanced impedance models of lines and cables. Some, such as EPRI’s OpenDSS program,
allow unbalanced models of arbitrary size (n-phase models) to be represented. For example, the
IEEE NEV distribution test feeder requires a program to model 17 coupled conductors to
compute stray neutral-to-earth voltages due to currents flowing in various neutral conductors.
The programs typically accept
• Series impedance and shunt capacitance data in matrix form, typically in units of ohms and
siemens (or farads).
• Line geometry data from which the program computes the matrices directly. With this
approach it is easier to accommodate advanced applications such as harmonics analysis that
requires frequency-dependent models.
The full phase-domain modeling approach to line modeling allows detailed and accurate
modeling of
• 1-phase laterals on 3-phase systems
• Laterals consisting of 2 phases of a 3-phase system
• 2, or more, 3-phase distribution circuits sharing the same pole, or tower, and possibly sharing
neutral wires
• 2, or more, 3-phase circuits of different voltage levels on the same pole, or tower, structure.
(per-unit values are generally not applicable to this problem).
Transformer Data Required
The most important data for representing transformers in power flow studies are generally the
leakage, or short-circuit, impedances. These are the series impedances of a transformer
responsible for the voltage drop (or rise) across the transformer. The shunt impedances of a
transformers are often neglected for power flow when performing capacity analysis, but may be

10253878 2-15
important for some calculations. For example, the core losses (no-load losses) are important for
studies of power delivery efficiency that involve simulations over a significant period of time
such as a year. While series impedance losses may dominate at peak load, the loading is closer to
minimum loading for most of the year. It should not be surprising to find that no-load energy
losses are dominant in an annual simulation.
The impedances for the transformer are the short-circuit impedances between each pair of
windings. For a transformer is n windings the number of short circuit impedance measurements
required is
𝑛(𝑛 − 1)
𝑛𝑛𝑛𝑛𝑛𝑛 𝑜𝑜 𝑠ℎ𝑜𝑜𝑜 𝑐𝑐𝑐𝑐𝑐𝑐𝑐 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 =
2
These impedance values are generally expressed in percent on a winding’s base MVA rating.
The test report or nameplate must be carefully studied if the transformer has more than 2
windings to ensure the correct base is used. The impedance in ohms looking into a winding is
2
𝑍% 𝑘𝑘𝐿𝐿
𝑍𝑜ℎ𝑚𝑚 =
100 𝑀𝑀𝑀3−𝑝ℎ𝑎𝑎𝑎

Most modern distribution system analysis program will accept the data in percent or per unit and
perform the conversion to ohms automatically.
Winding connection configurations are essential to model when performing 3-phase power flows
in the phase domain. This is how the phase shift for transformers with delta and wye windings is
determined. For positive-sequence single-phase equivalents, the phase shift in mixed delta-wye
transformers is modeled by multiplying the short-circuit impedance by the exponential phase
shift operator ejθ. This is generally not convenient in 3-phase models and is not necessary if the
transformers are modeled explicitly.
Transmission System Source Equivalent
The distribution system model is generally connected to the transmission system or bulk power
system at one point. Distribution system analysis tools model the bulk power source in one of
two ways:
1. As an ideal voltage source, assuming the bus at the head of the feeder is fixed,
2. As a short circuit (Thevenin) equivalent (Figure 2-9).
The former is often adequate for power flow capacity studies where it is assumed that the
substation OLTC or feeder regulator will be able to regulate the voltage to the specified value.
Note that several of the IEEE Test Feeders take this approach. However, the latter approach is
necessary for many modern grid applications and for performing short circuit studies.

10253878 2-16
 Z (matrix)

3-phase
voltage
source

Figure 2-9
Bulk power system source modeled as a 3-phase short-circuit equivalent

Utilities can provide the data for the short-circuit equivalent in one or more ways, including:
• Short circuit MVA, 3-phase and 1-phase short circuits.
• Short circuit current, amps, 3-phase and 1-phase short circuits.
• Z 1 and Z 0 , in ohms (more common for distribution)
• Z 1 and Z 0 , in pu (more common for transmission). The MVA base is generally 100, but some
distribution system analysis programs assume 10 MVA as the base. It is important to verify
the base.
For models employing positive-sequence equivalents, this equivalent is a single-phase source
and the impedance is Z 1 . For 3-phase models, a grounded-wye source as shown is commonly
assumed. The short circuit impedance, Z, is often expressed as a coupled, symmetric matrix. If
the head of a distribution system is electrically close to a large generator (or the main generator
of a microgrid) it may be necessary for the short circuit impedance matrix to be asymmetric
because the negative-sequence source impedance is different than the positive-sequence
impedance. (For lines and transformers, the positive- and negative-sequence impedances are
equal.)
Most transmission system equivalents are generally Y-grounded systems and the Thevenin
source can be considered grounded. However, some lower-voltage subtransmission systems are
3-wire delta systems and the source equivalent should also be connected in delta, if the computer
software permits it. Some software requires a grounded source. It can be transformed to an
ungrounded source using a wye-delta transformer model.
Shunt Capacitors
Many distribution systems contain numerous shunt capacitor banks for power factor correction
and voltage support. It is important to model these correctly to properly represent the reactive
power flow in a power flow solution.
Capacitors are rated by kvar or Mvar, but are not constant power devices as this rating might
imply. They are constant impedance or admittance devices. The capacitance of a power factor
correction capacitor is not a function of the voltage applied, but the reactive power output is. It is
not a nonlinear device for which an iterative solution technique must be applied. In fact, it is one
of the most linear elements encountered on the power system and should be modeled as a
constant impedance or admittance for power flow analysis.

10253878 2-17
Data required:
• Mvar or kvar rating and the corresponding voltage rating (important!).
• The voltage rating of a 3-phase capacitor bank is often stated in line-to-line kV, but could
also be specified by the rating of each can. The majority of capacitor banks on the
distribution system are wye-connected, so the line-to-neutral kV rating is also common.
• Connection of the bank: Grounded-wye, ungrounded-wye, or delta. Each connection has a
unique impact on the solution.
• Some capacitor banks are configured as harmonic filters with a series reactor in each phase
or a reactor connecting the neutral to ground. These usually have a minor effect on the
fundamental frequency power flow, but can raise the apparent kvar output of capacitor bank,
which could be important.
• Losses in capacitor banks are generally ignored for fundamental frequency power flow
solutions, being quite low. It may be important to include losses for harmonic studies.
Capacitor bank switch controls may or may not be included in the power flow analysis. If
solving only the peak load case, the capacitor banks are generally assumed to be energized and it
is not necessary to model the control. That is, the capacitor is modeled as a fixed bank that is un-
switched. The controls must be modeled if a QSTS simulation is to be performed over several
days during which the capacitor would be expected to switch. The switching of a capacitor bank
often requires the rebuilding of a Jacobian, or other matrix, and additional iterations to solve the
power flow.
Modern capacitor switch controls can have a myriad of settings meant to overcome specific
application problems. The typical quantities for control include:
• Time: Switches the capacitors on each day at the time when the load is expected to require
significant reactive power support. This is typically ON in the morning and OFF in the
evening.
• Temperature: For systems in which the load can be easily correlated to temperature – such as
one with a large amount of air conditioning load – this control mode is used to switch the
capacitors.
• Voltage: The capacitor is switched ON when the voltage drops below a set point and OFF
when the voltage rises above a set point.
• Current: The capacitor is switched ON when the load current rises above a certain level and
OFF when it drops below another level.
• kvar: Similar to Current control except that the voltage is also sensed and the reactive power
is used to switch the capacitor.
• Power Factor: This control mode is more frequently applied to substation control than to
local bank control because the correlation between load and power factor is more predictable
• Voltage Override: Many capacitor switch controls employ this function to override the
control action requested by any of the other control modes. The capacitor is switched OFF if
the voltage is too high and ON when the voltage is too low, which is the opposite of the
Voltage control mode.

10253878 2-18
Figure 2-10 illustrates the typical electrical connections of capacitor switch control and the
corresponding switched capacitor. If the control uses current or kvar to determine switching, the
current is measured on the Load side of the capacitor. Otherwise the current and kvar would
change significantly each time the capacitor switched and it would be more difficult to determine
settings unless the control is designed for such an application. For power flow, the topology
shown in the figure is usually the simplest to model.
CT

PT

LOAD
CapControl

Figure 2-10
Capacitor control monitoring load side voltages and currents

Some capacitor banks are switched by centralized controls in a distribution management system
(DMS). The control action may be dictated more by what is happening on the transmission
system that the distribution system. It may be manually switched by an operator. It can be
difficult to capture this effect properly in a QSTS simulation and analysts will generally resort to
a kvar or power factor control mode assuming that the transmission system need can be
approximately correlated to these quantities.
Capacitor controls, like voltage regulator controls, also have time delay settings that dictate the
order in which the devices switch. This is important to model for QSTS simulations, but is also
important if static power flow solutions are being done at various load levels. Capacitors are
frequently timed so that the ones farther away from the substation switch ON first and OFF last.
For the power flow solution algorithm to automatically determine which capacitors are ON there
must be some kind of list that is sorted according to delay times. (That is the approach in EPRI’s
OpenDSS program.)
Voltage Regulators or Substation OLTC
Many distribution systems have tap-changing transformers called voltage regulators or substation
on-load tap changing (OLTC) transformers to provide better control over the voltage magnitude.
It is important to represent them in a power flow analysis to get the correct result. This is
particularly true for long, weak distribution feeders that might have two or three regulators in
series. Also, the appearance of distributed renewable generation, such as solar PV generation, on
the modern distribution system has caused much concern about the impact of this fluctuating
form of generation on the life of the tap-changing devices. The numerous and rapid voltage
changes that occur can cause the regulator to increase the number of tap changes and thus reduce
the life of these mechanical devices.

10253878 2-19
Regulators and OLTC are comprised of transformers with tapped windings, a switching
mechanism for changing taps, and a control for initiating the tap changes. A substation OLTC is
a conventional 3-phase substation transformer with standard impedance, commonly connected
delta/wye-grounded in North America, and having sufficient MVA capacity to serve the load on
1-6 feeders. The tapped winding is often the MV winding, but could also be the HV winding
depending on the design. The tapped winding typically has 8 taps in addition to the neutral
position. Combined with a switching mechanism that can bridge two taps and a small series
transformer sometimes called a “preventive autotransformer” there will be a total of 16 tap
positions. Adding a reversing switch allows for 16 taps in the opposite direction. This gives rise
to the common “32-step regulator” with 16 raise taps and 16 lower taps. The typical range is +/-
10% regulation, although there are exceptions to that rule. Thus, each tap step represents 5/8%
voltage regulation. Each phase is on the same tap position (ganged operation).
“Voltage regulators” in North America are typically single-phase autotransformers that may be
applied singly or in a bank of 3 regulators. Regulators also typically have 32 steps like substation
LTCs. They may be installed in substations on individual feeders where they are referred to as
“feeder regulators” (Figure 2-11). In this arrangement each phase has its own control and each
regulator may be on different taps. Of course, this requires a full 3-phase solution algorithm to
properly model this. Voltage regulators are also applied on the distribution feeder lines, often
mounted on poles.

Figure 2-11
Bank of three 1-phase voltage regulators deployed as feeder regulators (Courtesy of Eaton, used
by permission)

10253878 2-20
The characteristics of the transformer in 1-phase regulators are much different than the 3-phase
transformer. Since it is basically an autotransformer and is only called upon to transform 10% of
the line voltage, the transformer windings have only 10% of the kVA rating of the load that can
be regulated. That is, a regulator bank designed to regulate 5000 kVA of load is built as if it were
a 500 kVA transformer. The impedance of the regulator is also quite low compared to a standard
2-winding 5 MVA transformer. In fact, some distribution power flow programs neglect the
impedance of voltage regulators and consider only the tap ratios. However, the impedance is not
zero and may appear significant in a short-circuit analysis. The range is approximately 0 to 0.5%
on its through-kVA rating, depending on tap position and design details.
Both substation OLTC and line voltage regulators employ similar control logic. For power flow
purposes, the controls can generally be treated the same. One feature that can be a challenge to
model in a power flow is the “line-drop compensator,” or LDC. The reader is referred to the
instruction manuals provided by manufactures for more details. The idea of the LDC is the move
the regulation point downstream from the regulator position and thus flatten the voltage profile.
This is done conceptually by compensating for a certain amount of line impedance by mimicking
the line drop using a small model circuit in the control. The LDC control settings are labelled R
and X leaving the impression that they represent resistance and reactance of the line – until one
notices that the units on R and X are volts! The most straightforward way to handle the LDC in a
distribution power flow algorithm is as follows.
Regulators are generally rated in amperes and come with a CT with the same or similar rating.
Assuming R=2.5 V and X=3 V, when the current in the regulator reaches the CT rating, the
compensation circuit subtracts 2.5 V resistive and 3 V in quadrature as if due to a reactance from
the voltage measured by the potential transformer (PT) at the regulator location. This forces the
regulator tap higher to compensate for the presumed line voltage drop to the regulation point.
The reader is referred to Kersting [44] who elaborates on this in great detail for the FBS power
flow algorithm. Similar logic may be employed for other power flow algorithms. The LDC
settings in regulators are employed heavily in the IEEE Test Feeders [45] that Kersting and
others have submitted. Therefore, researchers wanting to validate their distribution power flow
methods against these test cases will have to model regulators with LDC.
Incorporating the LDC in the model requires the following data in addition to the tap changer:
CT primary rating (not ratio), PT ratio, R and X settings. If a reversible regulator, the reverse-R
and reverse-X settings are required as well as the power threshold at which the regulator
reverses.
A tap change effectively alters the voltage bases downline from a regulator or substation OLTC.
This requires some special fixups for FBS methods implemented in per unit values. However, if
the model is in actual values, only the impedance and taps of the tap-changing device needs to be
updated. This is one of the advantages of working in actual values. For algorithms employing
admittance or impedance matrices, the system matrix must generally be rebuilt whether in per
unit values or actual ohms/siemens.
Tap-changing devices also have time delays similar to capacitor controls. However, unlike
capacitor controls the ones closer to the substation tend to be switched first. FBS power flow
algorithms have an advantage modeling regulators because they can naturally incorporate the
effect of the usual time delays by executing the tap change in the forward sweep. This will

10253878 2-21
change the taps on the devices closer to the substation first and then execute the downline tap
changes if necessary. Other algorithms must employ other programming constructs to force the
tap changing to occur in the proper order.
There are two types of time delays associated with these devices:
1. The time delay before any tapping action is taken. This is usually on the order of 30-60 s.
2. The time delay between tap changes after the first. This is generally about 2 s although it
could be must faster on tap changers with fast runback features.
When there are several regulators and/or OLTC controls on a system it can be tricky for a power
flow algorithm to properly determine which taps they should be on. One approach is to assume
the taps are continuous rather than discrete. Then the values are rounded to the nearest integer
after the solution is converged. Another approach is to simulate the tap changes one at a time,
converging the power flow solution at each tap value until the voltage measured by the regulator
control gets back in band.
In a QSTS simulation, if the voltage comes back in band before the timer expires, the regulator
control will reset the main timer.
These approaches yield slightly different answers that have minimal impact on planning
decisions. However, they cause problems for researchers trying to benchmark against published
test cases such as the IEEE Test Feeders [45]. The discrete tap approach is adopted by the EPRI
OpenDSS program. It frequently gives results that are one tap different than results from
algorithms that compute the tap required to get to the center of the band assuming a continuous
control. If it starts from a tap position below the final tap, it will typically stop one tap lower; if
starting from above the final tap, it will stop one tap higher.
What is the correct answer for a power flow solution? There are often 3 tap positions on a
standard 32-step regulator that will produce a converged voltage within band. Each tap step is
5/8% of rated voltage. So the difference in the solutions is likely inconsequential with respect to
planning decisions. Researchers trying to benchmark their power flow algorithms might prefer to
get a definitive answer, but the real answer is ambiguous due to variation in source voltage and
other variables in the model. The goal should be to get the regulator/OLTC model to behave
similarly to the real thing. This is important for such studies as those to determine the number of
tap changes due to variation from solar PV generation.
Readers interested in learning how to implement a tap-changing transformer model with controls
that include the LDC are encouraged to study the open-source OpenDSS software code [46] that
EPRI has made public to advance grid modernization efforts. This program is the subject of the
next section.

OpenDSS Power Flow Solution Method

EPRI’s OpenDSS program is presented here as an example of an approach to the distribution
system power flow problem that was developed in the mid-1990’s in response to grid
modernization needs – specifically, to handle distributed generation in distribution planning. It is
formulated a bit differently than most power flow programs to achieve greater modeling
flexibility. In fact, its heritage is from a line of network harmonics analysis programs. Harmonics

10253878 2-22
flow analysis requires detailed modeling of multiphase power systems, so this heritage gives
OpenDSS extraordinary circuit modeling capability. While performing a power flow analysis is
the most common usage, it can be used for many other types of analyses including dynamics,
geomagnetically induced current (GIC), short circuit, electromagnetic interference (EMI),
transformer frequency response, stray voltage/current, transmission line ground voltage rise,
hybrid power/communications, losses, voltage sags, and harmonics flow in networks. It is
formulated like a power system harmonics analysis program, which is more like electromagnetic
transients (EMT) programs and dynamics programs than a traditional power flow.
With respect to power flow analysis, one thing that is unique about OpenDSS is that there are no
special buses. Buses are simply places where circuit elements are connected together and are
defined as collections of nodes (Figure 2-12). Note that “bus” and “node” are not synonymous as
in many power systems analysis programs. The object-oriented concept of “a bus has nodes” is
employed. That is, a bus consists of one or more nodes, which are numbered somewhat
arbitrarily as the user chooses. Conductors at the terminals of circuit elements (lines,
transformers, sources, loads, etc.) are connected to nodes and the equations for solving the power
flow are formulated to determine the voltages to ground (zero-volt reference) at each node.

Figure 2-12
OpenDSS bus and node definitions

As documented previously in this chapter most power flow programs are designed with the
concept that “a bus has load” with the load being defined in a limited number of ways. In
contrast, the corresponding concept in OpenDSS is “a load has a bus.” This may seem a subtle
difference, but one that is important to how the program is constructed.
A load is simply another circuit element like many others. It is a “power conversion” (PC)
element that is usually a nonlinear circuit element for a power flow solution. In contrast, lines,
transformers, and capacitors are “power delivery” (PD) elements that are modeled as linear
elements fully represented by their primitive nodal admittance equations.
The main requirement for a power conversion element is that it can return an estimate of the
current given some function of voltage, V, at its terminals and its internal state, β:
I = f(V, β)
This allows much more flexibility in modeling loads than typically found in power flow
formulations in which load is a property of the bus. There can be any number of load models
connected to the nodes at a bus in nearly any fashion. This has proven very useful for research
into planning methods for the modern grid in which new technologies are being considered.

10253878 2-23
 Connect to bus nodes

I = f(V, β)
Power Conversion
Element

Figure 2-13
Loads and other power conversion elements are typically modeled as being in shunt with the
power system and have one or more conductors that are connected to nodes at buses

As of this writing, there are 8 load models available in OpenDSS to represent how the active and
reactive power varies with voltage at the load:
1. Standard constant P+jQ load. (Default)
2. Constant impedance load.
3. Const P, Quadratic Q (like a motor).
4. Nominal Linear P, Quadratic Q (feeder mix). Use this model with CVRfactor.
5. Constant Current Magnitude.
6. Const P, Fixed Q.
7. Const P, Fixed Impedance Q.
8. Special ZIP load model.
These models can generally be mixed in any fashion with minimal effect on the power flow
solution convergence. There is no limit on how many Load elements can be connected to a bus,
so users can disaggregate the load into as many elements as desired and feasible. For example,
many studies have been performed with an aggregate load representing a home and a separate
Load object for an electric vehicle (EV) that has an entirely different load characteristic.
A Load object is a circuit element with a number of conductors in its one terminal that may be
connected in nearly any manner to a bus. Line-to-line and line-to-neutral-connected loads can be
connected at the same bus. While most distribution power flows are conducted with 1-, 2-, and 3-
phase elements, higher phase order models are also possible.
The reader is referred to the OpenDSS user manual for more details of the program. [46] The
remainder of this section is devoted specifically to how OpenDSS conducts the power flow
solution.

10253878 2-24
The Math …
Nearly all variables in the formulation result in a matrix or an array (vector) to represent a
multiphase system. Many of the variables are complex numbers representing the common phasor
notation used in frequency-domain ac power system analysis.
OpenDSS uses a fairly standard Nodal Admittance formulation that can be found documented in
many basic power system analysis texts. The Arrillaga and Watson text [67] is useful for
understanding this because it also develops the admittance models for harmonics analysis
similarly to how OpenDSS is formulated.
A primitive admittance matrix, Y prim , is computed for each circuit element in the model. These
small matrices are used to construct the main system admittance matrix, Y system , that knits the
circuit model together. The solution is mainly focused on solving the nonlinear system
admittance equation of the form:
I PC (V) = Y system V
where,
I PC (V) = compensation currents from Power Conversion (PC) elements in the circuit
The currents injected into the circuit from the PC elements, I PC (V), are a function of voltage as
indicated and represent the nonlinear portion of the currents from Load, Generator, PV system,
and Storage elements in the circuit.
There are a number of ways this set of nonlinear equations could be solved. The most popular
way in OpenDSS is a simple fixed point method that can be written concisely:
V n+1 = [Y system]-1 I PC (V n ) n = 0, 1, 2, … until converged
In words, after building Y system, start with a guess at the system voltage vector, V 0 , and compute
the compensation currents from each PC element to populate the I PC vector. Using a sparse
matrix solver, compute the new estimate of V n+1 . Repeat this process until a convergence
criterion is met.
If this seems familiar, it is because this iterative process mimics the FBS method described
earlier in this document. Instead of sweeping through the circuit line-by-line, the solution is
accomplished for all buses simultaneously using the sparse matrix solver. Instead of
accumulating the total load injection current at each bus, part of the load model is represented as
linear (constant Z) and included in Y system . Thus, this in naturally included in the solution. I PC
contains the difference between the total load current and the current in the linear part for all PC
elements connected to each bus. The OpenDSS method has better convergence characteristics
than the simple FBS for a wider range of distribution system models and has the added
advantage that arbitrarily-meshed networks can be solved as easily as radial circuits.
The fixed point iteration needs a good starting point to converge. Including the linear part of the
load and generator models in Y system makes it easier to get a good starting point for V 0 and keeps
the intermediate iterations closer to the final solution. Once a QSTS simulation has begun, the
solution at one time step is an excellent starting point for the next. For most time steps, the power
flow will converge in just two iterations – one to get to the next solution and one to prove the

10253878 2-25
solution converged. There is now an option for just doing 1 iteration per time step if that is all
that is needed.
Y system is not changed very often. It is updated for tap changes and capacitor switching and other
switch operations. Updating I PC at each iteration and performing back substitution step of the
sparse solve is sufficient to achieve a converged solution in the vast majority of cases. This
avoids the time-consuming matrix factorization.
Keep in mind that OpenDSS works in the phase domain, in actual volts and amperes.
Symmetrical components and per units are not used inside the program (only for input and
output). Therefore, the Y prim and Y system matrices are also constructed in actual units (siemens).
Yprim Examples
To understand how a primitive Y matrix is constructed, consider a simple resistor as shown in
Figure 2-14. Instead of simply writing Ohm’s law, we define the resistor as a two-terminal
element represented by a 2x2 system of nodal admittance equations. The voltages at each
terminal are defined with respect to the zero voltage reference, which is commonly referred to a
“ground,” or remote earth. In the nodal admittance reference frame all node voltage are defined
with respect to this reference. Currents are defined as positive going into the terminal. Thus,
when complex powers are computed at each terminal from S = P +jQ =VI* the positive
direction on P and Q will also be into the terminal.
R = 1/G
I1 I2

V1 V2

I1 G -G V1
=
I2 -G G V2

Yprim
Figure 2-14
Primitive Y matrix for a simple resistor

The elements of Y prim are computed from the following rules:

• The diagonal elements are the sum of all admittances connected to the corresponding bus
• The off-diagonal elements are the negative of the admittance between the two buses
involved.
To understand how this might be done for a more complicated circuit element, consider the
single-phase pi-section model of a line in Figure 2-15.

10253878 2-26
 I1 R + jX I2

jB1 jB2
V1 V2

I1 (R+jX)-1 + jB1 -(R+jX)-1 V1

=
-(R+jX)-1 (R+jX)-1 + jB2 V2
I2

Yprim
Figure 2-15
Primitive Y matrix for a pi-section line model

This is very similar to the simple resistor in Figure 2-14 except that the model contains X as well
as R and has half the susceptance of the line’s capacitance at each end of the line section.
Following the rules described above results in the primitive Y matrix shown.
This is basically how OpenDSS computes the Y prim matrices for each element in the circuit.
OpenDSS defaults to 3-phase models instead of the 1-phase models illustrated. The expansion of
the 1-phase model to 3-phases is straightforward: In the equations shown, let R, X, B, G, C, etc.
simply represent 3x3 matrices and I1, I2, V1, V2 become 3x1 vectors and the notation stays the
same. Now the model is for a 3-phase system and can even be scaled to an N-phase system with
NxN matrices. This is precisely what was done in OpenDSS when it was created. It employs
relatively straightforward linear algebra techniques to construct a multiphase model of an
arbitrary electric power system and solves it.
The OpenDSS Network Model
The several Y prim matrices are used to create the Y system matrix. This is illustrated in Figure 2-16.
Each element of each Y prim matrix maps into one and only one element of Y system . The elements
are summed into Y system . For linear circuit elements, that is all that is required to adequately
represent them in the system. Nonlinear elements (most power conversion elements are
nonlinear) such as loads, generators, etc., are converted into Norton equivalents with a linear
admittance being included in Y system and a variable injection, or compensation, current source
representing the nonlinear characteristic of the element. As illustrated, voltage sources are
converted from Thevenin equivalent to Norton equivalents and added to the system. This is
frequently the source of energy for distribution system analysis. Current sources, if any, can be
connected directly to any node in the network.

10253878 2-27
 Linear Part of Loads Included in YSYSTEM

VSOURCE

YSYSTEM
(Norton Equiv.)

Injection (Compensation) Currents from Loads, Generators, etc.

(Power Conversion Elements)

Figure 2-16
OpenDSS network model

Figure 2-17 show the basic concept behind the load models used in the OpenDSS power flow
algorithm. Like the basic power flow template in Figure 2-4, current sources make up an
important part of the power flow. A portion of the load’s complex power value is assigned to the
linear element that is included in Y system . The amount of power assigned to the linear portion is
not critical, experience has shown that values close to whatever is considered to be rated load, or
peak load, generally work better because the compensation current does not have to be very large
to achieve the desired terminal power for off-nominal voltages.

Figure 2-17
Basic concept for modeling nonlinear power conversion elements like loads

10253878 2-28
Figure 2-18
OpenDSS solution process illustrated

Figure 2-18 is a pictorial representation of the simple fixed-point iteration process for solving the
circuit model. First, the system Y matrix is constructed and an initial guess at the voltage vector,
V, is made. The simplest way to do that is to do a direct solution of the network with no
compensation currents in the I inj vector except for voltage and current sources. Thus, loads are
represented as constant impedances. This gets the initial voltage estimate sufficiently close to the
final answer for this process to converge quickly for most cases.
One advantage of this method is that nearly all load, generator, and transformer configurations
can be handled without doing anything special. For example, the solution algorithm does not
have to test for a delta/wye connection and do something special as some FBS algorithms do.
The system Y matrix handles it. It also allows for models and analyses that are atypical for
power flow. For example, the intrinsic load models in OpenDSS switch to linear models when
the voltage estimate is outside the normal band (Figure 2-19). The normal band defaults to +/-
5%, a typical design target for many distribution utilities. The band can be expanded at the user’s
discretion.

10253878 2-29
 DSS P,Q Load Characteristic

1.3

1.2
Const Z
1.1 105%*
1

0.9
|I| = |S/V| 95%*
0.8

0.7
PU Voltage

0.6 Const Z
0.5

0.4

0.3

0.2

0.1

0 (Defaults*)
0 0.2 0.4 0.6 0.8 1 1.2
PU Current

Figure 2-19
The standard OpenDSS load models switch to a linear model when voltage is outside normal
limits.

For the constant power load model shown the current magnitude follows a hyperbola with
respect to voltage magnitude when the voltage is in the normal range. The current must increase
as the voltage sags to maintain a constant power. This causes the voltage to sag even more and
will eventually cause convergence failure for many power flow cases when the system is heavily
loaded. The convergence failure typically occurs when the voltage drops to approximately 70%.
By switching to a linear model at the minimum voltage the OpenDSS algorithm is able to
achieve convergence for significantly lower voltages than many other power flow algorithms.
The transition to a linear constant impedance load model occurs in two stages as shown. At 50%
voltage, the model becomes exactly equivalent to a constant impedance model, which nearly
always converges for any voltage. The transition from the 50% point to the lower voltage of the
normal band is interpolated linearly. This provides a smoother transition less subject to iteration
jitter than switching completely to a constant impedance model at the bottom of the normal band.
This load model was first instituted to prevent convergence failure in the middle of long, multi-
year planning studies that might take hours to run. A typical example was a 20-year study in
which the load growth being modeled caused the system to become overloaded resulting in the

10253878 2-30
voltage sagging too low to achieve convergence in, say, year 13. The planner is not generally
interested knowing the exact voltage when this happens, being mostly interested in knowing that
the voltage is out of the normal band and capacity investments would have to be made before
year 13. The solution crash became a nuisance for the planning study and was fixed by the
transition to a linear model.
This modified load model also allows the program user to apply a short circuit fault without
leaving the power flow solution mode. The load models in the affected voltage sag area will
automatically transition to linear constant impedance models in varying degrees. This generally
allows the power flow solution to converge easily.
A practical application of using this feature for Modern Grid planning is the simulation of Low-
Voltage Ride-Through (LVRT) in DER inverters. The purpose of LVRT is to prevent inverters
on feeders not directly involved in a fault from unnecessarily disconnecting and creating a supply
deficit.
The transition to a linear load model can be accomplished by adjusting the compensation current
without changing the Y matrix. A few more solution iterations may be required, but generally
this requires less computational effort than refactoring the Y matrix. This is a key advantage of
this method besides allowing for a multitude of different load models. A general guideline for
implementing a load model in this method is that the model be physically realizable. However,
this is not an absolute requirement as several models have been developed that do not represent
any real circuit element but still converge in this model.
This solution method will not solve all types of power flow problems but can solve nearly all
power flow problems on distribution systems – radial or meshed networks. It works best for
systems with a dominant bulk power source like most distribution systems. It requires that the
first guess at the voltage be “close” to the final solution. This is not a problem for solutions in the
middle of daily or yearly sequential-time solutions; the present solution is a very good guess for
the next one. The initial solution is the most difficult. The solution initialization routine in
OpenDSS has evolved over the years so that it accomplishes this with ease in most practical
cases and works well for arbitrary unbalances in loads and networks.

10253878 2-31
10253878
3
DISTRIBUTION PLANNING FOR DER
It is becoming more common that utility distribution planners are faced with accommodating
widespread Distributed Energy Resources (DER) on their power distribution circuits. For
example, in many states, the renewable portfolio standards and incentives from various sources
have resulted in larger solar PV installations than previously experienced.

Planning Process
The distribution planning process has traditionally been focused on determining the least cost
alternative for meeting the peak load demand projected for some date in the future. The analysis
is often simplified because loading patterns have been the same for many years and there is much
experience with dealing with these loading patterns such as for the thermal rating of
transformers. By looking at how the system behaves at one loading point, the planning engineer
has a good idea of how it will respond at other times. The basic process can be summarized as
follows:
• Define a Distribution Planning Area (DPA) and model it for power flow analysis.
• Develop a load forecast for a selected planning horizon.
• Determine when planning limits on voltages and current capacities will be violated based on
the load forecast.
• Identify one or more alternatives for correcting the violations.
• Determine the least cost alternative over the planning horizon using approved economic
evaluation methods.
There are many things in this process that are somewhat ambiguously defined and open to
engineering judgment. Sometimes the DPA consists of just one feeder, but in other cases it could
be a well-defined geographic region bounded by several substations in which switching between
feeders may be accomplished.
Engineering limits on allowable voltages and current-carrying capacities also require the
application of engineering judgment. For example, current-carrying limits are generally based
on thermal limits of the current-carrying device. Limits are established based on the thermal
heating assuming a typical peak daily load shape like that shown in Figure 3-1. Utility
distribution planners have a historical base of decades of experience dealing with load shapes
like this. Perhaps, the biggest impact of adding DER to the planning mix is that the load shape is
changed and new planning criteria or more capable planning tools are required to deal with it.
Planning limits are established for the assumed load characteristic. Some utilities will allow for
considerable margin to allow more flexibility for reconfiguring circuits after a failure while
others will allow loading nearly up to the maximum limit. There are different limits for overhead
lines, underground cables, and transformers based on their capabilities to get rid of the heat from
losses incurred carrying the current.

10253878 3-1
Figure 3-1
Typical daily load profiles -- 3-day period

Voltage limits in North America are basically dictated by ANSI Standard C84.1. Most utilities
plan for voltages in the primary (MV) distribution system to be within ±5% of nominal rated
voltage. Voltage drop has traditionally been the main focus of concern, with minor attention to
high voltages from voltage rise. That is, the voltage is generally assumed to drop as one travels
from the substation to the feeder extremities and the feeder must be designed to supply the load
demand with a voltage higher than a specified minimum value.
The minimum voltage planning value varies slightly from utility-to-utility. Some utility planners
design for a minimum voltage of 97.5% (117 V on a 120 V base) to allow for inaccuracies in the
models and unexpected load growth that might cause the voltage to drop lower than expected.
Others will allow the voltage predicted by power flow analysis to drop to 95% at peak load. This
allows some room for additional voltage drop on the secondary (LV) system where the voltage is
permitted by ANSI Std C84.1 to drop to -8%.
Figure 3-2 shows an example 3-phase voltage profile for the IEEE 8500-Node Test Feeder.
There is no DER on this circuit and the load in this figure is at 100% of the projected peak load.
The profile is typical of many heavily-loaded feeders. The voltage starts off at 105% at the LTC-
regulated substation bus and generally declines until another voltage regulator is encountered.
Note that one phase actually sags below 95% at this loading level. Figure 3-3 shows the same
voltage profile for loading at 40% of peak demand, which could be a typical value for the
minimum load. A typical feeder would have a voltage profile closer to the minimum load profile
for most of the year; the peak load usually occurs only a few days per year.

10253878 3-2
Figure 3-2
Voltage Profile at Peak Load

This presents a challenge for the planner: how to design the voltage regulation scheme to
maintain the voltage within ANSI C84.1 requirements all year. The approach shown here is one
that many utility planners take for simplicity of operation. The voltage regulating apparatus is set
to regulate close to 105% at minimum load. Sufficient voltage regulators and shunt capacitors
are added to keep the minimum voltage on the feeder above 95%.
DER are often located in parts of the distribution system which are lightly loaded and voltage is
more difficult to regulate. Thus, voltage regulation is one of the more critical issues when
planning for DER on utility power distribution systems. DER affects the planning process in
ways that alter the long-established planning rules, requiring the planning process to be
amended. Not only can it alter the load shape and, thereby, the thermal capacity ratings, but it
affects the way voltage regulation must be designed.

10253878 3-3
Figure 3-3
Voltage Profile at 40% of Peak Load

Firstly, having a power producer on the feeder rather than a load tends to cause a voltage rise
rather than a voltage drop. Note the “Headroom” labels on the two voltage profile figures. This is
the margin that the planner has left for devices that raise the voltage. In this illustration, there is
little headroom for DER on this feeder – perhaps only about 1%. The headroom also varies by
location on the feeder. To allow for more headroom and increase the DER hosting capacity of a
feeder, the planner can no longer exploit the entire ±5% voltage band strictly for voltage drop.
This is particularly true for off-peak periods where the voltage profile for many distribution
feeders tends to be at a higher level. One implication is that there will necessarily be more
investment in voltage regulation devices and/or shorter feeders. Shorter feeders would imply
decreasing the distance between substations and, therefore, more substations.
Secondly, DER will change the shape of the daily load curve. DER is generally thought to reduce
the peak power demand seen by the utility supply and thus assist in serving the load. However,
that depends on the ability of the DER to supply power at a time coincident with the peak load.
Variable sources such as wind generation and solar PV generation cannot reliably achieve this at
all times. Wind sources tend to be strong in a diurnal pattern in the morning and evening while
the load peaks in midday. Solar DER generates in midday, but is absent by the time of the
evening peak which is the limiting loading condition on many residential feeders. Planners must
decide how they will take this variability into account.

10253878 3-4
DER – and any other modification to the distribution system that alters the effective load shape –
requires some additional steps in the planning process. For DER, these might include:
1. Screening for operating conflicts that might require additional planning studies. The main
operating conflicts are related to:
a. Voltage regulation
b. Conflicts with the utility fault-clearing process
2. Determining the incremental power delivery capacity, if any, that the option achieves.
Table 3-1
Summary of Typical Planning Limits for Distribution-Connected DER

Criteria Basis Limit

Voltage Change (Infrequent)
Voltage Change (Frequent)
Voltage Regulation
(Voltage Rise)
Generation Ratio
Feeder Design Capacity
System Stiffness
Short Circuit Current
Contribution
Ground Source Contribution
Infrequent (disconnection due to fault
< 5%
or inadvertent trip)
Fluctuating generation such as solar
< 1%
PV or wind generation
Normal Service Voltage ±5%
(ANSI C84.1) of nominal
DER kVA in % of < 20% for rotating machine
min Load kVA < 33% for inverters
DER kVA in % of
<15%
Feeder Design kVA
DER current in % of short circuit < 2% (wind and PV)
current at point of connection <4% (continuous generation)
< 5% (rotating machines)
% generator contribution to faults
------ (inverters)
% of ground fault current contributed
by DER and interconnection < 5%
transformer

* DER installations that exceed these limits warrant further study to determine feasibility and may require remedial
action.

Planning Limits for DER

Several planning limits have been identified from both North American and European practices.
Table 3-1 shows a summary of several key limits that have been applied for quantifying DER
hosting capacity on distribution systems.
The limits in this table are applied as guidelines for screening proposed DER installations. When
a proposed DER installation results in these limits being exceeded, it is possible that significant
changes will have to be made to the design or operation of the distribution system to
accommodate the DER. Further study should be performed, often with more sophisticated
planning tools than a simple power flow program. Remedial measures may be prescribed to
allow the hosting of the proposed DER.

10253878 3-5
The criteria that have been identified typically fall into one of the following general classes:
• Voltage regulation (e.g, voltage rise for the case of DER)
• Rapid voltage change (fluctuations, sudden loss of generation)
• Thermal limits (capacity, losses)
• Protection limits (overcurrent, islanding)
The first two categories typically dominate planning limit screens. The latter two are related to
the limits in Table 3-1 which are stated for Generation Ratio and Feeder Design Capacity. The
short circuit limits listed in the table fall into the protection limits category.
Each of limits listed in the above table may be evaluated in, at least, a simple, straightforward
manner with tools most distribution planning engineers have at their disposal. However, neither
passing nor failing one of these limits is guaranteed to correctly represent the proposed DER
configuration. The voltage change analysis, in particular, can be greatly enhanced by a tool
capable of quasi-static time series (QSTS) simulations in 1-second intervals. This allows the
engineer to observe the tap changer and capacitor switching response to DER output that might
be changing as much as 10% per second. More sophisticated tools would be required for the next
level of screening that might include islanding analysis and open-phase fault analysis.

Feeder Hosting Capacity

The DER planning limit is often referred to as the “hosting capacity” of the circuit. The hosting
capacity varies depending on what phenomena are being examined and the type of DER being
considered. For example, short-circuit contributions are very important for any type of rotating
machine generation, but less important for inverter-based solar PV generation. On the other
hand, voltage issues dominate planning limits for solar PV generation as well as being quite
significant for other forms of DER. European literature tends to focus on voltage rise with a
limiting factor being the “headroom” above the operating voltage with respect to the upper limit.
Service voltage limits appear to be more liberal in Europe (±10%) than in North American
systems that follow ANSI C84.1 (±5%). Thus, one might expect that North American systems
will be limited – perhaps, more limited – by headroom issues than European systems. However,
European-style systems around the world have very long LV feeders and suffer significant
problems with overvoltages from voltage rise due to solar PV generation.
In 2010, EPRI began a detailed investigation of the impact that increasing levels of solar PV
could have on distribution system performance. Employing a rigorous, stochastic-based approach
similar to that described in the last section of this chapter, EPRI analyzed more than six million
unique solar deployment scenarios across 35 distribution feeders throughout the U.S. Through
this exhaustive analysis technique that considers a wide range of voltage, thermal, and protection
issues, EPRI has been able to quantify the range of impacts that could be encountered with
increasing levels of PV deployment (see references [1] to [4]).
Based on the lessons learned from this rigorous analysis, a newer streamlined hosting capacity
method has been developed and validated against the detailed approach [5] [6] [7]. The
streamlined method efficiently determines hosting capacity and can be applied with existing
distribution planning software to analyze hundreds of distribution feeders throughout a service
territory. Planners are able to better quantify the impacts of DER across their distribution

10253878 3-6
systems. DER impacts derived from the distribution system analysis can then be used in system-
wide cost/benefit analyses (CBA).
Many of the impacts identified in the distribution analyses are expressed as costs or costs saved.
Costs are incurred to mitigate adverse impacts and costs saved are avoided costs that would have
otherwise been incurred. These are aggregated categorically, distinguishing between benefits and
costs. The cost-benefit step monetizes the accumulated costs and benefits to produce summary
cost-benefit metrics.
A planning approach for the electric power distribution grid that is integrated overall will fully
realize the value of DER while cost-effectively serving all customers at established standards of
quality and reliability. Such a holistic approach to distribution planning as depicted in Figure 3-4
is one of the core components of the cost-benefit framework for an integrated grid [8].
Voltage

voltage unacceptable
overvoltage
limit

time 
Thermal Capacity Protection
relay
desensitization
Load Only

Current
Watts

Integrated
Load and PV
Approach for
Cost/Benefit Impedance 
Impedance 
Analysis

unserved energy
Energy Losses

Energy

energy exceeding
normal
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Month Time 

Energy Reliability

Figure 3-4
Distribution Analysis Components for Integrated Grid Assessment

The amount of DER that can be accommodated is based on the feeder response to DER and
when adverse impacts begin to exceed a planning threshold.
Each distribution feeder is designed to reliably serve all customers in a cost efficient manner,
thus each feeder is designed differently. As a result, each feeder has a unique response to DER,
as illustrated by hosting capacity in Figure 3-5. Hosting Capacity Determination Factors. Voltage
class, feeder topology, and load location are just some of the factors that impact what level of
DER can be accommodated and where it can be accommodated.

10253878 3-7
 DER Size and Feeder Design
Location and Operation

Impact Impact
Impact
Below Above
Depends
Threshold Threshold

Figure 3-5
Hosting Capacity Determination Factors

In order to effectively consider the impact of DER across an entire distribution system, a method
must be utilized that has a number of key components, including the following:
• Granular: capture unique feeder-specific responses
• Scalable: applied across entire service territory
• Efficient: easily repeated as distribution changes to consider future changes and
reconfiguration (operational flexibility)
• Transparent: clear and open methods
• Proven: validated techniques
• Available: utilize readily available utility data and models
EPRI has developed a new, streamlined hosting capacity method derived from the detailed
analysis of over 6 million unique DER deployments across a wide range of distribution feeders.
A brief summary of the method is illustrated in Figure 3-6.
Input
The basic input for the methodology is the typical power flow model used by distribution
planners. In most cases, only the MV system is modeled. Loads are modeled and allocated based
upon existing planning practices for the utility. Depending upon the DER technology being
evaluated, various load levels may be considered (e.g., daytime minimum load for solar PV).
Methodology
The overall methodology applies selected distributions of DER deployments, considering highly-
distributed rooftop PV systems as well as single and distributed instances of other forms of DER.
Taking advantage of only the standard, commonly-available power flow and short-circuit
analysis functions available in distribution planning tools, calculations are performed on each
feeder for each discrete distribution of DER to determine the overall feeder hosting capacity. The
overall method considers various aspects of grid impacts that constitute voltage, thermal, and

10253878 3-8
protection-related evaluations as identified in [3]. The amount of DER is increased for each
assumed distribution until a planning limit is breached.
For each location within a feeder, the hosting capacity is calculated for each of the evaluation
criteria the planner chooses to consider. One particular location on a feeder may have a higher
hosting capacity for voltage than for thermal issues, whereas another location may be the
opposite. From those issues, mitigation options can be evaluated.
Output
The output of the methodology is the distribution system hosting capacity value for each node.
Each node is evaluated to determine the range of hosting capacity values that the specific feeder
can accommodate. Each evaluation criteria is then evaluated and provided separately, thus
giving the planner the option to include/exclude certain criteria or consider combinations of
criteria thereof. The output is often best understood by coloring the circuit diagram to indicate
the hosting capacity level. The feeder-level view also indicates locations on the feeder that may
be more optimal for DER interconnection. This can be helpful in directing DER to ‘better’
locations within a feeder.

Input

Method

Location √
no impact Output

case by case
look needed

Location X
Potential risk

Figure 3-6
Streamlined Hosting Capacity Methodology

10253878 3-9
Based on the analysis of a large system, Figure 3-7 illustrates the distribution of hosting capacity
values determined through the analysis for non-optimally-sited centralized DER. Voltage,
thermal, and protection-based hosting capacity values are shown separately. As shown, 50% of
the distribution feeders can accommodate up to approximately 2.5 MW regardless of where the
DER is located. For feeders that have minimum hosting capacity values below 2.5 MW, the
limiting factor is predominately voltage related. Above 2.5 MW, the primary limiting factor is
found to be thermal related.

Voltage Thermal Protection

100%

75%
Frequency

50%

25%

0%
0 1 2 3 4 5 6 7 8 9 10
Hosting Capacity (MW)

Figure 3-7
Cumulative Distribution of Non-optimally Sited Centralized DER Hosting Capacity for 185 12-kV
Distribution Feeders

Feeder hosting capacity is difficult to link to any one specific feeder characteristic. There are
dominant characteristics that influence hosting capacity such as voltage class, but as shown in
Figure 3-8 for 12-kV class feeders, there is a wide variation in hosting capacity. The variation is
due to the influence from all feeder characteristics. Minimum (lower) hosting capacities correlate
to less optimal DER locations while maximum (higher) hosting capacities correlate to more
optimal DER locations. The maximum hosting capacity values are much more distributed across
the analyzed spectrum compared to minimum hosting capacity.
60 1
Cumulative Distribution
Feeder Count

0.75
40
0.5

20
0.25

0 0
0.5

1.5

2.5

3.5

4.5

5.5

6.5

7.5

8.5

9.5
10
More
0

Minimum Hosting Capacity

Maximum Hosting Capacity

Figure 3-8
Maximum and Minimum Centralized DER Hosting Capacity for 185 12kV Distribution Feeders

This streamlined method has been demonstrated in an actual distribution system and used as the
basis for determining the impact and value of DER across the distribution service territory. The
process will ultimately reduce the engineering time required from traditional long-term planning
studies and aid in the evaluation of interconnection requests. Initial results indicate a single
feeder can be analyzed completely in under five minutes using a typical desktop computer.

10253878 3-10
Mitigation Analysis
A desired penetration level creates a planning target. The hosting capacity on some feeders will
exceed that target, while others may need upgrades/mitigation. These feeders can be identified
and mitigation options determined for the economic analysis.
Mitigation would be applied in the least cost order until the target penetration level can be
attained without violations. A hypothetical application of this may employ smart inverters until
the benefit is exhausted. Then mitigation such as reconductoring might be necessary. If the
necessary benefit from reconductoring cannot be attained because the original conductor is lower
impedance or the feeder length is short, more costly solutions would be evaluated such as voltage
class upgrades or creating new parallel feeders.
Energy Analysis
The extent to which DER can reduce losses also depends upon the coincidence of DER
generation output compared to load. Storage devices can also be a load on the system when
charging. A detailed loss analysis takes these two factors into account by performing a full 8760
simulation (annual analysis at hourly resolution) that captures the annual energy and peak
demand changes under varying DER penetration levels and deployment scenarios. This requires
a planning tool with sequential-time power flow capabilities.
Thermal Capacity Analysis
This analysis quantifies the influence of varying DER on distribution system capacity with
regard to asset thermal ratings and the potential deferral of capacity upgrades. The assessment
methodology develops daily and yearly asset loading profiles representing projected load
conditions combined with varying DER portfolios. The profiles are then used to quantify the
impacts on asset thermal ratings while accounting for different combinations of asset thermal
characteristics, load mix, and projected load growth.
Volt-var Control
Voltage regulation issues are the limiting factors for the hosting capacity of many distribution
feeders. If all the DER comes on line abruptly, it can push the service voltage over the limit until
voltage regulators bring it back down. Then when the DER generation exits the system, the
regulators may on a tap that is too low to support the voltage. This problem cries out for more
sophisticated inverter controls. The volt-var control [EPRI 1023059] and advocated in recent
grid codes is a potential solution to this problem. There remain several details to work out to
coordinate this with other active voltage regulation on many distribution systems.
Models of inverter controls have been implemented in the InvControl model in EPRI’s OpenDSS
program to support research into this problem [EPRI 1024353]. In addition to volt-var control,
the InvControl model implements a volt-watt control mode and a dynamic reactive power control
mode. Systems with long LV feeders generally have a low X/R ratio such that R is much greater
than X. Such feeders will often respond better to volt-watt control than volt-var control
Many who work in this field are finding that communications between the distribution system
and the generation are essential for a given distribution system to achieve its full DER hosting
capacity.

10253878 3-11
The rapid detection of inadvertent islands and abnormal conditions such as open-phase faults
remain important issues. Protection issues cannot always be resolved with autonomous relaying.
Communications, often implemented in the form of direct transfer tripping, will be required to
achieve successful operation of some DER installations without interfering with utility fault
clearing processes.
Capacity Provided by DER
Whether a given DER scenario provides beneficial capacity to the system that planners are able
to exploit depends on many factors. The key question to ask is not “What is the rating of the
DER?” but “How much more load can be served with this DER in this location at the same risk
of unserved energy?” Sometimes the incremental capacity measured by this standard is greater
than the power output rating of the DER. Sometimes it is zero.
The “15% Hosting Limit”?
A common way to express limits for DER in distribution systems is to limit the amount by a
percentage of an easily-measured quantity such as peak kVA load demand. This is usually a
conservative approach that requires little feeder analysis effort, but simply an inspection of
measured data. For example, NERC recommends, and a number of utilities have adopted a 15%
limit. This number is likely sufficient to ensure that this amount of DER can be hosted without
requiring modifications to the distribution system equipment or operating practices.
The 15% value shows up in more than one guideline but can have different applications:
1. 15% of minimum load. This could be a good approach for an urban LV network in which the
generation can never exceed the load. However, it is generally overly restrictive for radial
feeders.
2. 15% of peak load. This is typically three times more generous that the previous guideline.
3. 15% of feeder capacity, or all feeders supplied from the same bus. This is still more generous
than the previous two. It is also a more firm guideline than the other two because it doesn’t
suffer from a moving basis. Some analysis to determine actual capacity of a feeder may be
required.
None of these adequately account for differing feeder topologies. A lengthy 10 MVA feeder may
only support 1500 kVA of DER. However, a short system or a system with DER located closer
to the substation may support considerably more DER. For many types of DER, voltage limits
make more sense although more analysis is required to evaluate. The EPRI Streamlined Hosting
Capacity methodology described above is designed to provide more useful values for this
analysis.
Usually, when the statement is made that the feeder can “host” a certain amount of DER, it is
meant that it can do that without costly changes to the feeder or the operation of the feeder. It
may be possible to host much larger amounts of DER if modifications can be made to the
distribution system. That will require more detailed engineering studies than the relatively simple
screens described here.

10253878 3-12
Screening Proposed DER Installations for Operational Conflicts
An important first step in evaluating distribution connection requests for DER is to screen the
proposal to determine whether detailed study is required. The EPRI DGScreener program
performs the following five screens for this purpose:
1. Voltage change upon entering or exiting the system. This is often the best screen for
determining if the DER is too large to be hosted without making significant changes.
2. Voltage rise/fluctuations. This is an important screen for renewable generation such as solar
photovoltaics. Interaction with voltage regulation devices should be considered.
3. Short circuit current contribution. A large current magnitude change may indicate that fault
interrupters will see excessive duty. A large percentage change may indicate that the
overcurrent protection coordination will be disrupted
4. Open conductor screen. This is dynamic simulation that is a good indicator of the need for an
effective grounding solution to prevent overvoltage. (See Figure 3-12.)
5. Islanding voltage and frequency change. A dynamics simulation to identify whether or not an
unintentional islanding condition can be detected promptly.

Reactive Power Control

As the penetration level of DER on distribution systems has increased, reactive power control
has become more critical. Previous guidelines for DER interconnection have often stated that the
DER should operate at unity power factor (PF). This is generally an inert operating point that is
not likely to support unintentional islands because the reactive power demand will not be met
and the voltage will quickly collapse. Voltage rise issues from high-penetration solar PV
generation coupled with the desire to ride through voltage sags have placed a spotlight on this
issue. PF requirements are currently under scrutiny by the industry and advanced inverter
controls are being developed to address many of the concerns.
One simple approach to minimizing the voltage rise/fluctuation issue is to require the DER to
absorb reactive power from the distribution system at a PF of approximately 0.95 to 0.98. This
small amount of reactive power is often sufficient to keep the voltage change to a minimum.
Some of this effect is achieved naturally when an interconnection transformer is used. Figure 3-9
illustrates how a generator operating at unity PF at its terminal actually draws reactive power, Q,
from the utility system. The current flowing through the interconnect transformer results in some
var “losses” in the transformer leakage inductance. Even if the active power, P, is flowing out, Q
is flowing toward the generator. A correct model of the DER would take this effect into account.
The reactive power flowing into the generator tends to oppose the voltage rise due to the active
power and minimizes fluctuations.

10253878 3-13
 Unity PF
Q = I2X
P - jQ P + j0

GENERATOR

INTERCONNECT
TRANSFORMER

Figure 3-9
Impact of Interconnection Transformer on Apparent Power Factor

Helping to minimize the voltage fluctuations is just one good reason to require an interconnect
transformer. While one may purchase rotating machines rated for primary distribution voltage
(medium voltage, MV), it is frequently not a good idea to directly connect a generator to the MV
system. One key reason is a mismatch in the basic impulse insulation level (BIL). For example,
many utility 15 kV systems are designed for 95 kV BIL while the machine may be only 50 kV
BIL. This makes the machine the weakest link in the system and is subject to failure for lightning
surges. It is difficult to design arrester protection that will protect the machine and survive
overvoltages that occur from events on the utility system such as ferroresonance and open-
conductor faults.
Advanced PV Inverter Control
However, the transformer reactive power losses may not be large enough to mitigate the voltage
fluctuations and the generator itself must be capable of operating at a lower power factor
(absorbing reactive power). Solar PV generation has created much interest in this in recent years
because cloud transients can cause large voltage fluctuations (see Figure 3-13).
The PV inverter industry has begun to address this problem with grid codes that define volt-var
characteristics that can be implemented for autonomous operation. An example of a proposed
volt-var characteristic is shown in Figure 3-10. This characteristic uses basic dead-band
controller logic common to many devices on the utility distribution system. As the voltage varies
within the band around the target voltage the unit operates at unity PF. As the voltage tends to
rise outside the band, the inverter would begin to draw reactive power to minimize the voltage
rise. Likewise, as the voltage sags below the band, the inverter would attempt to produce more
reactive power – within its capabilities – to help support the voltage.
Obviously, there will have to be coordination with utility voltage regulating goals. Such things as
adding a new capacitor bank could disrupt the settings on this controller scheme. Also, a desire
to operate with reduced feeder voltage in a voltage optimization scheme could conflict with the
desire to have autonomous inverter controls. Accommodating two-way power flow on
distribution feeders can have some unexpected complications.

10253878 3-14
 Qmax Dead Band

-Qmax
Target Voltage

Voltage
Vmin Vmax

Figure 3-10.
Volt-var Control Characteristic Proposed for Inverters

As mentioned above, for solar PV generation on customer premises the utility source impedance
can appear mostly resistive. Reactive power control alone is less effective in controlling the
voltage with low X/R values. Volt-watt control schemes are being implemented as well as volt-
var schemes.

Capabilities of Planning Tools for DER Planning

The streamlined hosting capacity method was designed to be implemented with standard
distribution power flow and short-circuit analysis capabilities. For more detailed analyses,
enhanced distribution planning tools are required to include DER in the planning process. Some
of these advanced capabilities are described here.
Sequential-time Power Flow
This is the single most important capability of a planning tool for including DER planning other
than an accurate power flow solution capability.
The basic process described in the introduction to this chapter and in the streamlined hosting
capacity analysis can generally be accomplished with a power flow tool that can execute a single
snapshot power flow at a selected loading for a given year on the planning horizon. This is
adequate if the loading characteristic falls into the usual predictable patterns. It can also be
adequate for analyzing some DER applications where the DER is dispatchable, and is reliable,
leading to a predictable system loading characteristic.
The value of the impact of DER is dependent on location and time. The single snapshot power
flow solution can be effective in estimating the locational value by, for example, comparing
losses between two locations for a given DER size. This is what is often done in many technical
papers on “optimizing” DER on a distribution feeder. However, the impact of DER is not
constant with time, particularly if the DER is solar or wind generation. DER generation typically
has value only if its output is available at a time when it is useful to increasing the capacity of the
system. An effective distribution system analysis tool for planning with DER will have some
means of capturing the time-dependent value.

10253878 3-15
 70

60

50

Load, MW 40

30

20

10

Figure 3-11
Typical Distribution Feeder Annual Load Profile

This is illustrated in Figure 3-11.and Figure 3-12. Figure 3-11 shows a typical 8760-hour annual
load profile for a summer-peaking North American substation. Note that the weekly loading
cycles are clearly visible although the daily variation is obscured by the resolution of the
graphics. The load peaks for only a few days during approximately 5 weeks in the summer
months. That is when a failure of a major power delivery component is most likely to yield a
capacity problem that a simple load transfer switching operation cannot resolve.
Figure 3-12 shows another way of plotting an annual load profile. This plot was computed by
simulating a distribution system over a year, performing 8760 power flows. At each hour the
power exceeding the normal planning limit was accumulated, yielding the energy served above
the planning limits – the energy at risk of being unserved in case of failure of a key system
component. In this kind of evaluation DER would have real capacity value only if it is available
at the periods of high stress.

10253878 3-16
Figure 3-12
Energy Exceeding Normal Planning Limits

Besides annual simulations to capture the time value of DER power output for capacity, other
reasons to perform time-sequential power flow solutions include:
• Obtaining the proper sequencing of voltage control devices. At a minimum, a daily profile
can be executed to get all the voltage regulator tap changer controls and capacitor switch
controls sequenced properly. Of course, the control models must properly represent the time
delays.
• Evaluating distribution losses. The single snapshot power flow will often give a misleading
picture of the true annual losses. Planners have formulae that convert peak load losses to
annual losses that work if the load shape is consistent with historical data. However, DER
changes the load shape.
• Evaluating the voltage fluctuations that might be caused by large solar PV installations
during cloud transients. These simulations are typically performed at 1- to 5-s time intervals
(Figure 3-13).

10253878 3-17
Figure 3-13
Voltage Fluctuations Due to Cloud Transients

Sequential-time solution capability is a key element of a distribution system analysis tool that
can evaluate DER. For situations where DER significantly modifies the load characteristic, it is
not possible to guarantee a correct solution without simulating of a significant amount of time.
The EPRI open-source OpenDSS program was developed for this purpose and is available on the
Internet. This program was first written in 1997 to support all the aspects of the distribution
planning problem with DER that were known at the time. There have been many capabilities
added since then due to the emergence of other distributed technologies including storage. The
program was made open source in 2008 to speed the adoption of more advanced distribution
planning techniques in the power industry.
Dynamics Simulation Capability
A capability somewhat related to the sequential-time power flow analysis is the dynamics, or
electromechanical transients capability. This capability is not yet commonly used in distribution
planning; power flow applications dominate the practice area. However, it is becoming more
important as planners become more interested in understanding how such things as inverter
controls in solar PV and storage device will interact with the system. Microgrid modeling is
another important area where dynamics analysis is useful. DER with inverters often have
controls with time constants of 2 ms requiring simulation time steps of approximately 200 µs.
While this is a challenging modeling area, advances are being made that will bring this capability
to the distribution planner in a palatable form. There are more details on dynamics analysis in
distribution planning provided in Chapter 9.
An interesting dynamics problem related to DER is the open-conductor fault on a circuit feeding
a synchronous generator. Figure 3-14 shows results of a dynamics simulation made with the
EPRI DGScreener software. This condition can occur from a splice failure, conductor burn
down, or a blown fuse. The open phase is back-fed by the DER with varying voltages that

10253878 3-18
depend on loading and transformer connection. The oscillations are power swings in the DER
model. In some cases, the voltage can become high like that shown in the figure and cause
damage to customer equipment.

OPEN
PHASE
INTERCONNECTION
DG
TRANSFORMER

UTILITY
SOURCE

FEEDER
CAP. BANK

Figure 3-14
Open-Conductor Simulation Result, Delta-Y Transformer

Detailed modeling of Islanding phenomena would also require dynamics simulation capability.
Unbalanced, Multi-phase Line Models
Most distribution system analysis tools designed for the North American market can support
some form of a full 3-phase feeder model including 1-phase and 2-phase laterals. The better tools
for DER analysis would support unbalanced matrix models of lines and cables. Lines models
based on the familiar symmetrical component impedance values sometimes underestimate the
voltage unbalance that results from large 3-phase DER.
Some large DER facilities require two feeders to carry the power produced. If these feeders are
on the same power poles, the proper model is a 6-phase conductor model. This could be
important depending on the type of neutral and the interconnection transformer configuration.

10253878 3-19
Transformer Modeling
In the North American distribution system, DER can be interconnected with many different types
of transformer connections including:
• Delta/Wye
• Grounded-Wye/Grounded-Wye
• Grounded-Wye/Delta (i.e., grounding bank; with and without neutral reactor)
• Delta/Delta
• Ungrounded-Wye/Delta
• Open-Y/Open-Delta
• 1-phase, Line-to-Neutral
• 1-phase, Line-to-Line
These present a modeling challenge to the software developer, but should be in the repertoire of
a DER planning tool.
Sometimes it is necessary to model in detail the 1-phase, 120/240 V residential service
transformer commonly used in North America (Figure 3-15). This may seem an easy task, but
not all analysis tools can handle it because it is actually a 3-winding transformer. Most rooftop
solar PV generation is connected with 240-V inverters but 120-V inverters are possible. Also,
models of microgrids might require detailed modeling of this transformer for certain conditions.
So this is a nice capability to have in a DER planning tool.
1-Phase Residential Service Transformer
Center-tapped LV Winding
MV Phase
120V Loads

MV Neutral

LV Triplex

Figure 3-15
Residential Service Transformer, North America

Protective Relaying Simulation

Most utility distribution feeders are protected using series overcurrent protection. This is where
many of the potential conflicts between hosting DER and normal feeder operation arise.
Interconnected DER generally relies more on voltage relaying functions than overcurrent
functions. Since these cannot be coordinated with the conventional TCC curve approach, a nice
feature to have for DER planning is to simulate the actions of the relays and breakers by
applying a fault and observing the actions.
Also, many, if not most, utility customers have only overcurrent breakers in service entrance
panels. Utility engineers might be more concerned with DER contribution to the fault current for
exceeding breaker interrupting duty or otherwise interfering with the fault clearing process.
Customers might be more concerned with having enough short circuit capability to operate a

10253878 3-20
facility’s overcurrent breakers, particularly if there is an intention of operating the DER in a
microgrid for grid resiliency purposes. Inverter-based DER can typically supply only about
120% rated current to a fault before shutting down. This is often insufficient to operate
overcurrent protection. With DER there is more reliance on impedance and voltage relaying.
This simulation would be of the same nature as the dynamics simulation described above, being
conducted in very small time steps. Breakers, relays, and fuses can operate in a few milliseconds.
If machine dynamics enter into the analysis, it would be a dynamics simulation.

Initialize

Solve Electrical
Circuit
(Controls Static)
Push delayed
actions onto
Sample the Control Queue
Control Elements

Execute Control
Actions, If Any Pop delayed
actions off
Control Queue

No Control Yes
Actions Solution Done
Done?

Figure 3-16
OpenDSS Solution Loop with Control Modeling

Control Simulation
A key capability for supporting the sequential-time simulations is to have models of the various
switch and tap-changer controls. Such utility devices usually have dead-band type control logic
and time delays ranging from several seconds to a few minutes before acting.
This capability was implemented in the EPRI OpenDSS program by modeling controllers
separately from the power-carrying devices they control. Figure 3-16 shows the algorithm used
to solve the power flow and include delayed control actions such as tap changes and capacitor
switching.

Evaluating Impact of DER on Capacity

The classic question of how to evaluate the impact of DER on distribution system capacity can
be illustrated with the next two figures from a solar PV generation simulation. To contribute to
the load-serving capacity of the system, DER must be available and producing at the time it is
needed for capacity support. Some metric is required that represents this.

10253878 3-21
One simplistic way to quantify the incremental capacity provided by DER is illustrated in Figure
3-17. This method assumes the DER output is available and answers the question: How much
more load, ∆P LOAD , can be served by adding a given amount of generation, ∆P GEN , to the feeder?
In this example, the constraint is downline from the substation. There is no benefit to having the
DER in the substation because it doesn’t relieve any constraints on the distribution system. As
the DER is moved out onto the feeder, a seemingly strange thing occurs at some locations: the
amount of additional load that can be served is actually larger than the generator output. These
“sweet spots” are generally the result of relieving the most restrictive constraint and allowing the
load all across the feeder to grow. Values of 1.4 or greater are frequently found on radial
distribution feeders.

Figure 3-17
Amount of Additional Load Served Depends on Location

The value of ∆P LOAD / ∆P GEN can actually be much higher on meshed network systems where a
little voltage support at a sagging bus can do wonders in terms of capacity. Of course, it assumes
100% coincidence between load and DER output.
Distribution planners generally want to have sufficient wire-based power delivery capacity to
meet the peak load demand without violations of engineering limits. Some planners try to meet
this objective assuming one failure of a major component – the so-called “N-1” planning
contingency. For distribution systems, this often translates into meeting the demand in the
backup, or alternate feed, configuration. Figure 3-18 illustrates the difficulty of achieving this
objective with solar PV generation. On many feeders serving predominantly residential load the
peak demand occurs in the evening after the solar PV generation output has declined. Therefore,
the peak demand is unchanged from the case without PV generation. Thus, most distribution
planners are reluctant to credit solar PV generation with any benefit for capacity. Distribution
engineers often refer to the “holy grail” quest to find economic storage to extend the impact of
solar PV generation by two hours to cover the residential evening peak.

10253878 3-22
Figure 3-18
Example Impact of Solar PV Generation on Peak Substation Demand

Does this mean there can be no distribution system capacity enhancement from unpredictable
renewable generation such as solar PV? It all depends on what is constraining the capacity. As
pointed out in the introduction to this chapter, the current-carrying elements are constrained by
thermal limits that depend on the capability of the environment to dispel heat generated by
losses. Capacity limits are typically set based on historical experience with the normal expected
load shape. While solar PV generation may not reduce the peak current because its peak output is
not coincident with the peak load demand, it may significantly reduce the duration of the peak
loading period. This is illustrated in Figure 3-19 on the same simulation shown in the previous
figure.

Figure 3-19
Example Impact of Solar PV Generation on Peak Substation Demand

10253878 3-23
The width of the daily load shape is cut by nearly half on peak load days from approximately 4
hours to approximately 2 hours. If the constraint is the loading on the substation transformer, this
is likely very helpful. The thermal time constant on a power transformer might be 2 h, or more,
so a higher MVA of load can be served without overheating the transformer.
However, if the constraint is the rating of the getaway cable in the substation, which is quite
common, reducing the duration of the peak to 2 hours may not be much help. The cable
ampacities frequently reflect what the cable can withstand for 1 hour. The DER power output
would have to be more coincident with the load demand to provide benefit for that case.
There are a number of approaches a planner might choose to express the coincidence of load and
DER power output. A simple method would be to compute some sort of statistical coincidence
factor and derate the DER capacity accordingly. For solar PV generation, the coincidence factor
is typically around 30% for residential load and perhaps as high as 80% for downtown
commercial load the peaks about 2:00 – 3:00 PM. Using this philosophy, 1 MW of residential
rooftop solar is equivalent of approximately 300 kW of dispatchable generation. This will vary
with latitude and time of year. When the sun sets early in the winter in northern latitudes, there
may be virtually no capacity provided by solar generation without storage.
Some planners might say that the incremental capacity provided by DER matters only during
those times that it can affect the reliability of the system. One approach supported by the
OpenDSS program is based on the concept illustrated in Figure 3-20. The most frequently-
implemented application of this approach is to define a “Normal” rating for each element. The
basic idea of the Normal rating is that the load exceeding this value in the normal circuit
configuration would have to be disconnected if a major failure in the supply to that feeder
occurred. If the load is below this limit, a failure can occur without significant loss of service by
simple reconfiguration of the system. Each day is simulated with the load and DER output while
tabulating the energy served above the normal limit (Energy Exceeding Normal – EEN). If the
DER is effective in serving the load during the times when the load exceeds the Normal rating,
the EEN value will be reduced.

Energy Exceeding Normal

Year n+1

Maximum Rating

Normal Rating

Year n

Figure 3-20
Tabulating Energy Exceeding Normal (EEN) Rating

10253878 3-24
Figure 3-21 shows a classic EEN chart for a summer peaking distribution system. While a
proposed DER application may have other value to the system in terms of losses or offsetting
energy purchases, its impact on reliability will be when generation availability is coincident with
the times when the EEN values greater than zero.

Figure 3-21
EEN Plot for Summer Peaking System

EEN is a measure of the risk that unserved energy will occur due to overloads during outage
contingencies and can be used to compare planning alternatives. It can be used to help answer
the question of incremental capacity:
How much more load can be served by having DER present with the same
risk of unserved energy?
Figure 3-22 shows the results of an incremental capacity calculation for a 2 MW combined heat
and power (CHP) generator that is located near the “sweet spot” in a distribution system. The
generator is assumed to be running continuously at constant output. Thus, it is coincident with all
load although it produces a little too much power at light load conditions. The load in the DPA is
assumed to grow from 168 MW to over 200 MW during the time period the CHP unit is active.
Based solely on EEN comparisons, the DPA can serve about 4 MW more load at the same risk of
overload-based reliability degradation. This is more than double the power produced by the CHP
unit.

10253878 3-25
 Capacity Gain for
2 MW CHP

7000 14

6000 12
Base_Case
5000 2MW_CHP 10
Incr. Cap.

Incr. Cap., MW
MWh EEN
4000 8

3000 6

2000 4

1000 2

0 0
150 160 170 180 190 200 210
MW Load

Figure 3-22
Example Incremental Capacity Calculation for a 2 MW CHP Unit Using EEN

In contrast to the CHP unit with perfect coincidence with the load and located in a good spot on
the system to address overloads, Figure 3-23 shows the results for a similar analysis for 20 MW
of rooftop solar PV distributed over a different DPA. As the total load grows from 700 MW to
almost 1100 MW over several years, the incremental capacity hovers around 8 MW. This is 40%
of the installed capacity and is actually quite good coincidence for solar DER. This suggests that
for each MW of widely-distributed solar PV, the DPA can serve 400 kW more load at the same
reliability level. The main reason it cannot do better is that solar PV lacks meeting the evening
load demand peak by about 2 hours.
This general result would not apply to large concentrated solar generation, which would have to
be evaluated on a case-by-case basis.

Figure 3-23
Example Incremental Capacity Calculation for 20 MW Solar PV Using EEN

10253878 3-26
Stochastic Analysis
This section illustrates the detailed hosting capacity analysis that was used to develop the
streamlined hosting capacity analysis described earlier in this chapter. The process for
determining a particular distribution feeder’s hosting capacity for PV must consider the wide
range of potential spatial interconnections in addition to a variety of PV system sizes. The
likelihood of a simulated PV deployment resembling the actual future deployment increases with
the number of cases considered, therefore a large number of actual case “scenarios” must be
considered and a stochastic analysis technique was employed.
The Analysis
The stochastic analysis is the process that analyzes the impacts from the large number of
randomly-generated future PV deployments. This process determines a specific feeder’s hosting
capacity which depends upon:
• Criteria considered for the limit
• Where the PV is located
• How much PV is deployed
During this process, particular key deployments are flagged for exceeding specific circuit
monitoring criteria limits. In some feeders, the hosting capacity can be closely tied to total PV
penetration while in others, this capacity can be more closely tied to specific deployments. The
stochastic analysis conceptual deployment characteristics can be used to identify with actual PV
as it is deployed on the feeder.
Simulation
The stochastic analysis is an exhaustive process that compares the large number of PV
deployments to the feeder characteristics prior to adding PV. To alleviate some of the burden
associated with analyzing large numbers of scenarios, the solution process is scripted to decrease
overall simulation time. In the scripts the feeder control elements are locked in their present state
after the feeder is initially solved without PV. When PV is added to the feeder, control elements
are assumed temporarily fixed. This solution determines the maximum deviation in feeder
voltage and flow prior to control action. This is appropriate when considering the fast ramping
capability of solar with respect to typically slower delays on feeder controls.
Processing Results
The primary goal of the stochastic analysis is interpreting the simulated response such that the
feeder hosting capacity can be determined. This involves a detailed post-processing of the
simulated feeder voltage, current, and power measurements at each node from the substation to
all service endpoints. The post-processing gives the ability to aggregate and visualize the results
from a large number of simulations.
Figure 3-24 illustrates a specific feeder’s maximum voltage deviation for all primary and
secondary buses, or nodes, in each PV deployment simulation. Each primary and secondary point
represents a single PV deployment scenario. A total of 5,000 possible scenarios are considered
and shown here. As PV penetration increases, so do the deviations.

10253878 3-27
The trend is a linear increase for this particular feeder. However, the spatial location of PV does
contribute to the range of deviation values at each particular penetration level.

Figure 3-24
Maximum Primary and Secondary Voltage Deviations for Each PV Deployment Scenario

As shown in Figure 3-25, the deployments exceeding a 1% deviation threshold begin at 1 MW,
while all deployments exceed the threshold by 1.5 MW. The reason some deployments exceed
thresholds at lower limits is due to the spatial distributions of PV on the feeder and to system
sizes.
Figure 3-26 illustrates the weighted average short circuit impedance at the locations of the feeder
PVs for each deployment. Each marker indicates a deployment while the blue circles represent
cases where the threshold is not exceeded and the red crosses represent cases where the threshold
is exceeded. Deployments with high weighted impedance typically cause thresholds to be
exceeded at lower penetration levels.

Figure 3-25
Total Number of PV Deployment Scenarios Exceeding 1% Deviation Threshold

10253878 3-28
Figure 3-26
PV Deployment Characteristics that Cause > 1% Primary Voltage Deviation

The results shown here indicate only one piece of a multipart stochastic analysis. However, it
does shed some light on the type of advanced analysis capabilities that are necessary to consider
the wide range of potential spatial interconnections in addition to a variety of PV system sizes.
Similar approaches are also used for assessing overvoltages, loading, protection, and power
quality impacts.

10253878 3-29
10253878
4
RELIABILITY ANALYSIS FOR DISTRIBUTION
PLANNING
Distribution planning is sometimes characterized as selecting the least cost design that has the
capacity to serve the load with acceptable reliability. The regulatory agencies that oversee the
utilities will generally measure how well a utility has performed with respect to reliability by
using a prescribed set of reliability indices. In some locales with performance-based ratemaking,
there can be penalties if the reliability goals defined with these indices are not met.
The material in this chapter is updated and expanded from EPRI report 1012450 (2006) [47] on
an assessment of tools available to distribution planners for including reliability analysis in
planning. The report described reliability planning methods and needs in reliability analysis
tools. Various commercially-available software packages were assessed. The focus of the report
was on the standard reliability indices used by the regulatory agencies to measure the
performance of utilities.
It was pointed out at the time of the report in 2006 that many of the methods and definitions
implemented in the tools were more than 25 years old. Software tools have continued to evolve
as the industry has changed and there are more capabilities for evaluating reliability. Given the
impact of Distributed Energy Resources (DER) on the operation of distribution systems, tools are
now required to include DER in the planning process. The indices first proposed in 1981 are not
necessarily as useful when considering DER, although the basics of service reliability remain.
Other methodologies like those described in the previous chapter have been developed to better
represent DER impacts.

Reliability Indices
The reliability indices in use by U.S. utilities have been defined in an ad hoc fashion for many
years. The 1981 EPRI Report EL-2018 [48] serves as a fundamental reference for performing
reliability analysis and describes many of the reliability indices and methods for computing
them. Some distribution reliability analysis tool vendors point to this work as the basis for their
algorithms. However, the definitions of some of the indices were changed, and some new ones
introduced when the IEEE standardized the definitions in IEEE Std. 1366-1998. IEEE Std 1366-
2003 expanded on the definitions by introducing the “2.5 Beta” method for utilities to handle
Major Event Days (MED). This was updated and superseded in 2012 by IEEE Std 1366-2012.
[49] A MED is due to a hurricane, ice storm, wind storm, earthquakes, terrorist attacks, or other
severe natural and man-made phenomena and is not necessarily reflective of the way the
distribution system is designed and operated. A MED is excluded from the reliability indices
reported to regulatory agencies.
An excellent detailed treatment of the reliability indices and MED may be found in Chapter 10 of
Short’s Electric Power Distribution Handbook. [50]

10253878 4-1
Sustained Interruption Indices
The standard reliability indices generally deal with sustained interruptions lasting more than a
specified time such as 1, 2 or 5 minutes. These are commonly referred to as a permanent fault
that requires human intervention to repair. There are also indices for momentary interruptions
that typically result when an auto-reclosing breaker or recloser interrupts the fault current and
successfully clears the fault. A successful reclose typically occurs within 5 s, or less.
The standard sustained interruption indices are defined as follows.
SAIFI - System Average Interruption Frequency Index
Total number of customer interruptions
SAIFI =
Total number of customers served
This index is computed by summing the number of customers interrupted for each event
and dividing (normalizing) by the total number of customers. The numerator, the
number of customer interruptions, is useful valuing options in recloser and automated
switch location optimization. See Chapter 6.
SAIDI - System Average Interruption Duration Index

SAIDI =
∑ Customer interruption durations
Total number of customers served
This index is computed by summing the restoration time for each event times the
number of customers interrupted for each event and dividing (normalizing) by the total
number of customers.
CAIDI - Customer Average Interruption Duration Index

CAIDI =
∑ Customer interruption durations
=
SAIDI
Total number of customer interruptions SAIFI
This index may be computed by first computing SAIDI and SAIFI and dividing as
shown.
CTAIDI - Customer Total Average Interruption Duration Index. This is the total average time
customers were without power for customers who actually experienced an interruption.

CTAIDI =
∑ Customer interruption durations
( Duration)
Total number of customer interruptions
This index is computed like SAIDI but normalizing by the total number of customers
who have experienced a sustained interruption. According to the standard, each
individual customer should be counted only once regardless of how many times
interrupted during the reporting period.
CAIFI - Customer Average Interruption Frequency Index. This is the average frequency of
interruption for those customers who actually experienced an interruption. The customer
is counted but once regardless of the number of times interrupted.

10253878 4-2
 Total number of customer interruption
CAIFI =
Total number of customers interrupted
This index is computed like one would be computing SAIFI but dividing by the total
number of customers who have experienced a sustained interruption. According to the
standard, each individual customer should be counted only once regardless of how many
times interrupted during the reporting period.
Other indices are, briefly (refer to the standard for details):
ASAI – Average Service Availability Index. This is the traditional index for indicating the
percentage of the time a given system is available each year or some other defined
reporting period. When someone refers to “three 9’s of reliability” it is often this index
to which they are referring. The standard normalizes the index to the total number of
customers (refer to the standard for the exact formula).
Customer hours service availability
ASAI =
Customer hours service demand
ASIFI – Average System Interruption Frequency Index. This index is like SAIFI, except that it
normalizes the frequency to the amount of kVA load rather than number of customers. It
is essentially the same as the ALIFI index in the 1981 EPRI EL-2018 report [48]that
some of the reliability analysis tools compute.
ASIDI – Average System Interruption Duration Index. Similar to SAIDI, but based on the
duration that an amount of kVA load was interrupted rather than the number of customer
interruptions. Similar to the ALIDI indec in the 1981 EPRI report. [48]
CEMIn – Customers Experiencing Multiple Interruptions. This index is designed to track the
number, n, of sustained interruptions to a specific customer. Averages tend to obscure
particular trouble spots that need attention. This is an example of an index that is
becoming more important for reliability-centered maintenance (RCM) analysis.

Indices for Momentary Interruptions

In addition, indices for momentary interruptions have been defined:
MAIFI – Momentary Average Interruption Frequency Index. This is like SAIFI except that it
counts the momentary interruptions instead of sustained interruptions. It is based on
interrupting device counts. A high MAIFI would indicate areas subject to temporary
faults such as lightning, tree faults, or animal faults, requiring improved grounding, tree
trimming, and animal guards, respectively.
MAIFI E – Momentary Average Interruption Event Frequency Index. Similar to MAIFI except
that it counts events. It does not include the events that lead to a breaker or recloser
lockout.
CEMSMIn – Customers Experiencing Multiple Sustained Interruptions and Momentary
Interruption Events. This index is designed to track the number, n, of both sustained
interruptions and momentary interruption events to a set of specific customers. As with
CEMIn , this index is designed to identify problems that are obscured by the average
indices.

10253878 4-3
Major Event Days (MED)
As mentioned, major event days are excluded from the calculation of the reliability indices. Short
[50] reports that the IEEE Working Group on System Design developed a statistically based
definition for MED in IEEE Std. 1366-2012 called the 2.5-beta method. This method is based on
fitting daily SAIDI values to a log-normal distribution. Then a threshold (T MED ) is found where
the daily SAIDI is 2.5 time the standard deviation for a log-normal distribution. Short
summarizes the method in Chapter 10 of his book concisely as follows:

𝑇𝑀𝑀𝑀 = 𝑒 𝛼+2.5𝛽
where
T MED = Threshold for daily SAIDI for MED classification
α = average of natural log of daily SAIDI values
β = standard deviation of the natural log of the daily SAIDI values
Using 5 years of data to establish T MED , if possible, days during the next year that have a daily
SAIDI exceeding T MED are considered MED and are excluded.
Work continues to better define a MED. Some investigators have found a value of β=4, or more,
to be more representative of their locale.

Power Quality Indices

The EPRI Reliability Benchmarking Methodology report [51] built upon the results of the EPRI
Assessment of Distribution System Power Quality project and defined several indices for each
category of power quality disturbance. [52]
There were approximately 25 indices defined in the report. One has found its way into common
usage:
SARFI x – System Average RMS Variation Frequency Index.
This index is analogous to SAIFI, except that it counts voltage sag events less than x percent or
voltage swells greater than x percent.
SARFI70 is one of the more popular indexes. That is, it counts the number of customers
experience voltages less than 70% and normalized to the total number of customers in the study
area. This value is critical for some industrial and commercial customers because it can lead to
motor contactors dropping out, which can be as severe and an actual interruption. SARFI80
would be an appropriate index for more sensitive customers.
This index is as important as SAIFI to some industrial and commercial customers. A voltage sag
can have as much effect on industrial process as an actual interruption because it causes a
shutdown to lasts longer than the sustained interruption criteria. It is also a key design indicator
for targeting problem areas in reliability-centered techniques.
For predictive reliability analysis, this index is computed in an entirely different manner than the
traditional sustained interruption indices. It is generally not possible to estimate SARFI from the
traditional predictive reliability analysis of simply summing failure rates. It requires some sort of

10253878 4-4
power flow that allows for faults to be applied so that voltage sag levels can be estimated. This is
not easy to do because most power flow algorithms will not converge with a fault on the system.
The voltage sags too low. The EPRI OpenDSS program has been adapted to perform such
calculations by altering the load model (see Figure 2-19).
The methods used to compute SARFI can generally produce most of the other indices as well.

Predictive Reliability Analysis

References [48], [53], and [54] are the fundamental references for computing measures of
reliability for distribution systems. These reports and papers focus on radial distribution systems
in North America. This could include substation and subtransmission reliability. However,
distribution planners often consider only the radial feeders starting with the primary distribution
(MV) bus at the substation. Therefore, the analysis is much simpler than if networked systems
were considered.
Reliability analysis tools with a European heritage have a different definition of “distribution.” In
those, distribution systems generally include a significant amount of HV (transmission or
subtransmission) systems supplying the local MV systems (i.e., primary distribution).This
requires a different approach to reliability analysis that is often quite similar to what American
utilities would do for generation and transmission system analysis. There are advantages and
disadvantages to each approach.
Brown, et. al, [55] identify four categories for distribution reliability assessment techniques:
1. Network modeling: This refers to modeling the system as a network of components
connected in series and parallel. It is a relatively simple technique commonly described in
basic reliability textbooks. However, it is sometimes too simple to properly account for the
complexities of the behavior of distribution system components.
2. Markov modeling: These techniques represent the various states of the system and the
transition rates between these states. This is a powerful framework for reliability analysis but
has some drawbacks. Matrix inversion is required to solve and the number of states in a
realistic distribution system can be enormous. Also, quantifying the probabilities for the
transitions is difficult. Gonen [56] gives a detailed example.
3. State enumeration: This is Brown’s terminology for methods that generate states for the
system by some means and then determine the impact of each state on system reliability. This
method is attractive if the system under consideration is in its normal operating condition
nearly all of the time. This is the case for U.S. distribution systems.
4. Monte Carlo simulation: This refers to representing possible events by probability density
functions and generating sequences of events randomly. There are several advantages to this
method, including the ability to represent complex system behavior. A drawback has been
that it is computationally intensive. However, it is gaining in popularity with the advent of
more powerful tools and with the promotion of well-known power industry reliability
analysts, e.g., Billinton [57]
Brown continues by defining a “systematic and cost effective process of using predictive
reliability models to improve distribution system reliability.” This is a reasonable breakdown
and, perhaps, there is some gathering industry consensus concerning reliability analysis. Vendors
of distribution system analysis software have implemented various aspects of these steps:

10253878 4-5
1. Create a reliability model of the system. This involves constructing a model of the
distribution system in some level of detail. Feeder topology is important, especially the
location of sectionalizing devices including breakers, reclosers, sectionalizers, line switches,
and tie switches. Reliability data is provided for each component. This typically includes at
least failure rates and repair times. Some vendors will break each of these down in greater
detail depending on the needs of their customer base. Model flexibility and customization are
highly valued characteristics of reliability analysis tools and one of the areas prime for
exploitation in future research.
2. Calibrate the system to historical performance. The exact failure characteristics of each
system component is generally not known and the reliability analysis typically begins with
default failure rates, etc. Reliability analysis software suppliers commonly provide easy
means to set the default rates and then allow users to specify exceptional data separately.
When historical system performance is known, the default values can be adjusted by
automatic, or semi-automatic, algorithms to better calibrate to known historical performance.
Sometime, this calibration step evolves slowly by using the tool over several years as the
accuracy of historical data improves.
3. Perform a root cause analysis. This step is often omitted in distribution reliability analysis,
but can provide very valuable information to the utility engineer about which component
failure issues are important to address. Root cause analysis can place an additional
recordkeeping burden on the utility to break down the various causes of failures. Some
software suppliers are now providing tools to ease this burden. This is one area where model
customization is essential because different utilities will have much different failure causes
due to local environment, construction practices, and operating practices.
4. Perform a sensitivity analysis. This refers to determining how much a particular index
(commonly SAIFI and SAIDI) will change given a change in a parameter. In typical usage of
distribution reliability analysis by distribution engineers, there is a strong emphasis on
topological changes such as adding a mid-line recloser. However, some analysts also
encourage varying other parameters such as failure rate and repair time assumptions. This
reveals more clearly where priorities should be placed on maintenance, scheduling of crews,
etc. For many, sensitivity analysis is the most import step in reliability analysis; it is less
important to calibrate to historical data. The key decision variable is how much does the
predicted reliability change with the proposed change to the system. There is validity to this
argument.
5. Perform an economic analysis. This final step in reliability analysis would ideally be used to
prioritize investment option and action plans. While some of the distribution reliability
analysis tools simply report raw reliability index values based primarily on technical
considerations, others allow various cost models to be applied to the results. There are likely
as many ways to apply cost models to reliability data as there are entities doing the analysis.
Because corporate and societal economic considerations can completely reverse
prioritizations based on purely technical considerations, this is one of the primary areas for
future research. Tools to capture all the economic perspectives must have good, flexible
facilities for customization.

10253878 4-6
Analytical Methods in Radial Systems
Most radial distribution system analysis tools are set up quite nicely to perform standard
predictive reliability analysis of distribution system using the methods of Reference [48] and
similar methods. The radial tree structure of the circuit is cataloged and indexed so that
algorithms can sweep through the circuit components in topological order with ease. The power
flow solution methods frequently employ some sort of forward-backward sweep method that
requires an efficient means to do this.
The implementation of a simple failure rate and duration calculation for elements in series is an
easy addition to the forward-backward sweep scheme. One basic algorithm is, in words:
1. Start at the ends of the radial feeder,
2. Advance toward the source (backward sweep), accumulating failure rate and duration values
from each series component (much like one accumulates currents in the power flow
solution). One might keep separate accumulators for permanent faults and temporary faults if
computing both sustained and momentary indices.
3. When a sectionalizing device (switch, fuse, recloser, breaker, or sectionalizer) is
encountered, the accumulators are reset. There are exceptions and refinements one might
make, e.g., only partially resetting a temporary fault accumulator when a fuse is encountered.
4. Continue until reaching the head of the feeder, or another element defined as the source.
5. Start at the source end with an assumed or known failure rate for the bulk supply system.
6. Advance toward the ends of the feeder (forward sweep), computing the net bus reliability
quantities based on the accumulators from the backward sweep.
While this may be similar to the technique employed by many vendors to compute the base
failure rate and duration data needed to compute the reliability indices (by adding data
concerning the number of customers and load), the reader is advised that each software
implementation may have features that complicate this algorithm.
For example, some algorithms allow for the assumption that the load downstream from a failed
section can be automatically restored to service during the repair time. This accounts for schemes
such as auto-loop schemes that automatically restore power downline of a failed section. If the
interrupting device at the upline bus is an automatic device (breaker or recloser, but not a fuse),
the downline bus will be not be interrupted for faults on branches upstream. That is, the upline
branch fault rate is not accumulated to the downline bus on the forward sweep. For more details,
see Chapter 6, and Figure 6-2, in particular.

Reliability Analysis by Simulation

Stochastic methods are gaining popularity due to the inability of simple analytical algorithms
such as the one described above to capture the complex behaviors of distribution protective
schemes. The increasing application of intelligent control systems to distribution systems also is
driving reliability analysis in this direction. Also, the impact of DER connected to the system can
be to invalidate the simple methods.
These methods are also favored by those who include Power Quality disturbance indices with the
Reliability Analysis. This is particularly useful for analyzing systems with characteristics that
make it difficult to analyze with simple analytical formulae. A case in point would be a

10253878 4-7
distribution system with a substation neutral grounding reactance and loads connected through
delta/wye transformers. An example is presented below.
Sequential Monte Carlo
Billinton [57] has been researching these methods for some time and some of his recent work in
this area serves as an excellent reference. The basic idea is to simulate the annual load cycle
sequentially while using failure probabilities to simulate random failures in various components
as the simulation progresses. Quite complex behaviors of automatic switches and breakers can be
modeled in this manner as long as the computing time is not prohibitively long. Fortunately,
computer tools have been getting faster and more efficient at performing these simulations.
One idea is to simulate many years of operation to achieve a better estimate of statistical results
describing the actual risk of unserved energy for a given set of circumstances. It is difficult to
incorporate the load growth that might occur over this same time period directly into such a
method because the topology of the system is likely to change every few years in unpredictable
ways. Each topology with its particular set of protective devices will yield different reliability
indices. Another approach is to maintain the same topology and simply perform the simulation
assuming perhaps 100 years. Then the results are averaged to estimate the expected SAIDI,
SAIFI, etc. for a distribution system design.
Stochastic Fault Simulation
This technique uses some stochastic means to determine when and where a fault occurs and then
simulates the fault and the subsequent response of the system. The fault type and location may be
some sort of Monte Carlo technique. The load is often not considered in this analysis. So it is an
evaluation of the topology of the circuit with fault interrupter locations represented.
This was the technique implemented in EPRI’s PQ Planner. [59][60] This tool included a tree
growth simulator and permitted definitions of such root cause variables as weather, vegetation
level, line type and design, and local topography. Because its focus was on Power Quality (PQ),
it computed many of the RBM indices, especially the SARFI family. It is also capable of include
solutions to PQ problems that are not typically considered as solutions to reliability issues, such
as
• Distributed resources
• Disabled fast recloser tripping during fair weather,
• Current-limiting fuses
• Fast transfer switches
• Fuse saving
• Single-phase tripping
• Line reactors
• Etc.
The OpenDSS program has absorbed many of the functions of PQ Planner and has the advantage
of being customizable by the user.

10253878 4-8
The user specifies data describing the probability of faults of a particular type at each location.
The user also specifies how long to allow the simulations to proceed to arrive at statistically
meaningful index values. As with many types of reliability analysis, it is the incremental
difference in certain indices resulting from a proposed solution that is more important than the
absolute value of the index.
This approach is illustrated in the example that follows.

Example Reliability Study By Monte Carlo Analysis

This is an example of using capabilities of the EPRI OpenDSS computer program to determine
Power Quality as well as Reliability indices for a distribution system by Monte Carlo analysis.
This example is imported and updated from reference [47]. This example is frequently presented
during OpenDSS training workshops to demonstrate the Monte Carlo Faultstudy mode
(mode=MF) capabilities. Even seasoned distribution engineers are often surprised by the results.

SUBSTATION

FAULT WHAT DO THE

480 V
and 208 V
LOADS SEE?

Figure 4-1
Circuit Diagram for Case Study

A number of things that makes the distribution circuit in Figure 4-1 a challenge:
1. The substation transformer is grounded through a reactance. So there is a certain amount of
neutral shift for single line-to-ground (SLG) faults and other ground faults.
2. The feeders were a mixture of three-wire overhead and underground cable. The cables have
grounded sheaths (concentric neutral) that provide a path to ground while there was not
neutral conductor on the overhead sections. This is just minor complication that makes fault
behavior not obvious.
3. There were several three-phase commercial loads on the system fed as shown by two
delta/wye-connected transformers in series. The first steps down to 480/277V from which
most three-phase and many lighting loads were served. The 120V loads were served from
208/120V step-downs also connected delta-wye as shown.
This is a common type of service to commercial buildings. For example, 480-V service may be
distributed to each floor of a multi-story office building where it supplies a transformer and,
commonly, the lighting on that floor. The office equipment (computers and peripherals, coffee
pots, etc.) requires 120-V service. Most people in the office will interact with the electricity
supply system at 120-V loads. However, the lighting that may blink in response to disturbances

10253878 4-9
and breaker operations on the primary distribution feeder is often supplied from the 480-V
service. So both voltage levels are important and have different characteristics due to the
delta/wye transformers.
Figure 4-2 shows the one type of result from the analysis of 1000 random SLG faults. This was
done with the OpenDSS monte carlo faultstudy mode. It is the distribution of voltages observed
at a 120-V load for each fault. Interruptions were not modeled in this analysis; only the voltages
during the SLG faults. Voltages range from approximately 75 V to nearly 140 V. Many
distribution engineers are surprised the range is so narrow and there are no zero voltages in this
result despite the fact that the faults on the main feeder yielded zero volts line-to-ground. This is
the effect of the delta/wye transformations – in this case two in series – and the neutral
grounding configuration of the MV system.

Voltages seen at 120 V

160
140
120
100
80
60
40
20
0
0 200 400 600 800 1000 1200

V1 V2 V3

Figure 4-2
Distribution of Voltages Observed at 120-V Loads for 1000 Random SLG Faults

While most of the faults resulted in a voltage drop, a few actually resulted in a voltage rise. This
would depend on the location of the fault and the phase that was faulted.
Another, larger and more detailed Monte Carlo fault study was executed on this circuit. The
main objective was to determine what voltages were observed at the various loads for various
faults on the primary system. The substation and feeder grounding scheme and the two delta/wye
service transformers complicate the analysis. Software for performing this kind of reliability
study must be able to properly take into account the full three-phase model with all the
connections.
The fault incidence rate for each of the primary feeders was determined from historical data. The
EPRI OpenDSS program was driven by a VBA program written in a MS Excel spreadsheet that
applied faults in a somewhat random manner across the feeders according to the historical fault
incidence rate. The resulting voltages were collected back into the spreadsheet and the sag and
outage statistics analyzed.

10253878 4-10
A typical Monte Carlo run consisted of applying 50,000 faults of various types in proportion
according to the historical data. Historical data indicated the fault resistances were generally low,
so a log-normal distribution was assumed and the fault resistances selected randomly according
to the distribution. Protective devices (feeder breakers) were modeled and interruptions were
captured.
The results are shown in Figure 4-3 and Figure 4-4. One of the advantages of this type of
reliability study is that not only is an estimate of the classical reliability indices obtained but the
Power Quality indices fall out naturally. This is important for distribution systems serving
industrial and commercial consumers because it gives the planner an idea of how sensitive
customers are likely to view the quality of service. A quite skewed voltage sag distribution is
predicted.

480-volt Expected Distribution of Voltages

8000 100%
Number of Samples 90%
7000 Cumulative Percentage

80%

Cumulative Percentage
6000
Number of Samples

70%
5000
60%

4000 50%

40%
3000
30%
2000
20%
1000
10%

0 0%
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95
Voltage (pu)

Figure 4-3
Expected Distrribution of Voltages at a 480V Bus

10253878 4-11
 208-volt Expected Distribution of Voltages

5000 100%

4500 Number of Samples 90%

Cumulative Percentage

4000 80%

Cumulative Percentage
Number of Samples

3500 70%

3000 60%

2500 50%

2000 40%

1500 30%

1000 20%

500 10%

0 0%
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95
Voltage (pu)

Figure 4-4
Expected Distribution of Voltages at a 208 V Bus

The 480-V bus sag voltages are clustered around 80% while the 120/208-V bus voltages were
lower in this case, around 70%. Many motor loads will drop out for voltages in this range. If we
assume 3-phase 480-V motors will drop out at 80% voltage, more than half the faults are likely
to cause motor dropouts even if there is no breaker operation.
The interruptions show up as the significant number of samples at 0V. This represents the
interruption that would occur when the fault is on the same feeder as the load. The way the faults
were distributed across the three feeders, less than 10% occurred on the feeder being monitored.
These data were normalized to the number of customers on an annual basis based on the
historical fault rate and the expected SARFIx indices computed. These are shown in Table 4-1.
Due to the way the system was configured, the only time the loads experienced a voltage lower
than 50% was for a feeder breaker interruption. Thus, SARFI50 and SARFI 10 are equivalent to
SAIFI = 0.6. This is an effective way to perform predictive reliability analysis and obtain more
useful information about the system.
Table 4-1
Annual SARFIx Results

SARFI Level (480-V V LN ) Number of Events (at or below this level)

SARFI 90 3.5
SARFI 80 2.3
SARFI 70 1.0
SARFI 50 0.6
SARFI 10 0.6

10253878 4-12
5
MODELS OF CIRCUIT ELEMENTS
Introduction
In this chapter, the models of lines, transformers, loads, and other components that need to be
included in planning studies for the modern distribution grid will be explored. Some of the
models have been used for decades, but others represent new kinds of devices such as PV
systems and storage systems that will be needed in the near future.
There is some overlap in the material in this chapter and that in the previous chapter on power
flow analysis. It could be useful to review that material.

Transformer Models for Distribution Analysis

The per-unit system was developed to remove the transformers from the circuit model to make
things easier for manual calculations. This is satisfactory for transmission systems that are
mostly balanced. However, transformers must be modeled in detail for most types of unbalanced
3-phase (or n-phase) analysis of distribution systems. This is perhaps the key element to model
correctly to enable the detailed models required for the more advanced analyses of the modern
grid. Modelers need to understand not only the winding connections and impedances, but the
type of core design as well.
The key data for modeling transformers for distribution are:
• Number of phases.
• Number of windings.
• Connection of each winding (delta, wye, single-phase L-N, etc.).
• Voltage rating of each winding.
• kVA or MVA rating of each winding.
• Short circuit impedances. If n windings, require n(n-1)/2 values. Usually found on the
nameplate and/or transformer test report.
• No load losses for distribution efficiency studies.
• Load losses (if not included in the short circuit resistances)
• Core construction (shell, 3-legged core, 5-legged core)
The development of the transformer model presented here follows McWhirter’s method of
building the primitive admittance matrix directly from short-circuit impedance measurements on
the transformer. [61] This is simpler and generally more accurate than computing the inductance
or impedance matrix from finite element field calculations.

10253878 5-1
The method has been updated with modern linear algebra notation [62]. The steps involved are:
1. Compute short circuit matrix, Z B , on 1-volt base.
2. Perform a reference frame transformation to compute Y on a 1-V base.
3. Apply turns ratios or voltage ratios to get actual siemens.
4. Apply winding connection transformation to obtain the primitive Y matrix for the complete
transformer.
Given the short circuit impedances between each pair of windings, Z SCi,j , compute the elements
of the Z B matrix as follows:

Diagonal Elements of ZB
zBii = zSC 1, i+1 × zbase for i = 1 to n-1
Off-diagonal Elements of ZB:
zBij = 0.5[zBii + zBjj - zScj+1,i+1 × zbase] i≠ j

Now, apply a series of power invariant transformations to obtain the complete Y matrix. Refer to
Figure 5-2 for the definitions of Y 1 and Y W.

Figure 5-1
Transformation of short circuit impedances into the primitive Y matrix for a transformer

The matrices A and B are simply incidence matrices that indicate how things are connected in
the model.
The N matrix captures the actual voltage ratings of the windings that turn the short circuit model
on 1-volt base into actual volts and siemens.

10253878 5-2
Figure 5-2
Three-phase Wye-Delta transformer schematic

Two-winding transformers are straightforward. Now let’s consider a more complicated

transformer, the Y-Delta-Y connection on a 3-legged core.
Since there are 3 windings, there are 3(2)/2 =3 short circuit measurements required. This is
shown in the X SC matrix in Figure 5-3.
The vast majority (likely more than 90%) of 3-phase utility distribution HV/MV substation
transformers 5 MVA, or larger, are core-form transformers (i.e., 3-legged core designs –
sometimes called “E” cores). The windings for each phase are installed over the three legs as
depicted in Figure 5-4.
The electrical behavior that sets this design apart from other designs is that it behaves as if it has
a delta-connected winding even if one is not physically present. This is called a “phantom
tertiary” winding. The reason for this is depicted in the figure. There is no return path for zero-
sequence flux totally within the core steel. The return flux path is through the insulating medium,
which is some combination of oil, paper, air and, possibly, other gases. This path has low
permeability (typically µr=1) and, therefore, higher reluctance, than the core steel. So the zero-
sequence magnetizing reactance is substantially lower than the positive-sequence magnetizing
impedance for which the flux stays in the core steel.

10253878 5-3
Figure 5-3
The 3 short circuit measurements required to model a wye-delta-wye transformer

Figure 5-4
Three-legged core transformers offer a low impedance to zero sequence currents because the flux
path has a large gap outside the steel.

It is relatively easy to model this. You can simply add a delta-connected winding to the
transformer model (Figure 5-5). If it already has a delta winding, you have the option of simply
reducing the impedance to the existing delta winding by 5-10% or adding another delta winding.
The leakage reactance to the equivalent, or “phantom,” tertiary is usually fairly high on the
transformer’s MVA base. For an existing 2-winding transformer, add a 3rd winding connected in
delta and set the initial guess for the percent reactances to the phantom tertiary to the following:
X HT = 100 %
X LT = 80 %
X HT is generally higher because the HV winding is usually wound at a larger radius than the LV
and encompasses more area – therefore, there is more flux in the space between the windings.
These values are frequently good enough if you have no other data. If you have zero-sequence
test data available, you can adjust the estimated values needed to provide the measured zero
sequence impedance. In programs like OpenDSS you can quickly and easily apply a 1-phase
fault to the load side of the transformer and adjust the values to give the expected results.

10253878 5-4
Returning to the 3-winding Y-delta-Y transformer, this kind of transformer is used in several
substations where the transmission system is weak to help hold the unfaulted phase voltages
down. The HV Y interacts with the delta winding to act as a ground source. Some other utilities
served from strong transmission systems also use this kind of transformer, being convinced that
they get a better, more robust transformer. Most of these transformers are built on 3-legged
cores. Thus, they also have the phantom tertiary winding effect.

Figure 5-5
The effect of a 3-legged core is like having an extra “phantom” delta winding

The building of this transformer model would start with 6 short circuit impedances between each
pair of windings (4(3)/2 = 6). The 3-winding model shown previously is expanded by adding a
row and column to account for the phantom delta winding (Figure 5-6). The task now becomes
deciding what values to use for the impedances to the phantom tertiary.

Figure 5-6
A 4-winding transformer requires 6 short circuit measurements (matrix is symmetrical; lower
triangle shown)

Estimating Impedances to the Phantom Winding

The values of the short circuit reactances from the actual windings to the phantom winding are
typically in the range of 75 – 200 % on the main transformer’s MVA base. In parallel with the
physical delta winding, for which the reactances are typically in the 7 – 30% range, this reduces
the net zero-sequence impedance by 5-10% from the positive-sequence impedance. To better
understand what values to choose, it is necessary to understand more about how transformers are
constructed.
As shown in Figure 5-7, the physical delta Tertiary winding is generally wound next to the core.
Then the Low voltage winding and the High voltage winding are wound on top of the Tertiary in
that order. The short-circuit reactance is proportional to the physical space between the
windings. The greater the space, the higher the reactance. Proportions vary between designs. The
default values in the OpenDSS transformer model are XHL=7%, XHT=35% and XLT = 30%,
This gives an idea for the range of values. The Low voltage winding is closer to the Tertiary
than the High voltage winding. Thus, XLT is lower than XHT.

10253878 5-5
For estimating the short circuit reactance to the phantom winding, we need to come up with three
values as indicated in Figure 5-6. One might think of the phantom winding as occupying the
space inside the Tertiary winding. The Tertiary (3rd) winding is closest, so X3-4 would be
smallest of the three and X1-4 (High to phantom) would be the largest. Knowing the reactances
must be in these proportions, we can make some educated guesses and can generally match test
results after a few quick sample short circuit calculations.

Figure 5-7
Typical construction of a 3-winding core form power transformer

Line Models
In this section, models of unbalanced line and cable configurations are examined. Lines and
cables, with neutrals. play an important role in full 3-phase modeling of distribution systems.
Overhead Lines
Figure 5-8 shows a conceptual drawing of the geometry of a typical distribution line construction
for a 4-wire multi-grounded neutral system with 8-ft crossarms. The neutral wire is mounted 5 ft
below the phase conductors.

Figure 5-8
Overhead line geometry for a horizontal crossarm construction for a 4-wired multi-grounded
neutral system [64]

10253878 5-6
Most distribution system analysis tools have the capabilities to compute the needed impedances
from the given geometry and wire characteristics. In the EPRI OpenDSS program, this geometry
is modeled by this script snippet:
set earthmodel=Carson

New WireData.OH_477_AAC diam=0.793 radunits=in GMRac=0.3 GMRunits=in Rac=0.1968

~ Runits=mi normamps=650

New LineGeometry.OH3P_FR8_N56_OH_477_AAC_OH_477_AAC_ABCN nconds=4 nphases=3

~ reduce=no
~ cond=1 wire =OH_477_AAC x=-3.67 h=31.33 units=ft
~ cond=2 wire =OH_477_AAC x=0 h=31.33 units=ft
~ cond=3 wire =OH_477_AAC x=3.67 h=31.33 units=ft
~ cond=4 wire =OH_477_AAC x=0 h=26 units=ft

// To show the impedance data calculated by OpenDSS

show lineconstant freq=60 units=mi

Horton, et. al., [63] provide the equations for computing the 4x4 impedance matrices for this
kind of pole configuration including the modeling of earth return.
To Kron or Not To Kron
Note the command “reduce=no” in the script. If it had said “reduce=yes” the program would
have executed the well-known Kron reduction on the impedance and capacitance matrices and
eliminated conductor 4, the neutral conductor. This raises an interesting question about when to
perform a Kron reduction and when not to.
For both transmission and distribution systems, power system analysts typically eliminate all
wires assumed to be grounded, leaving only the three phase conductors of 3-phase lines. An
exception would be lightning surge analysis using an EMT program. The assumption of the Kron
reduction is that the voltage on the neutral conductors is zero. This works acceptably for most
power flow capacity studies if there is not any interest in neutral currents. The reduced 3x3
matrices give the same answers for the phase voltages and currents as the 4x4 matrix for power
flow and other phasor-domain planning analysis. This allows the model to be reduced a bit in
size.
The problem arises when analysts want to know the currents in the neutral wire. It is not possible
to expand the 3x3 back to the 4x4 system. One reason for wanting to know the currents in the
neutral explicitly is to compute the neutral-to-earth voltages (NEV), or stray voltages. A
schematic of the 4-wire multigrounded system for NEV analysis is shown in Figure 5-9. For the
model, the zero-voltage reference for the circuit is remote earth. To model this properly, the
impedances of the pole downleads and grounding electrodes must also be modeled. The pole
grounds are generally assumed to be approximately 10-25 ohms, but could be as high as 100
ohms, so they are quite significant. Note that the load neutral point is connected to the line
neutral at the top of the pole and is certainly not zero volts.
Inexperienced modelers wanting to see the neutral conductor currents will often leave the 4x4
impedance matrices intact from the line constants calculation and simply connect the ends of the
neutral conductors to ground, or the zero-volt reference, without realizing there has been a
reference frame change. To get the correct answer it is necessary to model all the other

10253878 5-7
impedances shown in Figure 5-9. Fortunately, this detailed model is not necessary for most
distribution planning. The detailed model is necessary for some harmonics analysis and for
ground fault voltage rise calculations, which are performed routinely on long transmission lines.
The analysis is also sometimes performed on distribution lines with more than one circuit on the
same pole sharing a common neutral.
Note that when using the reduced 3x3 model, the sum of the three phase currents is not the
neutral current. It is the sum of the current in the neutral and the earth return path. On a large
distribution system with thousands of connections from neutral to earth, there can be more
current in the earth than the neutral.

Figure 5-9
Neutral-to-earth voltage (NEV) test feeder requiring modeling of pole downlead impedances and
grounding resistances [63]

Concentric Neutral Cables

Cable data models will be described from OpenDSS documentation by Ballanti and Dugan. [64]
This cable owes its name to the neutral conductors that encircle the phase conductors as shown in
Figure 5-10. [65] To describe a concentric neutral cable the following properties are required
(using OpenDSS property names):
DiaCable = Diameter over cable; same units as radius; no default
DiaIns = Diameter over insulation layer; same units as radius; no default. Establishes
outer radius for capacitance calculation
diam = Diameter of phase conductors; Alternative method for entering radius
DiaStrand = Diameter of a concentric neutral strand; same units as core conductor radius; no
default
Emergamps = Emergency ampacity, amperes. Defaults to 1.5 * Normal Amps if not specified
EpsR = Insulation layer relative permittivity; default is 2.3
GMRac = Geometric Mean Radius (GMR) at 60 Hz. Defaults to .7788*radius if not
specified.
GmrStrand = Geometric mean radius (GMR) of a concentric neutral strand; same units as
core conductor GMR; defaults to 0.7788 * CN strand radius

10253878 5-8
GMRunits = Units for Geometric mean radius (GMR): {mi|kft|km|m|Ft|in|cm|mm}
Default=none
InsLayer = Insulation layer thickness; same units as radius; no default. With DiaIns,
establishes inner radius for capacitance calculation
k = Number of concentric neutral strands; default is 2
normamps = Normal ampacity, amperes. Default equal to Emergency amps/1.5 if not
specified
Rac = Phase Conductor Resistance at 60 Hz per unit length. Defaults to 1.02*Rdc if
not specified
Radius = Outside radius of conductor. Defaults to GMR/0.7788 if not specified.
Radunits = Units for outside radius: {mi|kft|km|m|Ft|in|cm|mm} Default=none.
Rdc = dc Resistance, ohms per unit length (see Runits). Defaults to Rac/1.02 if not
specified
Rstrand = AC resistance of a concentric neutral strand ; same units as core conductor
resistance ; no default.defined. Ignored for symmetrical components
Runits = Length units for resistance: ohms per {mi|kft|km|m|Ft|in|cm|mm}

a) b)

Figure 5-10
Structure of a concentric neutral cable: a) cutaway view b) simplified cross section (not in
proportion)

The calculation of the impedances for a 3-phase cable installation yields a 6x6 matrix – 3 phases
and 3 equivalent neutral conductors. Unless the cables are quite long and the engineer is
considering cross-bonding the neutrals, the voltages on the neutrals are assumed to be zero and a
Kron reduction is performed. This is a good assumption in most underground distribution
systems because the neutral/shields are bonded to ground frequently at every padmounted
transformer and at riser poles.

10253878 5-9
Tape Shielded Cable
The physical features of tape shielded cable are shown in Figure 5-11 . [66] The following
properties are required, again using the property names in OpenDSS:
DiaCable = Diameter over cable; same units as radius; no default
DiaIns = Diameter over insulation layer; same units as radius; no default. Establishes
outer radius for capacitance calculation
diam = Diameter of phase conductors; Alternative method for entering radius
DiaShield = Diameter over tape shield; same units as radius; no default.
Emergamps = Emergency ampacity, amperes. Defaults to 1.5 * Normal Amps if not specified
EpsR = Insulation layer relative permittivity; default is 2.3
GMRac = Geometric Mean Radius (GMR) at 60 Hz. Defaults to .7788*radius if not
specified. Zero-Sequence capacitance, nF per unit length
GMRunits = Units for Geometric mean radius (GMR): {mi|kft|km|m|Ft|in|cm|mm}
Default=none
InsLayer = Insulation layer thickness; same units as radius; no default. With DiaIns,
establishes inner radius for capacitance calculation
normamps = Normal ampacity, amperes. Default equal to Emergency amps/1.5 if not
specified
Rac = Phase Conductor Resistance at 60 Hz per unit length. Defaults to 1.02*Rdc if
not specified
Radius = Outside radius of conductor. Defaults to GMR/0.7788 if not specified.
Radunits = Units for outside radius: {mi|kft|km|m|Ft|in|cm|mm} Default=none.
Rdc = dc Resistance, ohms per unit length (see Runits). Defaults to Rac/1.02 if not
specified
Runits = Length units for resistance: ohms per {mi|kft|km|m|Ft|in|cm|mm} Default=none
TapLap = Tape Lap in percent; default 20.0
TapLayer = Tape shield thickness; same units as radius; no default.

a) b)
Figure 5-11
Structure of a tape shielded cable: a) cutaway view b) simplified cross section (not in proportion)

10253878 5-10
Load Models
For power flow studies load is basically modeled as a current source as a function of the applied
bus voltage and the desired power (kW and kvar). EPRI has identified 8 types of load models
through its research and study experience that are important for distribution planning studies:
1. The standard constant P+jQ load typically expected for power flow analysis. Both P and Q
are varied separately or together for QSTS analysis.
2. Constant impedance load. This is a linear impedance, so the power varies by the square of the
voltage magnitude.
3. Constant P, Quadratic Q. This is used to model motors in some studies where the shaft power
is relatively constant but the reactive power is proportional to the voltage squared. Note that
for CVR studies the reactive power may vary by V3 or even V4, so the next model might be
better for that.
4. Nominal Linear P, Quadratic Q. This is a model that represents a typical feeder mix in some
utilities. The exponents one the voltage variation can be modified. Therefore, this model is
employed when doing conservation voltage reduction (CVR) studies and the exponents are
varied from the nominal assumption.
5. Constant Current Magnitude. The current changes phase angle with the voltage phase angle,
but does not change magnitude. Many distribution power flow tools use this model.
6. Constant P, Fixed Q. This is like the standard constant power load model except that for
QSTS simulations the reactive power, Q, does not change.
7. Constant P, Fixed Impedance Q. This is like the standard constant power load model except
that for QSTS simulations the reactive power, Q, varies with the square of the voltage
magnitude like a reactor.
8. ZIP: A combined constant Z, constant I (current), and constant power model like that
supported in many distribution power flow programs.
These 8 models are implemented in the OpenDSS program’s Load element model.
QSTS simulations are supported by providing a loadshape curve typically with per-unit
multipliers for both P and Q, which are specified with an initial base value – typically, the
forecasted peak load for the planning study. Actual P and Q values may also be used.
More detail on the load model is provided in Chapter 2. Many loading situations can arise that
will tend to cause the power flow solution to fail to converge. Load models should have some
protection against convergence failure, such as switching to a linear model when the voltage
goes outside the normal limits.

Storage Models
Storage devices are projected to play a big role in the modern integrated grid. Some governments
have mandated the installation of large amounts of storage. Even though it might be needed to
smooth renewable energy variations on the transmission grid, some of the storage will be
installed on the distribution system. Therefore, it is expected that modeling storage for
distribution planning studies will be essential for some utilities. This section will address some of
the requirements for modeling distribution-connected storage.

10253878 5-11
Applications of storage on distribution systems include:
• Smoothing solar PV power output.
• Extending solar PV output into the evening peak.
• Support of the Transmission grid.
• Extending capacity of existing assets.
• Supporting alternate feeds during reconfiguration after faults.
• Controlling frequency of a microgrid.
• Increasing the short circuit strength of a microgrid so that it is more stable and might be able
to operate overcurrent protection elements
These are excellent applications for storage, but there are certain distribution planning issues
created by storage that planners may need to address with appropriate planning tools:
• Overvoltages while discharging. This could occur if storage were to be dispatched to produce
power at a time when the load is low on the distribution system. This is the same problem
that is encountered with solar power on sunny days with low demand.
• Low voltages while charging. Like any other load on the system, there is the possibility of
storage pulling the voltage down as it charges and draws power from the distribution system.
• Voltage regulation while compensating for transmission grid support. If the storage is
primarily dispatched to charge and discharge to support the needs of the transmission system,
simulations of the charge/discharge cycles during normal and unusual load cycles will need
to be evaluated. This would involve QSTS simulations including regulator and capacitor
switching activity as well as actions of edge-of-grid technologies.
• Interference with overcurrent protection scheme. Storage will have many of the same issues
of contributing to short circuit faults as other DER.
• Insufficient short-circuit capacity in microgrid. The distribution system is currently protected
by a large number of overcurrent devices. If a small section of the system is carved out and
operated as a microgrid, what happens when a fault occurs? Will there be enough capacity to
operate the overcurrent protection or will the existing protection system have to be
supplemented by other schemes?
• Storage is not only a variable resource but it is also a limited resource. Existing planning
tools (power flows) are almost exclusively focused on power. They will have to be modified
to not only perform a single-shot power flow but to integrate over time to account for the
energy stored.
• Storage is effective providing energy on demand, but also can have a limited ramp rate. It is
difficult to model this aspect without getting into dynamics analysis, which is not common in
distribution planning tools.
• Storage has to be recharged at some time after it has been discharged. Is it done immediately
or do we wait until the sun comes up the next day or at 2:00 a.m. when there is very little
load? These are decisions planners will have to make as storage becomes an integral part of
the distribution system

10253878 5-12
• Storage has losses that need to be modeled in studies. The charge/discharge cycle typically
has an efficiency of 80-90%. The model can simply discount the power flow by the
efficiency or model the resistances of the battery and other devices. The former is more likely
to be compatible with the planning environment because planners are unlikely to have time to
track down all the component characteristics in a storage system. There are also idling losses
for such things as heating or cooling batteries that should be accounted for. Thus, these losses
are dependent on the ambient temperature.

Figure 5-12
Generic storage model used in the EPRI OpenDSS program

Figure 5-12 shows the generic Storage model from the EPRI OpenDSS program that has features
to address most of the issues mentioned in a distribution planning environment. This model can
be used in a standalone mode, but can also be controlled by the StorageController model
depicted in Figure 5-13. This model was designed to control a fleet of distributed storage
elements for several research projects, including references [12] and [69]. Storage is not
necessarily easy to model – the Storage and StorageController models contain the most complex
logic of any similar set of models in the OpenDSS program.

Figure 5-13
Storage controller model from EPRI OpenDSS program

10253878 5-13
EPRI has identified six operating modes for the storage controller to perform for planning
studies.
1. Static: Set the storage to charge or discharge at specific rate for power flow calculation.
2. Time: Set the storage to discharge or charge at specified times.
3. Peak Shaving: When the load being monitored reaches a selected value, dispatch the storage
to compensate for further increases in load.
4. Load Following: Dispatch the storage on at a selected time and hold the load at or below that
value.
5. Loadshape Following: Dispatch the storage in the simulation according to a defined
loadshape.
6. Dynamics (i.e., electromechanical transients)
Table 5-1
Typical model properties required for modeling storage elements for planning analysis

Property Description
Bus reference Bus to which the Storage element is connected.
Number of phases Number of Phases, this Storage element.
Connection Wye or delta or single phase.
Voltage rating, kv Nominal rated (1.0 per unit) voltage, kV, for Storage element.
kVA rating of power output. Used as the base for Dynamics and Harmonics
kVA rating
model values.
Rated kWh capacity Rated storage capacity in kWh..
kW rating kW rating of power output..
kWh stored Present amount of energy stored, kWh.
Percent of rated kWh storage capacity to be held in reserve for normal
%reserve
operation.
%Charge Charging rate (input power) in Percent of rated kW.
%Discharge Discharge rate (output power) in Percent of rated kW.
%Efficiency Charge Percent efficiency for charging the storage element.
%Efficiency Discharge Percent efficiency for discharging the storage element.
%Idling kvar Percent of rated kW consumed as reactive power (kvar) while idling.
%Idling kW Percent of rated kW consumed while idling.
Dispatch curve Dispatch shape to use for QSTS simulations.
%stored Present amount of energy stored, % of rated kWh.
Equivalent percent internal resistance, ohms. Placed in series with internal
%R
voltage source for harmonics and dynamics modes.

10253878 5-14
Table 5-1 (continued)
Typical model properties required for modeling storage elements for planning analysis

Property Description
Equivalent percent internal reactance, ohms. Placed in series with internal
%X
voltage source for harmonics and dynamics modes. (
{Yes | No*} Default is No. Force balanced current only for 3-phase
Balanced
PVSystems. Forces zero- and negative-sequence to zero.
ChargeTrigger Dispatch trigger value for charging the storage.
DischargeTrigger Dispatch trigger value for discharging the storage.
Dispatch Mode Reference to the mode being simulate – compatible with simulation tool.
Name of a DLL containing user-written dynamics model, which computes
Dynamics DLL the terminal currents for Dynamics-mode simulations, overriding the default
model.
kvar Alternative to specifying the output power factor
Terminal power, kW Terminal output power, kW
Limit Current option Limits current magnitude to a selected value.
The model to use for power output variation with voltage. Const kW,
Power model
constant Z, or other.
pf The nominal power factor for discharging (acting as a generator).
Reference harmonic voltage or current spectrum for this storage element.
Harmonics spectrum
Current injection is assumed for inverter.
One of {IDLING | CHARGING | DISCHARGING} In DISCHARGING
mode, the Storage element acts as a generator and the kW property is
positive. The element continues discharging at the scheduled output power
Operating State of Storage
level until the storage reaches the reserve value. Then the state reverts to
Element
IDLING. In the CHARGING state, the Storage element behaves like a Load.
The element continues to charge until the max storage kWh is reached and
then switches to IDLING state.
Time of day at which storage element will automatically go into charge state
Time of Charge Trigger
to top off storage element.
Name of DLL containing user-written model, which computes the terminal
UserModel
currents for both power flow and dynamics, overriding the default model.
Maximum per unit voltage for which the Model is assumed to apply. Above
Vmax pu
this value, the load model reverts to a constant impedance model.
Minimum per unit voltage for which the Model is assumed to apply. Below
Vmin pu
this value, the load model reverts to a constant impedance model.

10253878 5-15
PV System Models
Solar PV systems may be modeled for distribution planning purposes in at least two different
ways depending on the level of detail desired.
1. Generic generator models that simply inject power (P+jQ) for static power flow. Also, if the
planning tool has the capability, planners can provide loadshapes and perform QSTS
simulations to determine how the PV system output can affect the voltages. This is an
important part of PV system modeling because of the varying nature of solar generation.
2. Use a specify model of PV, such as the PVSystem model in OpenDSS (Figure 5-14). This
allows the modeler to supply more details about the system, primarily for QSTS simulations.
A simple generator model suffices for snapshot power flows. This model takes the irradiance
shape instead of the power output shape. Combined with the temperature shape and the
inverter efficiency (supplied as a curve), the model computes the power output. In addition,
OpenDSS has an inverter controller model for representing advanced inverter behavior such
as volt-var control.
Solar PV QSTS simulation can be conducted on different time scales. Each will yield a different
view of the operational behavior. Figure 5-15 shows the results of a simulation with 1-h time
steps. Therefore, the curves seem quite smooth. This, or a 15-minute time step, are good for
determined the effective capacity for solar PV. The yellow-colored shaded area is the resultant
load shape after adding the solar PV generation. It is an interesting plot to illustrate a point about
the distribution planner’s perspective. While there is sufficient PV output to create reverse flow
occasionally, and the width of the load peak is narrowed so that it should be easier to serve,
distribution planners are reluctant to give solar PV any capacity credit because it does not reduce
the peak demand. Like many distribution systems, the peak demand occurs in the evening after
sunset.
Other common time step sizes are 1 minute and 1 s. These allow planners to inspect the
interaction between the generation and equipment on the feeder such as voltage regulators that
are affected by rapid variations in voltage. Figure 5-16 shows an actual PV output sampled in 1-s
intervals. It is obviously not smooth like the one in Figure 5-15. If we were to feed this into a
power flow program capable of QSTS simulation at this small time step, we could see something
like the voltages shown Figure 5-17. In this simulation, there is initially much activity involving
switched capacitors and voltage regulators as they seek to find a new equilibrium. Then the
voltages follow the rest of the PV output power. This activity would not have been observed if
the time step were 1 h or if only a static power flow solution have been performed.
Thus, the time scale used for the simulation depends on what the planner wishes to see.
Table 5-2 lists the typical data required to perform this analysis, taken mostly from the EPRI
OpenDSS PVSystem model.

10253878 5-16
Figure 5-14
EPRI OpenDSS generic PVSystem model.

Figure 5-15
Solar PV simulation at a 1-hour step

10253878 5-17
Figure 5-16
Actual 1-second solar PV data with variations caused by cloud transients.

Figure 5-17
A possible impact on feeder voltages of the solar PV generation in Figure 5-16.

10253878 5-18
Table 5-2
Typical data for specifying a PV system for planning analysis

Property Description
Bus to which the PVSystem element is connected. May include specific node
Bus Reference
specification.
Number of Phases, this PVSystem element. Power is evenly divided among
Number of phases
phases.
Connection Wye or Delta or single-phase, etc
Rated voltage Nominal rated (1.0 per unit) voltage, kV, for PVSystem element.
kVA rating of inverter. Also the base for Dynamics mode and Harmonics mode
kVA rating
values.
Power-Temperature Curve An X-Y Curve that describes the PV array per unit Pmpp vs Temperature curve.
An X-Y Curve, that describes the efficiency vs per unit of rated kVA for the
Inverter Efficiency Curve
inverter.
PF The nominal power factor for the output power.
The rated max power of the PV array for 1.0 kW/sq-m irradiance and a user-
Pmpp
selected array temperature.
% Pmpp Upper limit on active power as a percentage of Pmpp.
% of kVA rating of inverter. When the inverter is OFF, the power from the array
% cut-in power
must be greater than this valuefor the inverter to turn on.
% of kVA rating of inverter. When the inverter is ON, the inverter turns OFF
% cut-out power
when the power from the array drops below this value.
Equivalent percent internal resistance, ohms. Placed in series with internal
%R
voltage source for harmonics and dynamics modes.
Equivalent percent internal reactance, ohms. Placed in series with internal
%X
voltage source for harmonics and dynamics modes.
Force balanced current only for 3-phase PVSystems. Forces zero- and negative-
Balanced option
sequence to zero.
The present irradiance value in kW/sq-m. Used as base value for shape
Irradiance
multipliers.
Irradiance curve Dispatch shape to use for QSTS simulations.
Temperature curve Temperature shape for use for QSTS simulations.
Get/set the present kvar value. Setting this property forces the inverter to operate
kvar
in constant kvar mode.
Indicates the maximum reactive power generation/absorption (in kvar) for the
Kvar Limit
PV System
Harmonic spectrum Name of harmonic voltage or current spectrum for this PVSystem element.
Temperature The present Temperature to use for a power flow solution.
Maximum per unit voltage for which the Model is assumed to apply. Above this
Vmax pu
value, the load model reverts to a constant impedance model.
Minimum per unit voltage for which the Model is assumed to apply. Below this
Vmin pu
value, the load model reverts to a constant impedance model

10253878 5-19
 1.00 kW/m2
0.75 kW/m2
mpp =

I,amps
0.50 kW/m2

0.25 kW/m2

V, volts

Figure 5-18
Typical i-V curves for values of irradiance

1.2

0.8
Factor

0.6

0.4

0.2

0
0 25 50 75 100
Temperature, C

Figure 5-19
Typical power vs temperature curve

0.98

0.96

0.94
Efficiency

0.92

0.9

0.88

0.86

0.84
0 0.2 0.4 0.6 0.8 1
Per Unit Power

Figure 5-20
Typical inverter efficiency curve.

10253878 5-20
Regulator Models
Table 5-3
Typical data required for simulation of a regulator or OLTC control

Property Description
Bandwidth Bandwidth in VOLTS for the controlled bus
Rating, in amperes, of the primary CT rating for converting the line amps to control
CT Primary rating
amps. Typically is the same as the regulator rating..
Time delay, in seconds, from when the voltage goes out of band to when the tap
Time delay
changing begins. This is used to determine which regulator control will act first.
The time delay is adjusted inversely proportional to the amount the voltage is outside
Inverse time option
the band down to 10%.
Max Tap Change Maximum tap change to allow in a power flow iteration
PT phase For multi-phase transformers, the designation of the phase being monitored.
Ratio of the PT that converts the controlled winding voltage to the regulator control
PT ratio voltage. If the winding is Wye, the line-to-neutral voltage is used. Else, the line-to-line
voltage is used.
When regulating a remote bus the PT ratio required to convert actual voltage at the
Remote PT Ratio
remote bus to control voltage.
Reverse Bandwidth Bandwidth for operating in the reverse direction.
Time Delay in seconds (s) for executing the reversing action once the threshold for
Reverse Delay
reversing has been exceeded.
Indicates whether or not the regulator can be switched to regulate in the reverse
Reversible option direction. Typically applies only to line regulators and not to the LTC on a substation
transformer.
Reverse Neutral option Set this to Yes/True if you want the regulator to go to neutral in the reverse direction.
Reverse R R line drop compensator setting for reverse direction.
kW reverse power threshold for reversing the direction of the regulator. Default is 100.0
Reverse Threshold
kw.
Reverse Voltage Target Voltage setting in volts for operation in the reverse direction.
Reverse X X line drop compensator setting for reverse direction.
Delay in sec between tap changes. Default is 2. This is how long it takes between
Tap delay
changes after the first change.
An integer number indicating the tap position that the controlled transformer winding
Tap Number
tap position is currently at, or is being set to.
Voltage limit Voltage Limit for bus to which regulated winding is connected (e.g. first customer)..
Voltage regulator setting, in volts, for the winding being controlled. Multiplying this
Voltage Target value times the PT ratio should yield the voltage across the winding of the controlled
transformer
R R setting on the line drop compensator in the regulator, expressed in VOLTS.
X X setting on the line drop compensator in the regulator, expressed in VOLTS.

10253878 5-21
One question that frequently comes up when modeling regulators is what impedance should be
assumed for the autotransformer. Table 5-4 shows the impedances of selected regulators in % on
the through-kVA base – the base of the rated load controlled. That is, if the regulator transformer
is 50 kVA, it will serve 500 kVA of load. The impedance is in % on the 500 kVA base.
Table 5-4
Example regulator impedances (%) on through-kVA base (Source: Siemens-Allis Regulator
Engineering Manual, 1978)

kVA (through) kV %Z Full Lower %Z Neutral %Z Full Raise

110 7.62 .40 0 .50
762 7.62 0.50 0 0.40
2500 2.4 0.45 0.35 0.45
7500 13.8 0.75 0.50 0.75

The impedances vary in a narrow range and are approximately 0.5% at the most. Close to the
neutral tap the impedance is approximately zero for many designs. Therefore, some distribution
power flow programs neglect the impedance of the regulator entirely. However, the impedance
of the regulator can affect short circuit calculations. The larger regulators can have a more
significant impedance at neutral. Short of putting every tap in the model, an average value for all
taps should suffice. The impedance will be swamped by line impedances in the model.

Capacitor Models
Figure 2-10 depicts a typical capacitor bank model with a switch control that is used for
distribution system analysis.
A capacitor is one of the most linear devices on the distribution system. Although specified by its
kvar rating, it is not a const power element like a load. It is a constant impedance element. We’ll
first look at the data typically used to describe the capacitor and then in the next section examine
the data for the capacitor control
Many capacitor banks have an intentional reactor in series. It is either for detuning, harmonic
filtering, or minimizing inrush. The EPRI OpenDSS model will be used as a reference. It models
a capacitor as a multistep filter bank.

10253878 5-22
Table 5-5
Data typically required to define a capacitor bank for analysis

Property Description
bus1 Name of first bus of 2-terminal capacitor.
bus2 Name of 2nd bus (often ground or earth).
Total kvar, if one step, or ARRAY of kvar ratings for each step.
Rated power, kvar
This is the usual power system way to define a capacitor.
Usually L-L kV for a 3-phase bank. Otherwise the actual can rating
Voltage rating, kV
is specified. This is an important property.
Number of phases Number of phases (1, 2, or 3).
Connection of bank Wye or delta, typically
Number of steps Number of steps in this capacitor bank if a mult-step bank.
Alternative ways to define capacitor
Alternate way to define a capacitor. ARRAY of Capacitance, each
Capacitance, microfarads
phase, for each step, microfarads.
Alternate way to define a capacitor. Nodal cap. matrix,
Nodal C matrix (Admittance matrix form)
microfarads.
ARRAY of harmonics to which each step is tuned. Or define XL
Harmonics for tuned filter
array.
ARRAY of series inductive reactance(s) in each phase (line) for
Reactor value, each step
filter, ohms at base frequency
ARRAY of series resistances in each phase (line), ohms of each
R
step
ARRAY of integers {1|0} representing the state of each step
states (on|off). Capacitor control will modify this array as it turns steps on
or off.

A power factor capacitor is typically defined by its kvar rating and the voltage rating, in kV. But
it is modeled as a constant capacitive reactance, Xc, computed as follows:
𝑘𝑘𝑘𝑘 2
𝑋𝑋𝑐𝑐 = × 1000
𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘
If line-line voltage is used, use 3-phase kvar. If the capacitor can’s voltage rating is used, use the
can’s kvar rating.
Otherwise, the capacitor is specified by its capacitance, typically in µF. Complicated bank
topologies can often be described by a capacitance matrix in nodal admittance form.
Capacitor Controls
A capacitor control typically monitors the voltage and current at a selected location in the
distribution system and determines when to switch capacitor banks as the load varies. This is an
important function for QSTS simulations when there are multiple capacitor banks on a feeder.

10253878 5-23
The ON setting and OFF setting are usually set with some hysteresis so there is less likelihood of
hunting.
Table 5-6
Typical capacitor control settings used in QSTS simulations for planning studies

Property Description
Capacitor Reference to the of Capacitor element being controlled
Current, Voltage, kvar, PF, time, temperature. Specify the ON setting and OFF
Type of Control
setting appropriately with the type of control.
Value at which the control arms to switch the capacitor OFF (or reduce by one
OFF setting
step)
Value at which the control arms to switch the capacitor ON (or ratchet up a
ON setting
step).
CT Phase Designation of phase(s) being monitored for current or kvar control.
Ratio of the CT from line amps to control ampere setting for current and kvar
CT ratio
control types.
Dead time after capacitor is turned OFF before it can be turned back ON.
Dead Time
Default is 300 sec.
Time delay, in seconds, from when the control is armed before it sends out the
ON Delay Time switching command to turn ON. The control may reset before the action
actually occurs. This is used to determine which capacitor control will act first.
OFF Delay Time Time delay, in seconds, for control to turn OFF when present state is ON.
Reference to the circuit element, typically a line or transformer, to which the
Monitored element capacitor control's PT and/or CT are connected.There is no default; must be
specified.
For PF control option, min percent of total bank kvar at which control will close
% Minkvar
capacitor switch.
PT Phase Designation of the phase(s) being monitored for voltage control
PT RATIO Ratio of the PT that converts the monitored voltage to the control voltage.
Override Bus Name of bus to use for voltage override function
Maximum voltage, in volts. If the voltage across the capacitor divided by the
Vmax PT RATIO is greater than this voltage, the capacitor will switch OFF regardless
of other control settings
Minimum voltage, in volts. If the voltage across the capacitor divided by the
Vmin PT RATIO is less than this voltage, the capacitor will switch ON regardless of
other control settings
Voltage Override Option True/False. Uses Vmin and Vmax.

10253878 5-24
6
OPTIMAL RECLOSER/DA SWITCH SITING
The optimal switch location algorithm was implemented in two main software components as
depicted in Figure 6-1. This depicts the typical situation in which the algorithm for computing
the objective function and controlling the positions of the switches in the circuit was
implemented in Microsoft Excel VBA or Python. Most of the effort was done using Excel VBA.
The versatility of VBA and the speed at which changes can be made to the code allowed the
investigators to modify the algorithm and test it very efficiently.
The Excel VBA implementation is actually a bit faster than a standard Python implementation
for this analysis. The implementation used exploits “early binding” to the COM interface, which
makes the function calls go much faster than the standard late binding employed in Python. This
enhances the ability to execute the two-at-a-time siting algorithm. Advanced Python users can
achieve considerable speed-up using the “makepy” application.
The circuit modeling and the computation of the reliability indices that are used to compute the
objective function are handled in the OpenDSS program. The optimization driver program
develops scenarios to test and controls the execution of the OpenDSS program through its COM
interface. As indicated, there were multiple connections between the VBA code and the interface
properties and methods in the COM interface.

Figure 6-1
Overview of the Test Bed for Developing the Optimal Siting Algorithm

The computation time for the combined predictive reliability and siting algorithm was found to
be relatively short – generally a few seconds up to a few minutes at most. Therefore, the
exhaustive search technique was employed to find the configuration that yields the highest
values of the objective function.
The circuit model was not reduced. The circuit models used in the examples had every pole and
load location explicitly modeled and each 3-phase location on the primary (medium voltage)
feeder was tested.

10253878 6-1
Two basic analysis schemes commonly used for device siting in the industry were investigated:
1. Adding a single automated switching device sequentially.
2. Adding two devices simultaneously.
Since the placement of one device has an impact on the placement of the next device, the basic
idea of researching the second option is that placing two at a time could yield a more optimal
solution.
In addition, algorithms were executed for each of 1) assuming no restoration of downline
sections and 2) assuming sections downline of an automated switch were automatically restored
by closing tie in sufficient time to avoid being counted as interruptions.

Objective Functions
The user has the option of choosing one or more of a number of the quantities available from the
OpenDSS predictive reliability calculation to form an objective function.
For the siting analysis demonstrated in this report, the primary objective function was the
number of Customer Interruptions saved by the proposed switch configuration. Algebraically,
this may be written:

𝐶𝐶𝑠𝑠𝑠𝑠𝑠 = 𝐶𝐶𝑏𝑏𝑏𝑏 − 𝐶𝐶
Where
CI = Number of Customer Interruptions for the present case
CI base = Customer Interruptions for a reference case, typically the case before adding
reclosers or other automated switches.
The usual practice for converting CI saved into a cost is to assume a fixed cost per customer
interruption saved. Values in use range from $20 to $50 per CI saved . This number has been
typically based on the cost to repair and restore after a failure divided by the number of
customers restored. It is essentially the amount the utility is willing to spend to prevent a
customer interruption. This is a developing field of study introduced to the industry by Short and
Taylor [[23]] and Duke Energy in 2006. There have been related panel presentations made at
both the 2014 and 2015 IEEE PES General Meetings. Research is expected to continue in the
value of a customer interruption.

Assumption of Automated Restoration

There are two ways the OpenDSS program computes the reliability indices: with and without
assuming power to feeder sections downline from an automatic protective device are
automatically restored after the protective device operates. The basic predictive reliability
algorithm is described in Chapter 4.
On the backward sweep through the radial feeder, the branch fault rates are accumulated up to
each protective device encountered. Then the accumulator is reset. The assumption of automated
restoration is applied during the subsequent forward sweep through the circuit when the number
of interruptions at each bus is computed. If restoration is assumed the number of interruptions is

10253878 6-2
simply the number of faults in each feeder section alone; all other sections are assumed to be
independently restored. If not restored, the number of interruptions at a particular bus is
accumulated from all sections upstream, which is the conventional way to compute the number
of interruptions.

N.O.

Assumed Tie to
Alternate Feed

N.O.
No. Bus Interruptions Due to
Failures in Each Section Only

Figure 6-2
For Assumed Automatic Restoration Downline Section are Assumed to be Restored by
Automatically Closing Normally-Open Ties

Sections with fuses at the head are not assumed to be automatically restored.

Initialization of Switch Placement Algorithm

It is assumed that there will be a feeder breaker or recloser at the head of the feeder. The optimal
siting problem is for devices placed on the line. So the first task of the placement algorithm is to
place a device at the head of the feeder if there is not already one there.
After the initial power flow solution there is a branch sequence list established in the
EnergyMeter element in the OpenDSS model when it is placed at the head of the feeder. This list
is ordered such that a lower sequence index number guarantees that the branch is closer to the
head of the feeder than one with a higher sequence index number. This makes it simple to
program forward and backward sweeps through the circuit.
The algorithm starts at sequence index=1 and looks for a 3-phase branch that does not already
have an overcurrent protection device. The algorithm expressed in Excel VBA code is:
DSSMeters.SequenceIndex = iStart ' sets active cktelement to first element in sequence

Do While ((DSSCktElement.NumPhases <> 3) _

And (iStart < DSSMeters.SeqListSize)) _
Or DSSCktElement.HasOCPDevice
iStart = iStart + 1 ' skip and move on down the feeder
DSSMeters.SequenceIndex = iStart
Loop

Where
DSSMeters = A variable pointing to the collection of the energymeters in the OpenDSS
COM Interface
DSSCktElement = A variable pointing to the active circuit element in the COM interface

10253878 6-3
This while loop moves the active sequence index down the feeder until it finds a 3-phase element
that does not have an automatic switch device already. Then it creates a Recloser at this location
(the OpenDSS Recloser object is used as a proxy to represent any type of automated switch for
this algorithm). Subsequent searches for another location will start at the next branch
downstream in the sequence list.
Most of the time, it is expected that the initial device added will be on the first branch in the
sequence list, but it wouldn’t have to be.
Note that only power delivery elements such as Line and Transformer objects will be in the
search list. Load objects and other shunt-connected objects will not be in the list.

Placement of One DA Switch Algorithm

This algorithm for placing one device is executed for each device sequentially. That is, one
device is added to the system at the optimal location for that condition. The next one is added
assuming the first one remains where it was placed.
The basic algorithm for finding a potential site is the same as the one above for finding the initial
site. Once a potential site is found:
1. Define an OpenDSS Recloser object and install a test Recloser at this site
2. Compute the predictive reliability indices for the present configuration
3. Compute the value of the objective function (e.g., customer interruptions saved) for this
location and save it.
4. Move the test Recloser to the next available site in the sequence list on the feeder.
5. Repeat steps 2, 3, and 4 until the end of the sequence list is reached, which signifies there are
no more viable locations.
6. Install the Recloser object permanently by moving it to the location with the highest value of
the objective function.
These steps are repeated for each recloser addition being considered. Reclosers would be added
until there is insufficient value to justify adding another. For example, if a new DA switch cost
$5000 and the value of a customer interruption was $50, the switch would have to save at least
100 customer interruptions to be justifiable.
One advantage of this algorithm is that it is fast. To add 2 Reclosers, there are only 2 passes
through this algorithm. Thus it takes only a matter of seconds even for large circuits like the
IEEE 8500-Node test feeder used to test the algorithm. However, it may not achieve a true
optimal configuration that might be achieved by considering the siting of more than one at a
time. An exhausted search algorithm for siting two at a time was also investigated in this project.
This algorithm required a few minutes to execute for the 8500-Node case.

Placement of Two DA Switches at the Same Time Algorithm

This algorithm was implemented essentially by placing a loop around the one-at-a-time
algorithm above. Every combination of device locations was tested.

10253878 6-4
1. Define an OpenDSS Recloser object and install at the first available branch location past the
substation breaker.
2. Define a second Recloser object and install it at the next eligible section past the first.
3. Move the second Recloser object to each of the remaining eligible sections, computing the
reliability indices and the values of the objective function for each configuration.
4. Move the first Recloser object to the next eligible location.
5. Move the second Recloser object back to the beginning of the feeder.
6. Repeat steps 3, 4, and 5 until all locations have been checked.
7. Install the reclosers to the locations that yielded the highest value of customer interruptions
saved. (If optimizing on SAIFI or SAIDI, it would be the lowest value.)

Required Input Data

• The circuit topology described in terms of bus and branch data required for power flow and
short circuit analysis studies. This would be typical data used in a distribution system
analysis tool.
• For each line and distribution transformer included in the circuit mode:
- Number of faults per year, including temporary faults. For lines and cables, this is usually
stated in faults per mile or some other unit of length
 Length of each line and cable
- Percent of faults that become permanent and result in a sustained interruption
- Time to repair a permanent fault on this branch
• Number of customers at each load point in the circuit model
• Location of existing automated switches, such as reclosers, that can isolate faulted sections.
• Locations of existing fuses and phases on which they are installed.
• Cost of a customer interruption
• Cost value to justify particular types of automated switching devices.

Algorithm Configurations & Settings

Once the circuit model is built for the distribution system analysis tool, and the data describing
the failure rates are established, the key optional setting is whether or not to assume automatic
restoration of downline sections.

Output Reports / Charts

Figure 6-4 and Figure 6-5 show typical plots generated in Excel from the values computed for
the objective function. For the two cases shown, the objective function was the number of
customer interruptions saved, which were valued at $50 each. This value is plotted vs distance in
miles from the substation. The circuit model was the IEEE 8500-Node Test feeder used in the
2014 research effort in this project. Figure 6-3 shows the two locations selected with the 2nd
location highlighted.

10253878 6-5
Figure 6-3
IEEE 8500-Node Test Feeder with Two Recloser Sited. Restoration Assumed.

These two figures show the results for the first two reclosers added sequentially. The first has a
maximum value approaching $60,000 at about 5 mi from the substation, which should be
sufficient to easily justify a recloser installation. The second reaches a value that is
approximately $15,000 at 2.5 – 3.0 mi from the substation.
The red arrows mark the locations of the maximum customer interruptions saved and therefore
the highest value.
There will be similar figures presented in the next chapter for the discussions of the actual test
cases for the 2015 work.
Note that there are multiple values at the same distance from the substation. When there are two
major branches in a feeder as well as minor ones there will often be multiple locations that are
the same distance electrically from the substation.

10253878 6-6
Figure 6-4
Value of Customer Interruptions Plot for Adding First Device

In addition to the objective function value at each location on the feeder, overall system
reliability indices for the present configuration are generated. After the second recloser was
added, the indices are computed as shown in Table 6-1, being computed with the assumption that
downline sections are automatically restored.
The customer interruptions saved represent the difference between the estimate of annual
customer interruptions for the present configuration and a previous configuration – in this case
after adding the first recloser. Adding the second recloser saved an additional 293 customer
interruptions.

10253878 6-7
Figure 6-5
Value of Customer Interruptions Plot for Adding 2nd Device

Table 6-1
System Reliability Indices Computed for 2nd Recloser Configuration

Quantity Computed Value

SAIFI 0.76
SAIDI 1.44 h
Cust Interruptions 900 /yr
Cust Interruptions Saved 293 ($14,650)

Visualizing Reliability Benefits

The reliability impacts of Distribution Automation can be impressive. Just saying that SAIDI
and SAIFI are improved by 40-50% sounds great, but it is hard to really internalize. The figures
below were prepared with the objective of helping to illustrate the value of adding a recloser/DA
switch in an optimal location on a feeder which is enabled with FLISR automated restoration.
In the figures, the thickness of the line is proportional to the accumulated number of Customer
interruptions. Both diagrams are drawn to the same scale. This analysis includes the impact of
the fuses, which are in the locations marked with a red ‘x’.

10253878 6-8
Figure 6-6
Illustrating Number of Customer Interruptions Prior to Adding Recloser/DA Switch

The optimal DA switch location is determined to be very near the first major bifurcation point on
the feeder. As illustrated by the thickness of the line upstream from the DA Switch location, the
number of customer interruptions is cut nearly in half to approximately 60% of the original. A
similar result is expected for most complex feeders of this nature for a single recloser, optimally
cited for saving customer interruptions.

Figure 6-7
Illustrating Number of Customer interruptions After Adding Optionally-Sited Recloser/DA Switch

10253878 6-9
Example 1 –Analysis of a Planning Area
The optimal siting method was applied to a planning are consisting of four major substations and
18 feeders as depicted in Figure 6-8. There are also 18 line reclosers in this area in the locations
shown – some feeders have more than one and some have none.

Figure 6-8
18-Feeder Planning Area with Existing Recloser Locations Shown

The analysis consisted of

1. Calibrating the fault rates to measured data.
2. Adding one recloser to each feeder.
3. Adding two reclosers simultaneously to feeders needing two reclosers.
4. Performing a “Greenfield” siting analysis in which existing reclosers were removed and an
entirely new scheme developed.
Calibration of Fault Rates
While the OpenDSS default fault rates and repair times are reasonable and typical for many
feeders in the US, it is always better to substitute actual values when known. The host utility
supplied actual failure data for 9 of the feeders in the planning area. Table 6-2 shows the number
of permanent faults per mile per year for overhead (OH) lines and underground (UG) cables. It
also shows the mean time-to-repair (MTTR) for OH and UG lines in these feeder in units of
hours.

10253878 6-10
Table 6-2
Actual Permanent Fault Data Provided for 9 of the 18 Feeders in the Planning Area

Faults/mi/year MTTR (h)

Feeder OH UG OH UG
1 0.1617 0.0539 1.75 2.63
2 0.1673 0.0558 2 3
3 0.1845 0.0615 1.1 1.65
4 0.367 0.1223 2.25 3.38
5 0.2152 0.0717 2 3
6 0.0645 0.0215 0.15 0.23
7 0.0283 0.0094 5 7.5
8 0.3478 0.1159 1.65 2.48
9 0.094 0.0313 1.65 2.48

The Line data in OpenDSS were given in ohms/ft so the fault rates in Table 6-2 were converted
to faults per ft per yr and the default fault rates and repair times for each Line element were
replaced with the actual values by a simple OpenDSS Edit command, for example for an
overhead line in Feeder 1:
Edit Line.ohxxxxxxxx Faultrate=0.000153125 repair=1.75

Another critical piece of data for the optimal siting algorithm described in this report is the
number of customers at each load point. These data were provided by the host utility in the
circuit model database. These data fields are often not populated by planners for power flow
analysis but are critical for applying the basic optimal siting algorithm based on customer
interruptions. A typical 120-V load definition in the OpenDSS model, with actual bus names
obscured, would be:
New Load.BSxx_B phases=1 bus1=xxx_yyy.2.0 kV=0.120 kW=24.7 kvar=11.0 NumCust=15

Combining the Faultrate values with the number of customers yields a customer-interruptions
plot for the entire planning area as in the figure below (Figure 6-9). The thickness of each line is
proportional to the number of accumulated customer interruptions and gives a good idea where
the most fruitful locations might be in terms of preventing customer interruptions by adding DA
switch devices. The exiting line recloser locations are marked with the large lime-colored circles.
(Lateral fuses are also included in this model, but are not shown on the diagram.)

10253878 6-11
Figure 6-9
Plot of Accumulated Number of Customer Interruptions for All Feeders

Locating 1 Switch at a time sequentially

One of the 18 feeders in the planning area without any line reclosers will be used to illustrate the
method. This feeder has 2561 customers and its SAIFI based on the calibrated fault rates and
customer distribution is computed to be 0.73. There are no DA switches presently on this feeder.
The diagram below shows the feeder one-line with all fuse location marked. Representing the
fuse locations is very important to the reliability analysis and they were always included in the
calculations even if they are not shown in the circuit diagrams.

10253878 6-12
Figure 6-10
A Sample Feeder showing fused laterals

The optimal siting algorithm was run iteratively on every node to determine the node which
resulted in the most Customer Interruptions avoided (or saved). The graph in the figure below
plots the Customer Interruptions saved vs. the distance from the substation for every node on the
feeder. The most beneficial node resulted in 948 Customer Interruptions saved per year, and
lowered SAIFI from 0.73 to 0.37 and cut the estimated SAIDI in half from 1.4 h to 0.7 h.

Cust Int Saved

1000
900
800
700
600
CI-Seved

500
400
300
200
100
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Distance from Substation

Figure 6-11
Graph of Customer Interruptions Saved vs. Distance from Substation

10253878 6-13
The node that was identified on the graph above corresponds to the following location on the
feeder schematic.

Figure 6-12
Graphical Display of Optimal Recloser Location

If the customer distribution and fault rate were uniform along the feeder, the location would have
divided the number of customers evenly. However, non-uniform failure rates, lengths of line
sections, and customer counts skews the optimal location for saving customer interrupts to a
point farther out on the feeder.

Locating 2nd Switch sequentially

In keeping with the typical practice of sequential addition of DA switches, the algorithm is once
again used to determine the optimal location for a second switch. This second pass incorporates
the first recloser/DA switch previously sited, and finds the optimal location for a second switch
using the same criteria. This second switch is predicted to save an additional 264 Customer
Interruptions per year, totaling 1212 saved, reducing SAIFI further to 0.274 and SAIDI to 0.53 h.

10253878 6-14
 Cust Int Saved
300

250

200
CI-Seved

150

100

50

0
0 0.5 1 1.5 2 2.5 3 3.5 4
Distance from Substation

Figure 6-13
Locating the 2nd Switch - Graph of Customer Interruptions Saved

The feeder one-line diagram is updated to reflect the location of this second switch. While these
locations look relatively close, there is a large number of customers (703) served from the
section between the two automatic sectionalizing devices.

Figure 6-14
Graphical Display of Second Optimal Recloser Location

10253878 6-15
Locating 2 Switches at the same time
One result that was identified early in the analyses of example feeders, was that the optimal
location of 2 switches was better accomplished by evaluating two locations in parallel as
opposed to the sequential example just demonstrated. To confirm this result, the analysis was
conducted in both ways and consistently yielded higher levels of improvement by locating two
switches at a time. Of course, this compounds the number of cases to be evaluated geometrically.
For example, if 1000 nodes were evaluated on a feeder twice to locate two switches that would
result in 2000 iterations. If however, we iterated every combination of two locations on a 1000
node feeder, it would require 1000 x 999 or 999,000 iterations to complete. Thus, it is not trivial
to accomplish this goal, but the benefits are clear, and with modern computing technology, this
amount of iterations is (thankfully) not an issue.
The same feeder was analyzed locating 2 DA switches at a time, generating the result in the
graph below. The graph now requires 3 dimensions to reflect the two locations and the relative
benefits. A total of 432 locations (over 186,000 cases) were evaluated in 65 s using OpenDSS
driven by an Excel VBA program. In this configuration, the program can evaluate over 2800
cases per second. So while the time for executing this calculation is considerably longer than
sequentially adding one recloser at a time, it is not onerous.

Cust-Int Saved

1400

1200

1000

800

600

400 Series205
Series171
200 Series137
Series103
0 Series69
1
13

Series35
25
37
49
61
73
85
97
109

Series1
121
133
145
157
169
181
193

0-200 200-400 400-600 600-800 800-1000 1000-1200 1200-1400

Figure 6-15
Customer Interruptions Saved - 2 DA Switches at a Time

10253878 6-16
The results of this analysis are revealing. The total Customer Interruptions saved increases from
1212 locating 1 switch at a time sequentially to 1282 locating 2 switches at a time. This
corresponds to an improved SAIFI moving from 0.27 to 0.25 and SAIDI improves to 0.48 h.
This may not seem like much on one feeder for one year, but multiplying that number by
hundreds of feeders over a 20 year life cycle, and the benefits of optimal siting are compelling.
The diagram below shows these proposed locations compared to the locations selected by 1 at a
time analysis. The second recloser farthest from the substation is moved a few buses upstream
while the first recloser added in the one-at-a-time analysis is moved more significantly closer to
the head of the feeder, picking up an additional 464 customers in its protective zone.

Figure 6-16
Placement of Two DA Switches (2 at a Time left) vs. (1 at a Time right)

Planning Area Analysis

Using the principles demonstrated in siting a recloser/DA switch on this example feeder, the
analysis was performed on all of the feeders in the planning area.
Add One Recloser to Existing Configuration
Table 6-3 shows the results for adding one more recloser/DA switch to each feeder. The main
column of interest in this table, and the next two tables, is the column in red (“Total CI Saved”).
Assuming that it will take a minimum of 500 CI-saved to justify a new 3-phase device, the
feeders with more than 500 CI-saved are highlighted in bold-face font.
This illustrates the value of this siting method for evaluating a planning area. The planner can
quickly scan a table like this and identify the “low-hanging fruit” where it is most likely the
installation of a new switch will reap dividends in reliability improvement. This is not to say that
other locations may also be worth pursuing, but it will take more careful analysis to determine
the justification.

10253878 6-17
Table 6-3
Reliability Indices Computed for the Feeders - Adding One DA Switch

SAIFI CI Total SAIDI

Feeder SAIFI Existing CI Existing CI Saved SAIDI Existing
a 0.024 0.045 15.9 30.1 14 0.123 0.234
b 0.360 0.675 595.5 1116.8 521 0.647 1.213
c 0.123 0.189 478.1 733.1 255 0.212 0.324
d 0.071 0.180 311.6 790.9 479 0.034 0.172
e 0.227 0.432 277.9 528.9 251 0.503 0.955
f 0.341 0.563 616.6 1017.8 401 0.859 1.449
g 0.373 0.731 990.0 1938.2 948 0.717 1.424
h 0.428 0.821 1846.6 3537.7 1691 0.864 1.655
i 0.258 0.442 553.5 947.4 394 0.312 0.538
j 0.370 0.691 1244.6 2321.9 1077 1.111 2.073
k 0.657 1.486 510.0 1153.1 643 1.972 4.458
l 0.345 0.465 1317.5 1774.2 457 1.035 1.394
m 0.721 1.391 2548.1 4914.7 2367 2.163 4.172
n 0.596 1.375 584.8 1350.2 765 1.787 4.125
o 0.528 1.093 1771.6 3665.4 1894 1.585 3.279
p 0.801 1.685 2166.7 4559.0 2392 2.403 5.056
q 0.375 0.630 1003.6 1688.9 685 1.124 1.891
r 0.650 1.668 2172.0 5573.0 3401 1.950 5.003

Add Reclosers Two-at-a-Time to Existing Configuration

Table 6-4 shows the same analysis except that two reclosers were added simultaneously using
the two-at-at-time algorithm. It will require more justification to install two reclosers. Those
feeders with more than 1000 CI-saved from adding two reclosers are highlighted in bold-face
font.
Table 6-4
Reliability Indices Computed for the Feeders - Adding 2 DA Switches Simultaneously

SAIFI CI Total SAIDI

Feeder SAIFI Existing CI Existing CI Saved SAIDI Existing
a 0.045 0.045 30.1 30.1 0 0.234 0.234
b 0.268 0.675 443.9 1116.8 673 0.481 1.213
c 0.102 0.189 398.0 733.1 335 0.177 0.324
d 0.051 0.180 222.0 790.9 569 0.022 0.172
e 0.227 0.432 277.9 528.9 251 0.503 0.955
f 0.276 0.563 498.6 1017.8 519 0.698 1.449
g 0.248 0.731 656.3 1938.2 1282 0.478 1.424
h 0.298 0.821 1286.7 3537.7 2251 0.602 1.655
i 0.182 0.442 390.0 947.4 557 0.218 0.538
j 0.287 0.691 963.1 2321.9 1359 0.860 2.073
k 0.424 1.486 329.2 1153.1 824 1.273 4.458
l 0.281 0.465 1074.4 1774.2 700 0.844 1.394
m 0.497 1.391 1755.4 4914.7 3159 1.490 4.172
n 0.410 1.375 402.9 1350.2 947 1.231 4.125
o 0.329 1.093 1103.8 3665.4 2562 0.987 3.279
p 0.466 1.685 1260.7 4559.0 3298 1.398 5.056
q 0.288 0.630 770.3 1688.9 919 0.863 1.891
r 0.468 1.668 1564.4 5573.0 4009 1.404 5.003

10253878 6-18
Greenfield Analysis
The preceding two cases dealt with the problem of adding more reclosers/DA switches to an
existing switch configuration. Another interesting exercise to perform with the siting algorithm is
the so-called “Greenfield” siting analysis. That is, assume all existing line reclosers are removed
and the planner has the opportunity to start over, replacing the existing configuration with a new
one.
Table 6-5 shows the results of a Greenfield analysis in which one or two reclosers were added to
the circuits, approximately matching the existing number of devices. In this case, the number of
CI-saved is an indication of the improvement that might be achieved over the existing case. This
value must justify all that is involved in moving or removing an existing recloser and replacing it
with another. This can be quite expensive. Those feeders with CI-Saved values greater than 2000
are highlighted. It is less like that there will be sufficient value in the other feeders in moving or
removing existing reclosers.
Table 6-5
Reliability Indices Computed for the Feeders – Greenfield Analysis

SAIFI CI Total SAIDI

Feeder SAIFI Existing CI Existing CI Saved SAIDI Existing
a 0.045 0.045 30.1 30.1 0 0.234 0.234
b 0.488 0.675 808.1 1116.8 309 0.877 1.213
c 0.169 0.189 656.5 733.1 77 0.291 0.324
d 0.096 0.180 422.9 790.9 368 0.042 0.172
e 0.432 0.432 528.9 528.9 0 0.955 0.955
f 0.540 0.563 975.8 1017.8 42 1.378 1.449
g 0.373 0.731 990.0 1938.2 948 0.717 1.424
h 0.347 0.821 1495.2 3537.7 2043 0.706 1.655
i 0.277 0.442 593.4 947.4 354 0.336 0.538
j 0.464 0.691 1559.0 2321.9 763 1.392 2.073
k 0.657 1.486 510.0 1153.1 643 1.972 4.458
l 0.338 0.465 1290.5 1774.2 484 1.014 1.394
m 0.721 1.391 2548.1 4914.7 2367 2.163 4.172
n 0.596 1.375 584.8 1350.2 765 1.787 4.125
o 0.357 1.093 1196.4 3665.4 2469 1.070 3.279
p 0.493 1.685 1334.3 4559.0 3225 1.480 5.056
q 0.516 0.630 1381.1 1688.9 308 1.547 1.891
r 0.468 1.668 1565.3 5573.0 4008 1.405 5.003

10253878 6-19
10253878
7
ENERGY STORAGE MODELING FOR DISTRIBUTION
PLANNING
Background
Energy storage devices are being actively promoted as the solution to various operational and
reliability problems on power systems mainly due to the rapid growth of variable and uncertain
power sources such as wind and solar generation. Some locations such as the US state of
California and the Canadian province of Ontario have been particularly aggressive in adding
relatively large amounts of storage to the grid (100’s of MW of capacity) to counter the
anticipated problems from these sources of generation. In October 2013, the California Public
Utilities Commission (CPUC) adopted an energy storage procurement framework and
established an energy storage target of 1,325 megawatts for three investor-owned utilities
(PG&E, Southern California Edison (SCE), and SDG&E) by 2020, with installations required no
later than the end of 2024. A certain amount of this will undoubtedly be installed on distribution
systems.
Recent storms in the US such as the occurrence of the historic derecho (straight-line winds) and
Storm Sandy in June and October of 2012, respectively, drew considerable attention to the topic
of resiliency of the grid. Many customers were without electric power for several days and some
for several weeks. This prompted calls for microgrids to be built so that parts of the distribution
system can be operated in islands until power lines damaged by the storms can be repaired to re-
establish grid connection. For example, the Connecticut legislature enacted PA 12-148 which
established a pilot program to provide grants and loans to help develop microgrid infrastructure.
Energy storage is a key part of such efforts.
An interesting aspect of these developments is that while a great deal of the storage will be
installed on the distribution system it will be controlled by the area grid operator for the benefit
of the transmission grid. The distribution system is simply a host for the storage. If the grid
frequency begins to droop due to loss of wind generation, the distributed storage will be called
upon to provide power to counter the downward ramping of the wind turbines. If the frequency
increases above nominal, the storage elements can called up to switch to charging mode and
absorb power up to maximum storage capacity to slow the frequency to normal. This gives time
for more conventional generation sources to re-dispatch to meet the load.
Storage is energy storage, measured in kWh or MWh. Most distribution planning studies are
focused on the capacity to deliver power, measured in kW or MW. Thus, storage naturally adds
the time dimension to the planning problem, which is both useful and challenging at the same
time. One ramification is that static power flow solutions will no longer provide adequate insight
into many of the planning problems that planners will face. Planners must simulate the system
over a reasonable amount of time to get the correct answer when there are time-variant resources
such as solar PV and wind generation.

10253878 7-1
Applications of Storage on Distribution Systems
Some of the applications proposed for storage installed at the Distribution primary (MV) or
secondary (LV) level include:
• Compensating for, or smoothing, solar PV power output ramping. On cloudy days the solar
PV output power can vary quickly, potentially interfering with the operation of other devices
such as capacitors and regulators that are attempting to regulate the voltage.
• Extending the power output from solar PV to meet the early evening peak demand. On many
distribution systems the peak load occurs after the sun has gone down. The idea is to charge
the storage either off peak or while the irradiance is high and then discharge it during the first
few hours of the evening demand peak.
• Support of Transmission grid. This can take the form of compensating for the loss of solar
power at the end of the day to reduce the need to for fast ramping of conventional sources. It
can also be a stabilizing mechanism during periods of high variability of renewable
resources.
• Extend the capacity of an existing distribution substation or feeder. This is typically where
the storage is controlled to discharge to clip the peak demand seen at the substation or feeder
while charging during off-peak hours when the system is not constrained.
• Supporting alternate feeds for temporary reconfiguration.
• Controlling the frequency of a microgrid. Storage devices are controlled to maintain
frequency within acceptable limits by either a built-in droop control or centralized dispatch.
• Increase the available short-circuit current of a microgrid so that it is more capable of
operating distribution system and customer protective devices.
• Reducing the cost of electricity to a given power purchaser by charging off-peak when the
energy is cheaper and discharging to supply load during peak demand periods. Of course, this
requires a substantial price difference between peak and off-peak energy prices that may not
commonly exist at present, but is a potential application for efficient storage in the future.
There are undoubtedly many more potential applications for storage, but this list gives an idea of
why there is much interest in storage on utility power distribution systems.

Distribution Planning Issues Introduced By Energy Storage

Installing energy storage devices on the power distribution system introduces several issues to be
considered by planners. These issue include:
• Overvoltages while discharging. The impact depends on the location and capacity of the
storage devices, and is similar to the impact which is evaluated for hosting capacity analysis
of inverter-based solar PV systems. This could be a particular problem when storage is
dispatched for purposes other than the benefit of the local feeder. The problem would be
similar to have maximum power from solar PV on a bright sunny weekend afternoon when
there is very little load on the distribution system. The difference is that the storage dispatch
could occur at night or any other time.
• Low voltages while charging. To reach the goals of several 100 MW of storage across a
utility service area, one can easily imagine storage devices with aggregate capacities of
several hundred kW to a few MW distributed over a given distribution feeder. This presents

10253878 7-2
 the same kind of challenges that serving a load with similar demand and the capacity of the
system to deliver the required power must be checked using power flow calculations.
• Voltage regulation while compensating for transmission grid support. Regulating the voltage
on the distribution system is likely to be more challenging when the storage devices are being
controlled to support power deficits or surpluses on the transmission grid. The power
produced or consumed may very well have no relevance to the behavior of the load on the
distribution system.
• Interference with overcurrent protection practices. All distributed power sources have the
potential to disrupt long-standing utility practices during fault clearing operations. Since most
storage devices currently being considered have inverter-based interfaces to the utility grid,
short-circuit current contributions are expected to be 110-120% of rated current. With several
devices this is sufficient to disrupt fast tripping/fuse saving coordination. On the other hand,
it is insufficient to operate conventional overcurrent devices such as fuses, reclosers, and
circuit breakers (see next bullet).
• Sufficient short circuit capacity to operate overcurrent protective devices when operating as a
microgrid. This is a common problem with all smaller resources and calls for a different
approach to distribution system protection based more on voltage quantities and impedance
relaying. Even if the protection on the utility-owned distribution grid is modified to
accommodate low-capacity sources, the vast majority of consumers will still have
conventional overcurrent breakers that require a strong short-circuit current to operate. So
planners will have to consider adaptable protection schemes that can adjust based on the
strength of the power source.

Simulation Modes
Distribution system load shapes have historically been quite regular with a daily or weekly cycle.
Many assumptions that planners make regarding such things as thermal ratings of equipment are
based on the expectation these load shapes will continue. Since the mid-1990’s, EPRI
researchers have recognized that it is not possible to get the correct answer for distribution
planning problems involving distributed variable resources unless a series of power flow
solutions covering a significant time period were computed. Sequential-time power flow
capability is now accepted practice in advanced distribution planning methods for analyzing any
resource or new load that significantly alters the typical load shape on the distribution system.
The EPRI OpenDSS tool was designed from its inception in 1997 with sequential-time analysis
capability to handle this aspect of the distribution planning problem with distributed resources.
Several other distribution planning tools in the industry have now incorporated this capability.
It will be necessary to continue to exploit this capability to accurately account for storage in
distribution planning. Storage is not only a variable resource, but it is a limited resource and
planning tools must keep an accurate accounting of the amount of energy stored and available to
supply power to meet demand. Some storage technologies also have a limit on the ramp rate.
Storage not only provides power when called upon by must be recharged from some resource –
usually the same grid – at some time after discharging the stored energy. In addition, the charge-
discharge cycle is lossy, which may play a significant role in the economics and, thus, dictate the
application. The planning problem now is more than simply determining the power-delivery

10253878 7-3
capacity to meet the forecasted peak demand from a bulk power source that can be assumed to be
always available.
EPRI has been studying the planning problem with storage since the beginning of the Smart Grid
Demo initiative. One of the first demo projects was the AEP Community Energy Storage (CES)
demo (2009-2012), which helped researchers understand what capabilities planning tools will
need for distribution planning tasks. The tools must aid the planner in understanding the
interactions and limitations of storage with respect to capacity, reliability, and power quality.
Six basic simulation modes were identified for modeling storage for distribution system analysis:
1. Static: This is the conventional solution of one power flow state with storage device model
manually set to discharging or charging at specified rates, or idling. This mode can be
represented in all commonly-used distribution planning tools at present by defining an
appropriate load or generator. This provides a partial planning picture by simulating limiting
conditions but does not reveal issues that are exposed through time-series simulation.
2. Time: Trigger the storage element at a specified time of day to discharge or charge at a
specified constant level.
3. Peak Shave: Triggers the storage element to discharge when the load measured at a selected
control location exceeds a specified peak load value and attempts to produce sufficient power
to limit the net load current or power to the peak load value. Charging is performed
separately at a scheduled time.
4. Load Following: Similar to peak shave mode except that the storage element is triggered to
discharge at a specified time predicted through a short-term load forecast when it will be
necessary to offset load demand. Then the storage controller attempts to produce sufficient
power to limit the net load to the value at the time of triggering.
5. Loadshape: The storage charge and discharge cycle is determined by a predefined shape.
The shape would be computed using planning criteria of interest to test planning
assumptions, size storage elements, etc. This capability gives the planner the flexibility to
investigate many different scenarios without requiring hard-coded computer algorithms to be
implemented in the planning tools. This is important to accommodate new types of storage
technologies as they evolve.
6. Dynamics: This is an advanced mode of analysis for modeling fast-changing phenomena
such as frequency control on microgrids. It is not a type of analysis commonly performed in
distribution planning, but will be important for some specialized applications and research
studies.
EPRI has implemented examples of these simulation modes in its main distribution system
analysis research tool, the open-source OpenDSS software available on the internet. Interested
researchers and software developers may examine the code to get an idea how these modes
might be implemented. Modes 2-5 are designed for sequential-time power flow simulations with
a time step of typically 15 minutes to 1 h for common power and energy capacity evaluations. To
study using storage to compensate for rapidly-varying power sources such as solar PV
generation, a time step as small as 1s is common. Much more information is now available on
this subject due to research sponsored by other organizations such as the US DOE. An internet
search on “Quasi-static Time Series,” or QSTS, will reveal additional up-to-date resources on
this topic.

10253878 7-4
Dynamics mode, mode 6, is often executed in time steps ranging from 0.2 to 1.0 ms. This mode
is needed to study such topics as microgrid frequency control and behavior during faults and
other disturbances. The Dynamics storage modeling capability in OpenDSS has been
demonstrated by implementing a storage model developed by EDF R&D, a description of which
is contained in the 2012 EPRI report 1024285. [12] Additional papers have been written on the
subject: references [13] and [14]. Dynamics simulations require modeling skills that are, at
present, not commonly utilized in the distribution planning process. The analysis is generally
performed on an exception basis, often by consulting engineering firms with the necessary
expertise.
The OpenDSS STORAGE element model is a generic, technology-independent model intended
to be suitable for planning studies (Figure 7-1). It is not intended to be a specific model of any
particular technology, but it should suffice for most planning studies involving one or more
storage devices on a distribution system. A DLL interface is provided for those cases where it
becomes necessary to model a specific storage technology in fine detail. A skilled programmer
would be required to create the DLL. This is addressed later in this chapter during the discussion
of Dynamics mode simulation.

Idle | Discharge | Charge % Eff. Charge/Discharge

Other Key
kW, kvar Properties
% Reserve
kWhRated
kWhStored
%Stored
Idling Losses kWh STORED kWRated
etc.

Figure 7-1
Basic Concept of the EPRI OpenDSS STORAGE Model

Static Analyses
The basic distribution planning analysis function is a static power flow. This is the capability to
solve for the power flow with the storage device set to either charging or discharging at specified
power values. This allows assessments of the limiting values for basic issues such as voltage
rise/drop and current-carrying capacities during operation. However, it does not permit
assessment of the energy storage requirement unless very simple charge and discharge shapes are
assumed. The time element is not considered.

10253878 7-5
The distribution planner will be interested in studying scenarios that could conflict with normal
operation. Two obvious screening scenarios are:
1. At minimum load, study maximum power output from the storage units to evaluate the
potential for overvoltages during such conditions. One case where this might happen is
dispatching storage late at night to support a transmission system need.
2. At maximum load, study the condition where the storage units are dispatched to charging
mode to investigate potential undervoltages and thermal overloads. The weakest bulk power
system source if often used for this analysis depending on individual planning philosophies.
These are two relatively simple screens that can be accomplished with existing distribution
system analysis tools. If the storage configuration being proposed passes both tests with a
satisfactory margin, the storage can likely be accommodated without interfering with the
operation of the distribution system. The screens are not comprehensive, however, and other
issues may show up during sequential-time or dynamic simulations.
Another important static analysis is a short-circuit study. This requires some model of
contribution of the inverter-based resources on the system to short-circuit faults. The analysis
would be quite similar to that required for modeling solar PV inverter contributions. When
inverter-bases resources were first considered, it was common to assume the current contribution
into a fault was a conservatively-high value of 2.0 times rated current. More recent data,
including research done by EPRI, has shown that 1.2 times rated current is more realistic. If the
voltage sag is minor, most inverters may be assumed to continue to produce rated current.
Static analysis will be adequate for many storage applications on distribution systems for the
near term. When storage devices become more prevalent or great in capacity relative to the
strength of the distribution system, some sort of sequential-time simulation will be required to
correctly represent the operation of the storage element and the state of stored energy. Examples
are described in the following sections.

Capacity Evaluations
Figure 7-2 shows a result from a simulation performed during EPRI’s Community Energy
Storage (CES) Smart Grid Demo project with AEP. [15] This simulation was designed to study
the feasibility of using a number of 25 kW, 25 kWh distributed battery storage units to shave the
peak substation demand load each day using energy stored in off-peak hours. The simulation was
performed using actual 15-minute feeder demand data.

10253878 7-6
 Load Following with Storage, 25 kWh Storage Units
Discharge Trigger @ 1300, 30% charge rate @ 0200
10000 30

25
8000

20
6000

kWh
15
kW

4000
10

2000
No Storage 5
Storage
kWh Stored
0 0
170 180 190 200 210 220 230
Hours into Simulation

Figure 7-2
Using storage for daily peak shaving

Just having the storage available is not sufficient to guarantee that it will be useful for capacity
needs. Like many variable resources, it must be available when it is needed to gain capacity
credit. Storage is a limited resource. When the limited storage resource is used to clip the daily
peak, the timing of the daily peak must be predicted accurately. In the “Load Following” control
mode shown, the storage controller attempts to keep the feeder demand at, or below, the demand
level at the time the storage discharge is triggered ON. If the storage element is triggered to
discharge too quickly, the storage is depleted prior to the peak and not all the benefit from
shaving the peak power on a feeder or substation transformer can be realized. This occurs on the
first of the two days shown in Figure 7-2. The dashed curve shows the state of charge in the
battery as it is discharged and charged over two peak-load days. The timing of dispatching the
storage on the second day (the larger peak) yields a more successful result. This finding would
not have been as easy to determine from simple static simulations.
Notice the difference in the sizes of the areas between “No Storage” curve and the “Storage”
curve during the charge and discharge cycles. The storage model used for this analysis represents
the losses in the storage element whether charging, discharging, or idling. Losses must be
represented in a simulation that captures the time-dependent nature of the problem. In this case,
there is approximately a 20 % difference in the areas between the curves because the model
assumed 10% losses on both charging and discharging. This is visibly noticeable in the relative
sizes of the areas between the curves. The model also assumes a 1% loss when idling. This is
power required to either heat or cool a battery or compensate for losses due to friction, windage,
etc. in rotating storage technologies. Such a small loss does not seem like much, but it can add up
over a year when the storage device is idling most of the time. This can affect the economics
significantly and must be accounted for in some way during the planning process.
In the OpenDSS implementation of this simulation, each STORAGE element takes care of
computing its own losses. An energy meter placed at the head of the feeder records the net effect
as the simulation proceeds.

10253878 7-7
The discharge and charge states of all storage units in this problem would be determined by a
supervisory controller. There is insufficient local intelligence available for each STORAGE
element to determine what its state should be while interconnected with the grid. Modeling
controllers presents a new challenge to distribution planners because controller models can have
quite complicated algorithms. In this example, discharging is started by a simple time control
trigger at 13:00 each day. Then the controller manages the power dispatch of each of the
distributed storage units to maintain an approximately constant demand at the head of the feeder
where the controller is located. A relatively simple deadband controller with a 2 % band around
the target value was used in this example.
Deadband controllers generally work well on power systems and are also generally well-behaved
during simulations. Discharging continues until either the feeder demand drops below the target
or the storage elements have reached their minimum allowable kWh storage level. For the CES
design modeled in this simulation, reserve energy levels of 20% to 50% were investigated during
peak shaving applications in case it is needed to cover outage contingencies. This is another
design decision a distribution planner must make. In the OpenDSS implementation, the storage
device models manage their own storage levels and simply cease to respond to the controller
when the storage is depleted to the reserve level or is fully charged.
Charging is assumed to begin at 02:00 in this example and proceed at a rate of 30% of the power
rating of the storage element until the storage unit is fully charged. In this example, it takes about
3 h to recharge from a 20% reserve level to full capacity.
The OpenDSS program uses two separate models to represent storage for this kind of simulation:
1. A STORAGE element representing the device that stores energy, and
2. A STORAGECONTROLLER element that contains the algorithms for the six solution
modes mentioned earlier and controls one or more STORAGE elements.
This is depicted in Figure 7-3. The STORAGE element models the behavior of the storage
medium and keeps track of the energy storage level, losses, etc. For the power flow solution, it
acts as either a generator or a load depending on whether it is discharging or charging. That is,
the power output of the STORAGE element can be either positive (+) or negative (–) or nearly
zero (there is always some power consumption when idling).
The STORAGECONTROLLER element neither produces nor consumes power but simply
monitors a location in the active circuit and send messages to selected STORAGE elements to
get the desired power from the fleet or recharge the fleet at a suitable rate for system conditions.
If the controller is monitoring a solar PV installation, the charging rate can, for example, be set to
a portion of the PV output (see next section).
Separating the functionalities in the distribution planning tool in this manner simplifies the
implementation in computer code. Nevertheless, the STORAGE-STORAGECONTROLLER
combination is currently the most complicated model in OpenDSS as of this writing. This fact,
along with some of the complexities of the Dynamics mode model, suggests that acceptance of
sequential-time simulation of storage into distribution planning may proceed slowly until
distribution system analysis tool vendors can provide suitable interfaces and the industry gains
more experience with storage.

10253878 7-8
 Storage
Controller

Comm Link
V, I

Substation
Storage “Fleet”

Figure 7-3
OpenDSS Storage Controller Concept

Compensating for Renewable Generation

Storage elements may be used for a variety of purposes on distribution systems. One commonly-
cited example is to compensate for variable renewable generation. Example results are shown in
Figure 7-4 for a 2-MWh storage device simulated at a one-minute resolution. [16] The storage
controller is programmed to charge during the peak solar PV output and then discharge at the
load peak, which lags the solar output by a few hours. This helps solve a common capacity
problem faced by distribution planners: solar generation output frequently lacks about 2 h of
meeting the evening peak load on residential feeders. This is frequently the peak load for which
distribution planners design the distribution system to deliver. Having useful storage could
potentially delay capacity upgrades and allow more efficient operation of the system. Having the
sequential-time simulation capability to analyze this problem will help planners better evaluate
the true impact of such a resource on system capacity.
8000 2000
Demand_wPV+Storage
7000
Demand_wPV 1500
6000 Storage Ouput
PV and Sotrage Output (kW)

5000 PV Output
Demand (kW)

1000
4000

3000 500

2000
Charging 0
1000
Discharging
0 -500
1 Minutes 1441

Figure 7-4
Using storage to shift PV generation

In the scenario shown in Figure 7-4, a Time-based control (simulated with Mode 2) is used to
shift the energy output from the PV to higher demand periods. While this control may be simple
and cost effective to implement, it is not as effective as the Peak Shave and Load Following
methods. Without direct observation, the charge/discharge durations of the time-based method

10253878 7-9
must be of sufficient length to capture the expected variations in both generation and peak
demand. Consequently, the discharging/charging rate tends to be shallower than may be achieved
via other control methods. Furthermore, a simple time-based dispatch method performs a full
discharge/charge cycle each day regardless of the energy generated or consumed.
It is important for the distribution planning tool to provide the ability to evaluate the performance
of different storage control options while accounting for interaction with renewable sources. In
this example, Time-based control of the storage was shown to reduce the effective energy
supplied by the PV by 3.3 % due to the operational losses incurred by the storage unit each day.
In contrast, a possible Peak Shave control (Mode 3) was shown to decrease the effective PV
generation by only 0.44 % while further reducing the peak demand by 740 kW beyond that
achieved with the Time-based control.
Another storage control function commonly considered when pairing with renewable generation
is smoothing the fluctuations in the variable generation. These fluctuations can result in
significant voltage regulation problems on distribution systems. In fact, this is often the most
significant problem limiting the PV hosting capacity of distribution systems. It is very important
to have the sequential-time power flow capability to expose potential voltage regulation issues.
The net PV output with and without the matched storage is illustrated in Figure 7-5 for one
potential control algorithm. In this case, the energy storage is dispatched based on a moving
average target for the net output of both PV and storage.

Figure 7-5
Smoothing variations in PV generation

The simulation was performed in OpenDSS using the Loadshape mode. A Loadshape object in
OpenDSS is a per-unit curve describing the variation of a quantity over time. The loadshape was
derived separately using the target value while also taking into account operational constraints
including the rated kWh and kW associated with the modeled storage. The loadshape derived to
emulate the smoothing operation is plotted in Figure 7-6.

10253878 7-10
Note that this operation requires the storage unit to alternate frequently between discharging and
charging while the PV is producing varying amounts of power due to the cloud transients. The
storage unit is charged to half its rated capacity at 02:00 while the PV is not producing. This
allows the storage unit to absorb or inject power as needed when the PV system begins to
produce power. The charging cycle is clearly visible in both Figure 7-5 and Figure 7-6. Given the
variable output of the PV and the nature of smoothing functions, the amount of energy required
to “top off” the storage device each evening is a direct function of the losses incurred during the
previous day’s operation. In this simulation, the losses incurred by the smoothing operation
decreased the overall generation from the PV and storage by 1.8 %.
1500

1000

500
ES Output (kW)

0
0 1440 2880 4320 5760 7200 8640
-500

-1000

-1500
Minutes

Figure 7-6
Power output smoothing operation

This illustrates the level of detail that could go into a planning study for representing schemes to
smooth PV output. The Loadshape mode is quite useful for analyzing this because the planner
needs flexible power shape modeling capability for this kind of problem.

Dynamics Simulations
Most of the other time-varying simulation modes identified here can be executed by supplying
the load and storage power charging/discharging characteristic using power data in time steps
ranging from a few seconds to one hour. Then the analysis tool must simply perform a series of
power flow solutions, which is generally an extension of the single snapshot power flow solution
supported by nearly all distribution system analysis tools. This assumes the time step size is
longer than the transients within the storage element or control.
However, to study the interaction of inverter-based storage devices during disturbances or for
such things as frequency and stability control on microgrids requires dynamics, or
electromechanical transients, analysis capability to simulate storage in very small times steps of
1 ms, or less. The development of models suitable for distribution planning is still very much in
the embryonic stage and continues to be an active research topic at EPRI.
The simple Dynamics mode originally implemented within the OpenDSS software was intended
to model rotating machine dynamics for islanding analysis sufficient to evaluate a distributed
generation (DG) interconnection against the frequency and voltage requirements of IEEE Std
1547™. It is the only mode of OpenDSS that implements the solution of differential equations.
For rotating machines, the differential equations represent the so-called “swing equation” for
each machine.

10253878 7-11
In 2012, the Dynamics mode was adapted to implement a model of a storage device developed
by EDF R&D called DESS (for Distributed Energy Storage System). EDF R&D developed a
dynamic model of the DESS for the electromechanical transient simulation software Eurostag, a
positive-sequence modeling program for transmission systems. OpenDSS uses a multiphase
model of the power system as do most distribution system analysis programs. The OpenDSS
version of the DESS model was implemented by coding the differential equations for several
proportional-integral (PI) control loops into a special dynamic-linked library (DLL) that can be
linked to the OpenDSS STORAGE element “DynaDLL” model.
Figure 7-7 illustrates the results of a dynamics simulation using the OpenDSS implementation of
the EDF DESS. It is a 6-s simulation performed at a fixed 0.2-ms time step, or 5000 steps per
second. This is the type of simulation that would be needed to represent using storage to help
control frequency in a microgrid. Time constants in the inverter controls are frequently on the
order of a few ms requiring simulation steps less than 1.0 ms to accurately represent the
transients. Simulated time durations are much shorter than with the other two storage simulation
examples described herein due to the prodigious volume of results data that is produced.
However, it can be quite important to determining the stability control algorithms for a
microgrid.

Figure 7-7
6-second DESS simulation

While sequential-time power flow modes often execute in faster than real time, Dynamics-mode
solutions are often slower than real time on typical desktop computers due to the tiny time steps.
This is one reason why some researchers performing storage simulations employ special real-
time simulators built for that purpose. This work is mostly done at universities and the national
laboratories. A few large utilities have acquired such computer tools expressly for performing
simulations related to renewable generation and storage. However, distribution planners
performing off-line planning simulations are not likely to have access to real time simulation
capability to design their systems, at least in the near term.

10253878 7-12
The example illustrated in Figure 7-7 assumes the existence of a master controller that sends out
the power reference signal, P reference , based on frequency stability requirements. If the frequency
droops too low, the controller will ask for more power. If the frequency increases too much, the
controller will ask the DESS to absorb power from the grid by charging the storage medium. The
DESS output (blue line) lags a few ms behind the P reference signal (red line). In this case, the
power levels for the discharge and charge cycles are identical in magnitude and duration. Thus,
the kWh stored level indicated by the dashed line is gradually decreasing due to losses incurred
for each cycle – another complication of controlling storage elements. A more intelligent master
controller would periodically extend the charge cycle to ensure the storage level remains
sufficiently high.
On the Complexity of Dynamics Simulation Models with Respect to Distribution
Planning

Figure 7-8
Layout of an ultracapacitor-based Distributed Energy Storage System (EDF)

Figure 7-8 shows a simplified block diagram of the EDF DESS. This device uses an
ultracapacitor instead of a battery, flywheel, or some other storage technology. Keeping track of
the state of charge and the charging losses as the voltage on the capacitor varies is challenging. It
is put forward here as an example of the complexity of the control of storage devices should
distribution engineers wish to represent one in detail.

10253878 7-13
To define the DESS model, the user must supply 31 constants in addition to defining how the
various control signals connect to the system. Control systems for other types of storage devices
are expected to have similar complexity for models used in dynamics analysis. Given that
distribution engineers today have difficulty obtaining a handful of values for common rotating
machines for models of conventional generators, distribution planners are likely to find making
detailed models of devices like the DESS a discouraging task.
One solution often proposed is for manufacturers of storage devices to supply the models (see
below). However, there is much work to be done by the power industry in general before this
will become commonplace.

Electromagnetic Transients Simulation

Using electromagnetic transients (i.e., time-domain) simulation for this problem takes the
modeling complexity to a higher level. Not only must the control loops in the controller be
represented, but the switching of the inverter and the detail of the storage medium must be
modeled. Those performing real-time simulations are currently doing this. Typical time step
sizes for this simulation would be on the order of 10-50 µs. This limits the size of problem that
can be executed in a reasonable amount of time and may not be practical for distribution planners
accustomed to making models of 5000-10,000 nodes.
This is expected to have limited application in the utility distribution planning process for the
foreseeable future unless there is a technological breakthrough in terms of modeling ease.
Electromagnetic transient analysis of the energy storage problem is expected to be performed
primarily by industry consulting engineers, academic researchers, and equipment vendors.

Vendor-Supplied Model Interfaces for Different Simulation Modes

The modeling complexities described above for dynamics and electromagnetic transients
analysis are likely too onerous to be practical to include in the typical distribution planning
process. There is insufficient time in the planning process to build and calibrate such models. A
commonly-proposed solution to this problem is for manufacturers of energy storage systems to
supply the models along with the equipment. This seems like a good idea, but is not necessarily
very easy to implement.
Among the issues that must be dealt with are:
• The establishment of a common software interface between the distribution system analysis
packages and the vendor-supplied models.
• The interface would have variants for quasi-static power flow, dynamics, and
electromagnetic transients; each can have different modeling requirements. Standard-setting
working groups would have to be established to develop the interface specifications.
• The interface would have to be implementable on various computer platforms such as
Windows, Linux, OS/X, and possibly handheld platforms such as Android and iOS.
• Distribution system analysis vendors will have to implement a means of supporting the
interface in their software product.
• Storage system suppliers will generally want some means of protecting the proprietary design
of their product. This will generally rule out open-source models and require compiled

10253878 7-14
 libraries (DLL, SO, etc). This is not a fool-proof method of protecting the intellectual
property due to the existence of software for decompiling such libraries.
So there are quite a few development steps to be taken before this idea can become a reality.

Conclusions About Storage Modeling

Sequential-time power flow simulation is a relatively simple extension of the typical static power
flow solution in common distribution system analysis tools. Each of the sequential-time
simulation modes presented in this chapter requires more sophisticated models and more data
than a simple static power flow evaluation with which most distribution planners are familiar.
Dynamics models of inverter-based storage may require values of more than 30 parameters. This
is daunting for distribution planners. Some form of standard model framework for vendor-
supplied models must be developed to make this task easier and more attractive for distribution
system analysts. Substantial model development work for distribution system analysis tools
remains.
Lacking these advancements, the distribution planner will be forced into a default position of
building systems with greater power delivery capacity to handle energy storage wherever it
appears. A similar problem exists with hosting widespread solar PV generation. If the proposed
system fails either of the two simple tests proposed at the beginning of this chapter, the planner
has the choice of either reinforcing the system so that it can accommodate the proposed system
or declining the interconnect it. Like DG interconnections, this would move the proposal from a
fast-track interconnection process to one requiring detailed studies to determine interconnection
requirements.
Another area of concern among distribution planners is related to reliance on high-speed
communications for controlling microgrids, multiple storage devices, and other fast-acting
distributed resources. It is a legitimate question as to whether or not the latency of the
communications network will be a major roadblock to implementing many of the ideas for
storage controllers – especially those applications requiring high-speed response to frequency
changes and power ramping. The capability of a distribution planning tool for this would have to
include some model – perhaps simplified – of the ICT network as well as the power network
with its controllers. Work on this has only just begun and it is still very unclear as of this writing
what will be a suitable form of this capability for distribution planners.

10253878 7-15
10253878
8
PLANNING FOR HARMONICS
Introduction
A good assumption for most utilities in the United States is that the sine-wave voltage generated
in central power stations is very good. In most areas, the voltage found on transmission systems
typically has much less than 1.0 percent distortion. However, the distortion increases closer to
the load. At some loads, the current waveform barely resembles a sine wave. Electronic power
converters can chop the current into seemingly arbitrary waveforms.
While there are a few cases where the distortion is random, most distortion is periodic, or an
integer multiple of the power system fundamental frequency. That is, the current waveform is
nearly the same cycle after cycle, changing very slowly, if at all. This has given rise to the
widespread use of the term harmonics to describe distortion of the waveform.
To some, harmonic distortion is still the most significant power quality problem. It is not hard to
understand how a distribution engineer faced with a difficult harmonics problem can come to
hold that opinion. Harmonics problems counter many of the conventional rules of power system
design and operation that consider only the fundamental frequency.
Therefore, the distribution planner is faced with unfamiliar phenomena that require unfamiliar
tools to analyze and unfamiliar equipment to solve.
Although harmonic problems can be difficult, they are not actually very numerous on utility
distribution systems. Only a few percent of utility distribution feeders in the United States have a
sufficiently severe harmonics problem to require intervention. In contrast, voltage sags and
interruptions are nearly universal to every feeder and represent the most numerous and
significant power quality deviations.
The utilization sector suffers more from harmonic problems than does the utility power delivery
sector. Industrial users with adjustable-speed, or variable-frequency, drives (ASDs or VFDs), arc
furnaces, induction furnaces, and the like are much more susceptible to problems stemming from
harmonic distortion.
Harmonic distortion is not a new phenomenon on power systems. Concern over distortion has
varied during the history of ac electric power systems. A problem would arise – usually with the
introduction of new technology – and then a solution would be implemented by the industry and
the concern would wane for a few years. In the 1930s and 1940s, there were many articles on
power system harmonics. At that time the primary sources of harmonic distortion were the
various transformers and the primary problem was inductive interference with open-wire
telephone systems. The forerunners of modern arc lighting were also being introduced and were
causing quite a stir because of their harmonic content—not unlike the stir caused by electronic
power converters in more recent times.

10253878 8-1
Fortunately, if the system is properly sized to handle the power demands of the load, there is a
low probability that harmonic currents will cause a problem with the power distribution system.
They may still cause problems with low power circuits such as telecommunications and
computer systems in the vicinity of the power system.
Harmonic problems in the electric power system arise most frequently when the capacitance in
the system results in resonance at a critical harmonic frequency. This dramatically increases the
distortion above normal amounts.
While these problems occur on utility power distribution systems, the most severe cases are
usually found in industrial power systems because of the higher degree of resonance achieved.
There is proportionately less resistance at the point of connection to power factor correction
capacitors and the “Q” of the resonant circuit is much greater than is typically found on the
distribution system.
The standard that governs harmonic distortion on the power system is IEEE Std 519-2014 - IEEE
Recommended Practice and Requirements for Harmonic Control in Electric Power Systems. The
standard breaks the problem into two parts:
• Limiting the amount of harmonic current that a load may inject into the power system.
• Limiting the voltage distortion on the power system.
Distribution planners usually do not have much control over the harmonic currents a consumer
may inject into the distribution system other than including references to IEEE Std 519-2014
limits in the interconnection agreement. It is always a good idea for planners to investigate the
types of loads that commercial and industrial customers are proposing to connect to the system
even if it is not explicitly the planner’s responsibility.
Distribution planners would have more control over the harmonic voltage distortion that results
from the harmonic currents originating in consumers’ load equipment. Part of that control comes
from designing the system with sufficient capacity to minimize the impact of the harmonic
currents. However, in North America most utilities apply capacitor banks to reduce losses and
increase power delivery capacity. The introduction of capacitance into a mostly inductive circuit
will always result in a resonance somewhere. The challenge to the distribution planner is to keep
the system out of resonance at a harmonic frequency present in the load currents.
This chapter deals with the essential harmonic analysis a distribution planner should perform to
complement capacity planning studies.

Harmonics Fundamentals
The ideal ac power system service voltage is a perfect sine wave – of one frequency – that has
constant magnitude and frequency. The sine-wave voltage measured at the central power station
generators for most utilities in the United States is very nearly ideal. As the power flows from the
generator to the load the voltage gets distorted due to serving loads that produce harmonic
currents. The voltage found on transmission systems is less than 1% distorted. Some areas have
higher distortion, but it seldom exceeds 2% on the transmission grid. The distortion increases as
the power flows through the distribution system and low-voltage system on its way to the load.
At some load points, the current waveform barely resembles a sine wave.

10253878 8-2
Harmonic distortion is caused by nonlinear devices in the power system. A nonlinear device is
one in which the current magnitude and shape is not proportional to the magnitude and shape of
the applied voltage. This is the source of the most significant harmonic distortion in power
system voltages and currents.
The theory of how a distorted waveform is decomposed into harmonics of the fundamentals is
well documented and is beyond the scope of this document. The reader is referred to standard
textbooks on power system harmonics. Here, we will assume the existence of harmonics and
focus on how a distribution planner must deal with them
As Figure 8-1 shows, voltage distortion is the result of distorted currents passing through the
linear, series impedance of the power delivery system, although, assuming that the source bus is
ultimately a pure sinusoid, there is a nonlinear load that draws a distorted current. The harmonic
currents passing through the impedance of the system cause a voltage drop for each harmonic.
This results in voltage harmonics appearing at the load bus. The amount of voltage distortion
depends on the impedance and the current. Assuming the load bus distortion stays within
reasonable limits (e.g., less than 5 percent), the amount of harmonic current produced by the load
is nearly constant.
While the load current harmonics ultimately cause the voltage distortion, it should be noted that
load has no control over the voltage distortion.
Ideal Voltage Distorted Voltage

+ Voltage Drop -

Distorting Loads
Distorted Load
Currents

Figure 8-1
Voltage Distortion is Due to Distorted Current Passing Through the System Impedance

Sources of Current Distortion

Nonlinear elements on electric power systems that are well-known for producing distorted
currents include:
1. Arcing devices such as arc furnaces, welders, and arc lighting.
2. Electronic power converters such as rectifiers, adjustable-speed motor drives, and induction
furnaces.
3. Ferromagnetic devices such transformers.
Ferromagnetic devices were quite important harmonic current producers early in the history of
US power systems. Transformers produced a significant amount of 3rd and 9th harmonic currents
that were notorious for producing telephone interference when the predominant type of telephone

10253878 8-3
line was an open-wire construction. Today, telephone interference is much less of a problem and
other sources of harmonic currents swamp out the transformer contribution. Transformers
produce harmonic currents at a level of approximately 1% of the rated current of the transformer.
This is still significant because there are so many transformers. A typical 10-MVA distribution
feeder in the US may have over 20 MVA of distribution service transformers connected, which
would be roughly the equivalent of 200 kVA of distorting power.
Historically, the next class of harmonic-producing load to cause problems on the power system
was arc lighting. In particular, the application of sodium-vapor arc lighting for street lighting and
other area lighting application resulted in a new flurry of papers in the AIEE Transactions on
harmonics in power distribution systems. [20] Fluorescent lighting has similar harmonic content
in the current as do arc furnaces. Of course, arc furnaces are large loads with quite volatile
current characteristics and continue to be difficult to handle even today. Arcing loads produce
harmonic current magnitudes on the order of 20-30% or rated current – considerably more than
what transformers produce.
Power distribution engineers have learned to cope with harmonics from ferromagnetic and arcing
devices. Then came electronic power converters with the potential for nearly an order of
magnitude higher magnitudes of harmonic currents.
Electronic power converters can chop the current into seemingly arbitrary waveforms. Figure 8-2
shows the typical “rabbit ear” current waveform from a 3-phase variable-frequency drive, which
will have a schematic circuit diagram similar to Figure 8-3. The “ears” come from current pulses
for charging the dc bus capacitor. The magnitude and duration of the pulses will depend on the
equivalent source impedance. When the impedance is low, distortion can exceed 100%. One of
the difficult issues with serving this type of load is that the displacement power factor – the
fundamental frequency power factor – is near unity while the true power factor including all
harmonics is very low. This requires the supply system to be overbuilt with larger conductors
and transformers to support this kind of load. Power factor correction capacitors will often not
help because they act on the fundamental frequency power factor, which is already near unity. In
fact, if the application of capacitors results in harmonic resonance, capacitors will exacerbate the
problem.
When such power converters were first installed in large numbers in the late 1970s, utility
engineers became quite concerned about the ability of the power system to accommodate the
harmonic distortion. Many dire predictions were made about the fate of power systems if these
devices were permitted to exist. These concerns have proven to be somewhat overstated and the
power delivery system has proven remarkably robust in the presence of this distortion. A
statement that is frequently made about this observation is:
Without resonance, if the power system is built with sufficient kVA capacity to
supply the load kVA demand, the harmonic voltage distortion will generally be in
the acceptable range.
This is part of the basis for the harmonic limits stated in IEEE Std. 519-2014. The current limits
were established assuming normal power delivery system capacity. Exceptions to this statement
are where the system capacity for serving the load is marginal. Then one can get excessively
distorted voltages even without resonance.

10253878 8-4
Figure 8-2
Current waveform from a variable-frequency drive and its corresponding harmonic spectrum.

Figure 8-3
Schematic of a 3-phase pulse-width modulated variable-frequency drive. [18]

While the extent of the harmonics problem is not as great as many power engineers had feared,
all those who now earn their living solving power quality problems owe a great debt of gratitude
to those engineers. Their concern over this “new” problem of harmonics sparked the research in
the late 1980’s that eventually led to the great increase of knowledge about all aspects of power
quality that we now take for granted.
To some, harmonic distortion is still the most significant power quality problem. This is not hard
to understand. Finding a solution to a difficult harmonics problem can seem like chasing ghosts
and gremlins. Some days it works and then something fails. A fix that works in one location will

10253878 8-5
not work in another. The primary planning tool of distribution engineers is power flow analysis.
Harmonics problems counter many of the conventional rules of power flow analysis and other
planning analysis that consider only the fundamental frequency. Therefore, the engineer is faced
with solving an unfamiliar problem with unfamiliar software tools and unfamiliar mitigation
equipment to install and maintain.
To perform network analysis at harmonic frequencies, most power system harmonic analysis
computer software tools take the approach illustrated in Figure 8-4. The linear impedances of the
system are adjusted for frequency and the nonlinear, harmonic-producing loads are nominally
replaced by a current source. Then a separate solution is performed for each frequency of
interest. The magnitude and phase angles of the harmonic current source is determined from a
fundamental frequency power flow solution. A measured, or assumed, harmonic spectrum is
used to determine the magnitude of the current source at harmonic frequencies. Some tools
assume that the current source is constant while others iterate and adjust the current for the
voltage distortion. Either approach is generally acceptable for planning purposes.

System One-Line Diagram

Model

Figure 8-4
Replace the harmonic-producing device with a current source in the model.

The concept of replacing the nonlinear loads with a current source will work in most cases for
computing the harmonic flows in a power distribution system, but not all. As we shall see, the
exceptions involve resonance.
Capabilities of an adequate harmonic analysis tool for distribution planning include:
• Solve at frequencies other than fundamental power frequency.
• Perform a solution at individual harmonics or interharmonics with multiple harmonic sources
simultaneously to estimate total harmonic distortion.
• Perform a frequency-response scan at small frequency interval (e.g., 5 Hz) to identify
resonances.
• Produce graphs and tables of multi-frequency analyses.
• Automatically adjust the impedances of lines, transformers, capacitors, etc. frequency.
• Adjust phase angle of current source by the base power flow and by frequency.
• Model all transformer connections because different connections have different responses to
each harmonic.

10253878 8-6
• Multi-phase coupled lines. This is important where there are multiple circuits sharing the
same right-of-way and the same neutral conductor. Triplen harmonic currents tend to flow in
the neutrals and can couple different circuits.
• Frequency-dependence of line and transformer impedances.
• Models for filters.
• Can model large networks of several hundred nodes
• Allows both current-source and voltage-source models of harmonic sources.
Rms and Power Values
Much of distribution planning revolves around accounting for the power flow through the circuit
elements. It is important to understand how power is computed in the presence of harmonic
distortion.
There are three standard quantities associated with power:
1. Apparent power S [voltampere (VA)]. The product of the rms voltage and current.
2. Active power P [watt (W)]. The average rate of delivery of energy.
3. Reactive power Q [var (var)]. The portion of the apparent power that is out of phase, or in
quadrature, with the active power.
The apparent power S applies to both sinusoidal and nonsinusoidal conditions. The apparent
power can be written as follows:

For the sinusoidal condition, P resolves to the familiar form,

Where θ 1 is the phase angle between the fundamental-frequency voltage and current.
For the distorted, or non-sinusoidal, condition, the rms quantities are computed by the square-
root of the sum of the squares of the individual harmonic components:

10253878 8-7
Determining Capacity with Distorted Currents
There is some disagreement among harmonics analysts on how to define Q in the presence of
harmonic distortion. If it were not for the fact that many utilities measure Q and compute
demand billing from the power factor computed by Q, it might be a moot point. It is more
important to determine P and S; P defines how much active power is being consumed, while S
defines the capacity of the power system required to deliver P. Q is not actually very useful by
itself. However, Q1, the traditional reactive power component at fundamental frequency, may be
used to size shunt capacitors.
The reactive power when distortion is present has another interesting peculiarity. In fact, it may
not be appropriate to call it reactive power. The concept of var flow in the power system is
deeply ingrained in the minds of most power engineers. What many do not realize is that this
concept is valid only in the sinusoidal steady state.
When distortion is present, the component of S that remains after P is taken out is not
conserved—that is, it does not sum to zero at a node. Power quantities are presumed to flow
around the system in a conservative manner.
This does not imply that P is not conserved or that current is not conserved because the
conservation of energy and Kirchoff’s current laws are still applicable for a waveform of any
shape. The reactive components actually sum in quadrature (square root of the sum of the
squares). This has prompted some analysts to propose that Q be used to denote the reactive
components that are conserved and introduce a new quantity for the components that are not.
Some analysts call this quantity D, for distortion power or, perhaps more correctly, distortion
voltamperes. It has units of voltamperes, but it may not be strictly appropriate to refer to this
quantity as power, because it does not flow through the system as power is assumed to do. In this
concept, Q consists of the sum of the traditional reactive power values at each frequency. D
represents all cross products of voltage and current at different frequencies, which yield no
average power.
P, Q, D, and S are related as follows, using the definitions for S and P given previously as a
starting point:

Therefore, D can be determined after S, P, and Q by

Some prefer to use a three-dimensional vector chart to demonstrate the relationships of the
components as shown in Figure 8-5. P and Q contribute the traditional sinusoidal components to
S, while D represents the additional contribution to the apparent power by the harmonics.

10253878 8-8
Figure 8-5
The Components Of Power For Non-Sinusoidal Currents

Power factor (PF) is a ratio of useful power to perform real work (active power) to the power
supplied by a utility (apparent power), i.e.,

Many devices such as switch-mode power supplies and PWM VFDs have a near-unity
displacement power factor – the power factor at fundamental frequency due to the phase angle
displacement between the voltage and current. However, the true power factor may be 0.5 to 0.6.
The poor power factor is almost entirely due to the D term.
A traditional power factor correction capacitor will do little to improve the true power factor in
this case because Q 1 is zero. In fact, if it results in harmonic resonance, the distortion may
increase, causing the power factor to decrease. The true power factor indicates how large the
power delivery system must be built to supply a given load. In this example, using only the
displacement power factor would give a false sense of security that all is well.
The bottom line is that distortion results in additional current components flowing in the system
that do not yield any net energy except that they cause losses in the power system elements they
pass through. This requires the system to be built to a slightly larger capacity to deliver the
power to the load than if no distortion were present.
To supply a load with significantly distorted load current, the system current-carrying capacity
must be larger than for a sinusoidal current.

Resonance
The impedance of power delivery systems generally appears to be inductive, at least at power
frequency. Distribution engineers don’t worry too much about resonance when performing basic
capacity analysis such as power flow and short circuit. However, capacitances exist on power
systems either from being intentionally added to correct power factor of motor loads or then

10253878 8-9
naturally exist in lines and cables. When capacitors are added to inductive power lines and
transformers there will be at least one frequency at which the circuit is resonant. There may be
several resonant frequencies, as we shall see later.
If the resonant frequencies line up with frequencies that are being produced by harmonic-
producing devices on the power system, there can be severe consequences. Resonance can yield
persistent overvoltages or overcurrents or both. Figure 8-6 shows a calculated capacitor current
waveform for an industrial power system in which the power factor correction capacitors tune
the system near the 11th harmonic. The harmonic current in this case has approximately the same
rms magnitude as the fundamental, or power frequency, current. This the total rms current in the
capacitor is approximately 140% of rated current. Capacitor failure can result when such currents
are allowed to exist.

Figure 8-6
Simulation of current in a Capacitor Bank in an Industrial Power System that is in 11th-Harmonic
Resonance

This waveform can be created from either series or parallel resonance depending on the location
of the capacitor with respect to the main 11th-harmonic source.
It is common to find situations in the power systems where the power factor capacitors have
tuned the system to near the 5th harmonic. These situations often result in the failure of some
component and the 5th harmonic resonance problem is often discovered in the failure
investigation. This phenomenon was first noted by Dugan when performing an arc furnace study
in 1975. [17] Investigation showed that manufacturers’ capacitor selection chart for correcting
common load power factors frequently resulted in the selection of capacitors with a kvar rating
equal to approximately 1/25th of the inductive short circuit kVA at the bus at which it was to be
applied. This results in a 5th-harmonic resonance. Figure 8-7 shows a simulation of energizing a
capacitor bank in such a situation.

10253878 8-10
Figure 8-7
th
Simulation of a System Going into 5 -Harmonic Resonance when a Capacitor Bank is Energized;
Voltage across Capacitor and Current in the Capacitor.

This simulation shows voltages in which the crest is much higher than normal as well as currents
that exceed the capacitor rated current. If this condition is allowed to persist, the breakdown of
the insulation between the plates in the capacitor can be accelerated, significantly shortening the
life of the capacitors. The high currents may also cause capacitor failure from thermal causes or
result in capacitor fuse blowing. If the current also passes through a heavily-loaded transformer,
the transformer may overheat from the increased stray losses.
Thus, it is important to understand harmonic resonance and learn how to identify it so that
measures may be taken to prevent it. Before leaving the discussion of these two figures, notice
that the capacitor currents during resonant conditions generally contain just two key frequencies:
the harmonic at which the system is resonant riding on top of the fundamental frequency current.
There are no harmonic-producing loads in the power system that produces such a current
waveform. It will only appear in the capacitor that is participating in the resonant circuit. This
observation leads to one key rule for investigating power quality problems such as capacitor
failures when resonance is suspected:
Measure the power factor correction capacitor current!
Experience has shown that power engineers tend to focus on line currents in the main feed. There
may be existing current transformers so that it is relatively easy to measure the line currents and
it may take extra effort to measure capacitor currents safely. Unfortunately, line current
measurements can obscure the harmonic distortion problem if the harmonic currents are
swamped by power frequency currents from motors and resistive loads that are relatively clean.
If the capacitor current waveforms can be obtained safely, make it a high priority to measure
them.

10253878 8-11
There are two basic kinds of resonant circuits:
1. Series resonance in which the inductance and capacitance are in series (see Figure 8-8)
2. Parallel resonance in which the inductance and capacitance are in parallel (see Figure 8-9)
These are described in greater detail in the next sections.

Series Resonance
In series resonance (Figure 8-8), the apparent impedance across the resonant branch reaches a
minimum at the resonant frequency – the frequency at which the reactance of the inductor equals
the reactance of the capacitor. The net impedance of the branch is then the value of the resistor.
If one were to set the injection current, I(ω), equal to 1.0 + j0 (1.0 at 0 degrees) at all frequencies,
the voltage, V(ω), across the entire circuit will be equal to the total branch impedance, Z(ω), and
will have a frequency response characteristic similar to the shape shown in Figure 8-8. This is a
common way that power engineers perform frequency scans of networks, examples of which will
appear repeatedly in this document.
At the resonant frequency, the voltage across the capacitor reaches a maximum value. Likewise,
the voltage across the inductor also reaches a maximum and is nominally180 degrees out of
phase with the capacitor voltage. This is an important concept to remember for power system
harmonics: when this condition occurs it is frequently the voltage across the capacitor that is the
source of the harmonic distortion problems that are being experienced.

R
I(ω) V(ω) = Z(ω)

L
ω=2πf

Figure 8-8
Series Resonant Circuit

Because the impedance of the branch is low at resonance, the current in the branch for series
resonance can be high at the resonant frequency when there is sufficient harmonic source
strength to supply the current. One symptom of this condition can be capacitor failure by
overcurrent.
The low impedance of series resonance is often exploited to purposely create a series-tuned
branch to filter off harmonic currents from the rest of the power delivery circuit. This is the
essence of the series-tuned filter, which must be designed with care to minimize the risk that the
capacitors will see excessive current or voltage. This filter is also referred to as a “notch” filter
due the dip, or notch, in the impedance characteristic at the tuned frequency. It may also be

10253878 8-12
called a “shunt” or “resonant shunt” filter because it is generally placed in shunt with power
system loads to shunt away harmonic currents near the tuned frequency. However, this term can
be confusing because “shunt” often refers to devices in parallel and it is certainly not parallel
resonance, which is the next topic.

Parallel Resonance
In parallel resonance (Figure 8-9) the capacitive and inductive elements participating in the
resonance are in parallel with each other. As with series resonance, the capacitive reactances and
inductive reactances are equal at the resonant frequency. Instead of a notch in the impedance
characteristic at this frequency, the impedance reaches a maximum value as illustrated in Figure
8-9.
It is more difficult to push current through this circuit at the resonant frequency. The voltage
across the capacitor can be very high if the harmonic current source has significant strength. The
currents in both branches of this parallel circuit can be much larger than the injected current,
I(ω). This is an important concept because it results in the apparent “magnification” of harmonic
currents in many situations on the power system. That is, the harmonic currents measured at one
location are larger than those contributed by any actual harmonic-producing loads.
This can thwart attempts to locate sources of harmonic currents by following only the magnitude
of currents. The magnitude of the supposed troublesome harmonic current can drop off
dramatically as one moves past the capacitor location and the investigator can “lose the scent” of
the source in all the other currents in the system. It is generally better to follow the direction of
the active harmonic power if it is sufficiently large to be measured accurately. Harmonic-
producing loads on the power system tend to take in active power at fundamental frequency and
appear to inject powers at harmonic frequency back into the power supply system. This is simply
the result of the natural power balance in a circuit with highly distorted currents and lightly-
distorted voltages – the usual case in a power system.
Of course, tracing power flows requires simultaneous measurement of the voltage and current
waveforms to properly determine the phase angles between the harmonic voltages and current.
Many modern power quality meters can perform this task properly.

C I(ω) V(ω) = Z(ω)

L
ω=2πf

Figure 8-9
Parallel Resonant Circuit

Parallel resonance could be exploited to make blocking filters. That is, filters that block the flow
of certain harmonics because of the high impedance of a parallel-resonant circuit. This is
sometimes employed to block zero-sequence harmonic currents from flowing in the neutral of

10253878 8-13
grounded capacitor banks, but is used infrequently to block line currents. It is difficult to apply a
parallel-resonant filter in series with line current and properly account for other power system
concerns such as short circuit currents and insulation requirements. Both sides of the filter must
be insulated for full system BIL. Parallel filters are used extensively to help extract low-power,
high-frequency power line carrier (PLC) signals from power lines.

Situations with Both Series and Parallel Resonance

Many situations involving harmonic resonance on the power system exhibit both series- and
parallel-resonant characteristics; it is a matter of perspective of the observer. Figure 8-10
illustrates one situation that occurs in many places where there is a transformer between two
voltage levels with a power factor correction capacitor on the load side. This can occur either at
utility substations or at end-use, or customer, service connection locations.
Looking from the left side, the circuit appears to be a series-tuned resonant shunt. The
transformer is inductive with a high X/R ratio, so the tuning is relatively sharp and the filter
effect will siphon harmonic currents off the supply circuit to the left. Often this is the primary
distribution feeder that supplies many loads and collects the distorted currents from them. The
transformer-capacitor combination acts as series-tuned filter for the distribution feeder. The
result could be that the harmonic currents from so-called “background distortion” overload the
capacitor even if there are no harmonic-producing loads on the load side in the diagram. Also,
since it is a series-resonant circuit, the voltage at the junction of the inductive and capacitive
elements can be quite high. The typical case in which this occurs is when an end user with
relatively clean loads attempts to abide by utility power factor requirement by applying power
factor correction capacitors and finds significant voltage distortion and, perhaps, capacitor
failures due to overload or overvoltage.

EQUIVALENT SEEN EQUIVALENT SEEN

FROM THIS SIDE FROM THIS SIDE

(SERIES) (PARALLEL)

Figure 8-10
A Common Power System Configuration that Can Appear Either Series Resonance or Parallel
Resonance Depending on Perspective

From the right side of the diagram (load side), the source impedance appears to be a parallel-
tuned resonant circuit to any harmonic currents produced by the load. If we assume the distorting
source on the left side (primary) are minimal, the primary side of the transformer is effectively
shorted to ground at the resonant frequency, which is a non-power frequency. This is simply an
application of the principle of superposition, a basic electrical engineering concept. The
harmonic currents injected into this circuit at a frequency near the tuned frequency appear in

10253878 8-14
significant quantities in both the transformer and capacitor branches and will appear to be larger
on the primary side than can be explained by measurements of the actual load harmonics.
This circuit appears in many places in both transmission and distribution system. Whether or not
it will result in problems depends on the location of harmonic sources relative to the circuit and
many other factors including the nearby presence of resistive loads that tend to damp out
resonance.

Resonance and Harmonic Distortion Problems

The electric power system is remarkably robust. When it is mostly inductive and planned with
sufficient capacity to deliver the fundamental frequency power to the load the harmonic
distortion generally remains within bounds.
Problems in the electric power system with harmonic distortion are most frequently due to
resonance of some form when the capacitance in the system results in resonance at one, or more,
critical harmonic frequencies. This dramatically increases the distortion above normal amounts.
The standard that governs harmonic distortion on the power system is IEEE Std. 519-2014 -
IEEE Recommended Practice and Requirements for Harmonic Control in Electric Power
Systems. The recommended practice divides the problem into two parts:
• Limiting the amount of harmonic current that a load may inject into the power system.
• Limiting the voltage distortion on the power system.
Utility distribution planners usually do not have much control over the harmonic currents a
consumer may inject into the distribution system other than including references to IEEE Std
519-2014 limits in the interconnection agreement. Thus, it is always a good idea for planners to
investigate the types of loads that commercial and industrial customers are proposing to connect
to the system even if it is not explicitly the planner’s responsibility.
Distribution planners have more control over the harmonic voltage distortion that results from
the harmonic currents originating in consumers’ load equipment. The planners’ basic
responsibility is to design the system with sufficient capacity to supply the load kVA demand. As
mentioned previously, this is sufficient for accommodating most typical harmonic-producing
loads. The introduction of capacitance into a mostly inductive circuit will always result in a
resonance at some frequency. The challenge to the distribution planner to meet the utility’s
responsibility in IEEE Std. 519-2014 is to keep the system out of resonance at a harmonic
frequency present in the load currents.
Thus, the responsibilities in meeting IEEE Std. 519-2014 boil down to the follow two:
1. Consumers operate their loads so that the amount of harmonic current injected into the power
supply system is less than the current limits in the standard. Filters are applied if necessary.
2. Utilities build the power delivery system with sufficient capacitor to supply the load with
typical margins and operate the system to keep it out of harmonic resonance. Filters and
detuning devices are applied if necessary.

10253878 8-15
Sharpness of Resonance on the Power System
While resonance problems can occur anywhere on utility power distribution systems, the most
severe cases are usually found in industrial power systems because of the sharper resonance that
occurs. Power factor correction capacitors are typically applied at the secondary bus and there is
proportionately less equivalent resistance at the point of connection. Capacitors installed in
utility substations are subject to the same considerations. Engineers often describe the sharpness
of the resonance in terms of the “Q” factor. The Q of the resonant circuit is much greater at a
transformer location than is typically found when capacitors are placed on the distribution
system lines.
Q is typically defined for power systems harmonics resonance problems as:
𝑋𝐿
𝑄= 𝑅

Where
X L = reactance of the equivalent inductance of the circuit at the resonant frequency, and
R = resistance in the resonant circuit.
When Q is very high – as it is for a capacitor located on a transformer bus – the harmonic
resonance is very sharp and devices in the resonant circuit are more prone to failure. The X/R of
a substation transformer or a large industrial service transformer might be 20 at fundamental
frequency. If the transformer-capacitance combination is resonant at the 5th harmonic – which is
common – the Q might be approaching 100.
In contrast, the X/R on a distribution line might be 2-4 and Q at the resonant frequency being
less than 10. Load damping effects will further reduce the Q. Thus, distribution engineers
typically can place feeder capacitor banks anywhere they are needed without concern for
harmonic resonance. Resonance problems can arise, but they are generally less extreme than for
capacitors located near transformers.
Figure 8-11 illustrates this sharpness of the by plotting the magnitude of the impedance of a
parallel resonant circuit for different values of Q. Figure 8-12 shows the same data in another
way more pertinent to power system planning – the magnification of the injected current in the
inductive element (the transformer) for different values of Q.

10253878 8-16
 5000
4500
4000
3500
3000
|Z| Q=100
2500
Q=50
2000
Q=10
1500
Q=5
1000
500
0
0 2 4 6 8 10
Harmonic

Figure 8-11
Illustrating the effect of increasing the apparent resistance (decreasing the Q) of a resonant
circuit.

45
40
35
30
Magnification

25 Q=100
20 Q=50

15 Q=10

10 Q=5

5
0
0 2 4 6 8 10
Harmonic

Figure 8-12
Magnification factor: Amps through transformer per amp injected.

Harmonic Current Flow in Distribution Systems

Figure 8-13 shows the normal flow of harmonic currents in a radial distribution system when
resonance is not involved. The direction of the arrows is correct for the nominal case: harmonic-
producing loads take in fundamental current and effectively convert some to harmonic currents,
the largest portions of which tend to flow back out of the load toward the utility source. The
reason is simple: the utility source offers the lowest impedance return path.

10253878 8-17
 Normal Harmonic
Current Flow Paths

Distributed Harmonic
Current Sources

Figure 8-13
Harmonic currents from sources of harmonic distortion tend to flow toward the utility source.

Resonance changes things. Figure 8-14 shows a case where the power factor capacitor at the
substation is in resonance at a harmonic frequency with the substation transformer and source
inductance. This resonance is relatively high Q and it only takes a little excitation at the resonant
frequency to cause high current flow in both the transformer and capacitor. From the perspective
of loads on the feeder the resonance appears to be a parallel resonance. Thus, there are high
currents in both branches of the parallel circuit and significant voltage distortion at the bus at the
resonant frequency.

Harmonic Current
Magnified by Resonance

Distributed Harmonic
Current Sources

Figure 8-14
Parallel resonance at the substation magnifies the current injected into the power system

This is a common scenario on utility distribution systems and industrial power systems.
Measurements of line currents on the source side of the transformer will often have higher
harmonic currents at one frequency than can be justified by the known harmonic currents from
the various loads. The currents in the capacitors will be the clearest indication that resonance is
occurring.
Another scenario that results in altered harmonic current flow paths due to resonance is depicted
in Figure 8-15. An industrial facility has installed power factor correction capacitor on the load
side of its service transformer. As often happens without careful design, the capacitor and
transformer combination is tuned to one of the key harmonics produced by loads on the
distribution system. From the perspective of the distribution feeder, the facility appears to be a
series-tuned filter. This alters the normal flow pattern for harmonic currents of that frequency.
They now flow to the industrial site rather than into the utility source. Even if the industrial

10253878 8-18
customer has no harmonic-producing loads the service bus voltage could be significantly
distorted and the capacitor bank seeing severe duty from harmonic currents.

Altered Harmonic
Current Flow Paths Due to
Series Resonance

PF Cap.

Distributed Harmonic
Current Sources

Figure 8-15
The normal flow of harmonic currents are altered by series resonance from power factor
correction capacitors.

Modeling Nonlinear Loads When Resonance is Present

We have seen previously that it is common for harmonic analysis of power systems in the
frequency domain to replace nonlinear harmonic-producing by a current source at each harmonic
of interest. This is the default model in most harmonic analysis tools. However, when the system
is near resonance, a simple ideal current source model will give an excessively high prediction of
voltage distortion.

Figure 8-16
Replacing Simple Current Source Models with Thevenin or Norton Equivalents to Get Better
Answers for Simulations at Resonant Frequencies.

10253878 8-19
As depicted in Figure 8-16, this kind of model would try to inject a constant current into a
parallel resonant circuit which has a high impedance at its resonant frequency, which is not a
valid representation of reality. Our experience has shown that once the voltage distortion exceeds
approximately 15% the nonlinear load changes and no longer injects a constant current.
Sometimes the simple current source is adequate for planning purposes as long as the planner
does not try to justify the predicted voltages. The knowledge that the system is in resonance is
often sufficient to justify seeking a remedy. Once the resonance is eliminated by changing a
capacitor size or adding a filter, the simple model will give a realistic answer.
For the cases where a more accurate estimate of distortion is required during resonant conditions,
a more sophisticated model must be used. For many power system devices, a Thevenin or Norton
equivalent as shown in Figure 8-16. The additional impedance moderates the response of the
parallel resonant circuit. This is the default approach of the EPRI OpenDSS program, which
serves as the solution engine behind the Harmonic Evaluation Module. We will explore this
model in the next section.
Norton Equivalent Load Model for Harmonic Analysis
A simple load model commonly used in harmonics analysis is a Norton equivalent as shown in
Figure 8-17. [19] The current source in the model is set to the value of the fundamental current,
I fund , determined from a power flow solution times the multiplier for the harmonic spectrum
assumed for the load at each frequency. The load equivalent admittance, G + jB, may be
represented in the model as shown with only the susceptance, B, adjusted for frequency.

Figure 8-17
Simple Norton Equivalent Model of a Load for Harmonics Analysis

This model of a load may be derived directly from the typical power flow load model specified
by active and reactive load values, P and Q. G and B would be determined from P and Q
typically at rated voltage.
𝑉2 𝑉2
𝐺= 𝐵=
𝑃 𝑄
Where
V = 100% rated voltage in volts
P = Load active power in watts
Q = Load reactive power in vars
Admittances for models in harmonics analysis on distribution systems are typically used directly
in units of actual siemens. Per-unit values are not commonly used in unbalanced multiphase
harmonics analysis due to possibilities for errors.

10253878 8-20
One side effect of this approach to load modeling is that loads that are highly resistive may
provide significant damping of harmonic resonance if they are large enough. Whether or not this
is desirable depends on the motivation of the analysis. This will produce lower estimates of the
voltages and currents resulting from the resonance.
For frequencies where the system is not in resonance, most of the current produced by the
current source in the model flows back into the power system. The short circuit impedance of the
typical power system looking into it from a load is usually less than 5% of the load’s equivalent
impedance. Therefore, very little current is siphoned off into the shunt admittance branch of the
Norton model.
At frequencies where the system is near resonance, the driving-point impedance looking into the
system can be very high. A significant portion of the harmonic current is bled off into the shunt
admittance branch of the model. This keeps the predicted voltage distortion more reasonable than
an ideal current source, but it may also provide excessive damping.
A More Detailed Load Model
This describes the present load model in the EPRI OpenDSS program as modified per user
requests to have more control over the amount of damping caused by the load model at resonant
frequencies. [19] Figure 8-18 shows the revised model schematic diagram.
A series R-X branch was added to the existing parallel G-B branch. The program’s user can
specify the percentage split between the two. The default split is 50/50. The impedance of the
series branch behaves considerably different than the parallel branch. It becomes more inductive
and provides less damping to resonance as frequency increases. It also tends to shift the resonant
frequency slightly higher. The inductive part of the parallel branch becomes a high impedance at
the higher frequencies, resulting in the branch appearing more resistive.
Many analysts prefer to represent motor load as a series branch for harmonic analysis and the
OpenDSS program provides a special model for motor load as do other harmonic analysis tools.
Instead of determining R and X from the load P and Q, it is estimated from the equivalent
blocked rotor impedance. This results in a lower impedance at lower harmonics that rises with
frequency to be mostly inductive at the higher harmonics.

Figure 8-18
Load model with both series and parallel branches

10253878 8-21
Modeling Transformer Impedance Variation with Frequency
Generally, the resistance in a power system has a minor effect on the flow of harmonic currents
when the system is not in resonance. However, the damping of harmonic resonance by resistance
of loads, lines, and transformers can Loads are not the only elements in the power system that
have significant impedance variation with frequency. Substation transformers and larger
transformers supplying industrial consumers have a relatively high X/R ratio of 10 or greater at
fundamental power frequency and this contributes to sharp resonances. Distribution service
transformers such as those that serve residential loads can have a much lower X/R. A 25 kVA
transformer would have an X/R ratio only slightly greater than 1.0. In either case, there is a
question about what to assume for the variation of the equivalent resistance for harmonic
frequencies.
If no adjustment to the winding resistance for frequency is made, the equivalent X/R will
increase in proportion to the harmonic. Such a model predicts very little damping at harmonic
frequencies and excessively sharp resonances. For example, if a substation transformer has an
X/R ratio of 10 at fundamental, the model will have an X/R of 50 at the 5th harmonic, which
generally results in an unrealistically high-Q circuit model with highly exaggerated predictions
of voltage distortion.
The apparent resistance of transformers increases with frequency at a rate that is dependent on its
design. The chief component of the increase comes from the stray eddy current losses and can be
quite significant in transformers that have conductors with large cross-sectional areas. Also,
designs with conductors in parallel can have circulating currents within the windings that yield
an effective increase in resistance.
Utility engineers do not have time to measure the frequency response of each transformers, but
they know that the apparent resistance of the transformer increases with frequency and helps to
hold system resonances in check. A typical assumption of harmonics analysts when no other
data are available is to assume that the X/R ratio at fundamental frequency remains constant over
the frequency range of interest. This approximates what happens in larger power transformers. It
is not a good fit for some transformers, but at least it adds some damping to the model to nullify
the exaggerated high voltages and impedances that would otherwise be predicted.
Adjusting the resistance of small utility distribution transformers in the frequency range up to the
13th harmonic is generally not critical. The windings are constructed with wire having a small
cross section and the stray eddy losses do not generally increase as rapidly as for large power
transformers. The typical low X/R of these transformers tends to contribute to the damping of
resonance anyway, yielding results that are only moderately conservative.

Combined Effect of Load and Transformer Modeling

Figure 8-19 shows the effect of the various load and transformer modeling assumptions we’ve
discussed on the magnitude of the impedance looking into a typical parallel resonant circuit. The
all-parallel simple Norton equivalent yields the greatest damping and will be overdamped for
some cases. It is interesting to note that this model often matches well with measurements when
the capacitor banks are mostly on distribution lines rather than in substation. The likely
explanation is that the low X/R of distribution feeders damps resonance similarly.

10253878 8-22
 Comparison of Harmonic Scans with
Different Load Assumptions
1000 No Load
900
800
No Load, X/R
700
|Z| Driving Point

Const
600
500 All Series RL
400
300
Default
200
100
0 All Parallel RL
0 10 20 30
Harmonic

Figure 8-19
Comparing the impact of different load and transformer modeling assumptions on resonance

Assuming the X/R of the transformer is constant drops the magnitude of the driving point
impedance at the capacitor location by approximately 20% compared to the model that has no
load and no extra damping. This can be important in many cases where there is a sharp resonance
in a substation.
Assuming the impedance branch in the load model is all series RL shifts the frequency higher
and provides a little damping at the resonant frequency. Note that if this model is incorrect it
could shift the resonance either into or out of a troublesome harmonic frequency. This is
sometimes the reason that model predict a high resonance that is not observed in measurements.
The default 50/50 split in the load model admittance branch often a good compromise when no
better data are available. It will clearly show the resonance and will generally not exaggerate the
voltage distortion that would occur.
Figure 8-20 shows a one-line diagram of the OpenDSS REACTOR model. The model is
nominally a series, multiphase R-L with user-defined properties of R and X. In addition to scalar
values, R and X may also be defined as matrices. A feature of the model that, perhaps, is seldom
used is the parallel resistance, Rp, that is around the entire branch. Its default value is infinite
(open) so that it doesn’t enter into the calculations. However, it can be employed to model
frequency dependence of R-L elements, including transformers. This would require a separate
REACTOR to be added in series with the transformer and defined with an appropriate value so
that the total through impedance of the transformer is correct. Users may also define curves for
the resistance, R, and inductance, L, as a function of frequency when Rp is not defined.

10253878 8-23
Figure 8-20
One-Line Diagram of the OpenDSS REACTOR Object

Avoiding Resonance
The simplest way to solve problems with harmonic resonance is to make a determined effort to
avoid resonance at any of the critical harmonics such as 5, 7, 11, and 13. When planning for the
addition of a capacitor bank one should always check for the resonant frequency that will result
when it is installed. One quick way for most power engineers to estimate the resonant frequency
when installing a capacitor immediately downline from a transformer is to use the available short
circuit kVA (kVA SC ) and the capacitor kvar rating:

𝑘𝑘𝑘𝑆𝑆
ℎ≅�
𝑘𝑘𝑘𝑘𝑐𝑐𝑐

If kVA SC is not supplied by any other means such as a short circuit study, it can be estimated
from the transformer kVA rating and percent reactance, %X, at the base kVA rating, kVA TR :
𝑘𝑘𝑘 𝑇𝑇
𝑘𝑘𝑘𝑆𝑆 ≅ × 100
%𝑋

The actual value will be lower than given by this formula because there will be impedance in the
power supply to the transformer. One common approach is to use 90% or this value when more
accurate information is not available.
While special computer programs are generally required due to the complexity of many
distribution system circuit models, one circuit appears frequently in simple industrial systems
that is tractable by manual calculations (Figure 8-21). It is basically a one-bus circuit with one
capacitor. Two things may be done relatively easily:
1. Determine the resonant frequency. If the resonant frequency is near a potentially damaging
harmonic such as the 5th or 7th, either the capacitor must be changed or a filter designed.
2. Estimate the Voltage distortion at each frequency. The voltage across the potentially parallel
resonant circuit can be computed by a relatively simple formula.

10253878 8-24
Figure 8-21
A simple circuit that can be analyzed by manual calculations

The formula for estimating the voltage magnitude at each harmonic, h, is given below. Note
that this formula includes the effect of the resistance, R.

If the resonant frequency is not near a significant harmonic and the projected voltage distortion is
low, the application will probably operate successfully. Planners should avoid applications where
the resonant frequency is near the 5th or 7th harmonic especially. Tuning near even harmonics
like the 8th are often successful unless there are significant amounts of interharmonics. Tuning at
the 6th runs a greater risk of accentuating both the 5th and the 7th, so this application should be
designed with great care.
Adding just one more L-C loop to this circuit makes is difficult to do by manual calculations.
Typical distribution systems with power factor correction capacitor banks may have dozens or
even hundreds of such loops and require software for network harmonic analysis. There will be
not just one resonant frequency but several. Figure 8-22 shows the results of frequency scans of a
distribution circuit for all possible capacitor bank configurations using the EPRI Grid-IQ
Harmonics Evaluation Module. A number of configurations yield resonances in the 7th-9th
harmonic range with a relatively broad tuning that suggests that there will be accentuation of
these common harmonics even if the system is not tuned precisely to the 7th or 9th. The vertical
axis on this plot suggests that one should expect about 12 V per amp of injected harmonic
current in this range.

10253878 8-25
Figure 8-22
Positive-Sequence frequency scan of all possible capacitor configurations with the EPRI Grid-IQ
Harmonics Evaluation Module.

When there are this many possibilities for resonance, how can resonance be avoided? This is
where harmonic distortion studies come in. Engineers will want capable tools for this job so that
they can quickly investigate possible solutions.

Harmonic Studies
Harmonic studies in combination with measurements play an important role in characterizing
and understanding the extent of harmonic problems in power systems. Harmonic studies are
performed for the following purposes: [18]
1. Finding a solution to an existing harmonic problem
2. Installing large capacitor banks on utility distribution systems or industrial power systems
3. Installing large nonlinear devices or loads
4. Designing a harmonic filter
5. Converting a power factor capacitor bank to a harmonic filter
Harmonic studies are very important when the conditions exist for harmonic resonance. Without
resonance, most power systems with sufficient capacity to serve the load demand can also handle
typical distorting loads without excessive harmonic distortion. Harmonic studies are often
neglected because distribution engineers, in particular, can usually apply capacitor banks where
needed for loss reduction, voltage profile improvement, or power factor correction without
concern for resonance. The main reason is that capacitor banks distributed on distribution lines
are in locations where the X/R ratio of the equivalent short-circuit impedance is relatively low.
Thus, resonance is heavily damped for the majority of the installations. However, harmonic

10253878 8-26
studies should always be performed when applying large capacitor banks at or near transformers.
The transformer X/R is much higher than power lines and dangerous harmonic resonance can
easily occur.
Harmonic studies range from relatively simply resonant frequency calculations to quite
complicated simulations of large networks requiring sophisticated computer models. The studies
provide a means to evaluate various possible solutions and their effectiveness under a wide range
of conditions before implementing a final solution.

Harmonic Study Procedure

The ideal procedure for performing a power systems harmonics study can be summarized as
follows: [18]
1. Determine the objectives of the study. This is important to keep the investigation on track.
Objectives could include identifying resonances and correcting the system frequency
response.
2. Make a preliminary computer model to identify likely resonance situation.
3. Make measurements of the existing harmonic conditions, characterizing sources of harmonic
currents and system bus voltage distortion.
4. Calibrate the computer model using the measurements.
5. Study the new circuit condition or existing problem.
6. Develop solutions (filter, detuning options, etc.) and investigate possible adverse system
interactions. Also, check the sensitivity of the results to important variables.
7. After the installation of proposed solutions, perform monitoring to verify the correct
operation of the system.
Admittedly, it is not always possible to perform each of these steps ideally. The most often
omitted steps are one, or both, measurement steps due to the cost of engineering time, travel, and
equipment charges. An experienced analyst may be able to solve a problem without
measurements, but it is strongly recommended that the initial measurements be made if at all
possible because there are often surprises when performing harmonics analysis of power
systems.
If the subject power system is complex, it is often economical to make an initial computer model
prior to making measurements using the best information available. Harmonic simulation
technology is now sufficiently advanced that models can often make fairly good predictions
without measurements. Measurements are very beneficial but are very expensive in terms of
labor, equipment, and possible disruption to plant operations. It will generally be economical to
have a good idea what the likely problems will be and where to look before beginning the
measurements. Then the investigation team can take the monitoring equipment directly to the
likely problematic locations.

10253878 8-27
Principles for Controlling Harmonics
When a problem occurs, the basic options for controlling harmonics are:
1. Reduce the harmonic currents produced by the load.
2. Add filters to either siphon the harmonic currents off the system, block the currents from
entering the system, or supply the harmonic currents locally.
3. Modify the frequency response of the system by filters, inductors, or capacitors.
Of these, the 3rd topic is of most interest with respect to the subject here.

Modifying System Frequency Response

There are a number of methods to modify system frequency response when resonance occurs at
harmonic frequencies: [18]
1. Change the capacitor size. This is often the least expensive options for both utilities and
industrial customers. Simply move the resonance frequency away from a harmonic.
2. Add a shunt filter. Not only does this shunt a major troublesome harmonic current off the
system, but it completely changes the system response. When properly designed the change
is beneficial
3. Add a reactor to detune the system. Harmful resonances generally occur between the system
inductance and shunt power factor correction capacitors. The reactor must be added between
the capacitor and the supply system source. One method is to simply put a reactor in series
with the capacitor to move the system resonance without actually tuning the capacitor to
create a filter. Another is to add reactance in the line.
4. Move a capacitor to a location on the system with a different short-circuit impedance or
higher losses. This is also an option for utilities when a new bank causes telephone
interference—moving the bank to another branch of the feeder may very well resolve the
problem. This is frequently not an option for industrial users because the capacitor cannot be
moved far enough to make a difference.
5. Remove the capacitor and simply accept the higher losses, lower voltage, and power factor
penalty. If technically feasible, this could be an economic choice.
The X/R ratio of a utility distribution feeder is generally low. Therefore, the magnification of
harmonics by resonance with feeder banks is usually minor in comparison to what might be
found at an industrial facility. Utility distribution engineers are accustomed to placing feeder
banks where they are needed without concern for harmonics. When problems do occur, the usual
strategy is to first attempt a solution by moving the offending bank or changing the capacitor size
or neutral connection.
Harmonic problems on distribution feeders often exist only at light load. The voltage rises,
causing the distribution transformers to produce more harmonic currents while at the same time
there is less load to damp out resonance. Switching the capacitors off at this time frequently
solves the problem.

10253878 8-28
When harmonic currents from widely dispersed sources require filtering on distribution feeders,
one approach is to distribute a few single-tuned filters toward the ends of the feeder. With the
ends of the feeder “nailed down” by filters with respect to the voltage distortion, it is more
difficult for the voltage distortion to rise above limits elsewhere.
When harmonic resonance is suspected in an industrial facility, the first step is to confirm that
the main cause is resonance with power factor capacitors in the facility. This is done by
measuring the current in the capacitors and looking for the telltale waveform such as the one in
Figure 8-6. One should first attempt a simple solution by using a different capacitor size.
Some automatic power factor controllers with multi-step capacitors may have control logic that
allow them to avoid the capacitance value that causes resonance. In other cases, there will be so
many capacitors switched at random with various loads that it will be nearly impossible to avoid
resonant conditions. Filtering will be necessary, possibly with broadband filters.
Resonance problems are often less severe in factories when capacitors are located out on the
plant floor on motors and in motor control centers. This assumes that the cables are sufficiently
long to introduce enough resistance into the circuit to dampen the resonance. In plants with short
cables, it may not be possible to achieve significant harmonic reduction benefit by distributing
the capacitors.

Filters
When simple changes are not feasible, harmonic filters can be added to the system to alter the
frequency response, either moving or damping the resonance. Figure 8-23 shows some
commonly-applied filter topologies.

(a) (b) (c)

nd
Single- 1st- Order 2 - Order
Tuned High Pass High Pass
Resonant
Shunt

Figure 8-23
Common harmonic filter topologies

The most commonly-applied is the single-tuned resonant shunt. It is generally the least expensive
and the most efficient. The main intent with this filter is create a low impedance for a
troublesome harmonic current and short-circuit it off the system. This filter also changes the
overall system frequency response, which could cause problems at other frequencies. Figure 8-24

10253878 8-29
shows what happens when an existing capacitor that was causing a resonance at the 9th harmonic
was converted to a 5th harmonic filter. It is usually good practice to design the filter for one of
the lower harmonics on the system. The reason is that the filter creates a new, sharp resonance
below the notch frequency. This resonance should be moved to a frequency where it is not likely
to cause a problem. In this case, the new resonance occurs near the 4th and there is generally little
excitation of this resonance unless the system is serving cyclo-converter type loads.
The other two filters in Figure 8-23 introduce intentional resistance. Resistance helps to damp
out resonance so this can be quite useful if the losses are affordable. The 1st-order high pass filter
simply inserts a resistance into the resonant circuit sufficient to suppress the resonance. This is of
course quite lossy. However, there are applications where the heat off the resistor can be used
effectively and this simple filter works effectively. The 2nd-order high pass is a little more
sophisticated. It is typically applied in smaller sizes for 11th harmonic and higher where the
fundamental frequency voltage across the inductor is relatively low.

Figure 8-24
Converting an existing capacitor bank to a single-tuned filter

An increasingly popular filter is the C Filter (Figure 8-25). This yields a frequency response
similar to the 2nd-order filter, but can be tuned to a low harmonic such as the 3rd without
suffering the high losses of the 2nd-order filter. It does an excellent job of suppressing all
frequency above its main tuning frequency. The key feature is the series-tuned combination of
Lm and Ca tuned to the fundamental frequency that shorts the resistance, R, at the fundamental,
thus avoiding many of the losses.

10253878 8-30
 Lm
R

Ca

Cm

Figure 8-25
C Filter configuration.

The frequency-scan response of a C Filter is illustrated in Figure 8-26. It is tuned to the 5th
harmonic where the notch occurs. It presents a high impedance to fundamental frequency but has
a low impedance with resistive damping above the notch frequency. This can be effective in
filtering the higher frequency components of harmonic-producing loads.

C Filter Characteristic
1000

100

10

1
0 200 400 600 800 1000 1200

Figure 8-26
C-Filter Characteristic

A final type of filter to consider for controlling resonance is the broadband filter (Figure 8-27).
The application for this would be where there are multiple resonances and, perhaps, cyclo-
converter type loads that produce varying interharmonics such that it is almost impossible to
avoid a resonant condition. The main idea is to use a relatively large inductance to keep all
harmonic currents on the right-hand side of the diagram by forcing them through the capacitor.
The filter is thus tuned to a low frequency such as near the 2nd or 3rd harmonic. A significant
voltage rise at power frequency occurs at the capacitor due to the size of the capacitor relative to
the short circuit strength of the system. Therefore, a tap changing transformer is often employed
to control the voltage level. Such devices are available for industrial low-voltage applications

10253878 8-31
and are effective in minimizing the distortion over a wide range of frequencies. Such filters could
also be constructed on utility distribution systems using transformers for the inductance and
voltage regulators to control the overvoltage.

Harmonic Current
Sources

Figure 8-27
Broadband Filter Schematic

Figure 8-28 shows a broadband filter current magnification frequency response – the current
observed on the source side of the filter per ampere of current on the load side. The harmonic
current source sees a typical parallel resonance characteristic with a high impedance at the tuning
frequency. However, at higher frequencies the impedance approaches the impedance of the
capacitor which shunts the current off the system and strongly attenuates the currents about 300
Hz in this case (the magnification factor is less than one). This is a useful characteristics for
loads that produce harmonic and interharmonic currents that are constantly shifting frequencies.
Obviously, the tuning frequency must be lower than the lowest harmonic produced by the load.

Current Magnification
100

10

1
0 200 400 600 800 1000 1200

0.1

0.01

Figure 8-28
A Broadband Filter Characteristic

10253878 8-32
9
DYNAMICS
Dynamics Mode vs. Power Flow Modes
The modeling described in the preceding chapters was focused primarily on power flow analysis.
Even when performing QSTS simulations, which some analysts call “long-term dynamics”, the
time step is greater than 1 s and transients in rotating machines and inverter-base DER devices
are assumed to be settled out within the time step.
Dynamics, or electromechanical transients, analysis involve simulations with time steps of
approximately 1 ms. This is not to say the analysis recreates the 50 or 60 Hz waveform, but that
it performs all calculations in the frequency domain with complex phasor values. That is, with
average rms values instead of instantaneous value. Nevertheless there are differential equations
to be solved, which introduced a whole new level of complexity.
This analysis is very similar to transient stability simulations routinely performed on
transmission networks and a few engineers in the transmission planning department, or with the
Independent System Operator, would be expected to be experts in this analysis. However, this is
a type of analysis that is done infrequently in distribution planning and distribution planners are
typically unfamiliar with it. However, that could change with the next generation of engineers
because of the DER devices such as storage, PV systems, and generators in microgrids that have
important dynamic responses in less than 1 s.
It is unclear how much dynamics analysis will be done by utility distribution planners as the
modern grid advances. Microgrids will likely have a greater need for it than today’s
interconnected distribution grid. There are industry efforts like IEEE Std 2030 to develop stands
for microgrid controllers. What may transpire from this is that microgrids will be designed and
maintained by non-utility entities that are responsible for the control and stability of the
microgrid. The suppliers of the microgrid controllers will be the ones performing dynamics
studies to make sure the system is stable and sufficiently protected. This would remove the
burden of doing dynamics studies from the distribution planners, but it remains to be seen how
this planning function will evolve.

Figure 9-1
Generator model for power flow analysis

10253878 9-1
For power flow analysis, loads, generators, PV systems, and storage elements are treated
essentially as current sources that are a function of the voltage across the device so that they
produce or absorb the specified active and reactive power. A generator is shown in Figure 9-1
that is producing the power output P + jQ.
Figure 9-2 shows the generator model converted for dynamics analysis. This is a simple single-
mass model to illustrate the key points. Instead of a current source, we now have a voltage
source connected to spinning shaft. The voltage, E, and phase angle, θ, are initialized to match
the power flow solution obtained with the current source model.

Figure 9-2
Simple single-mass generator model for dynamics analysis

A differential equation is now introduced to the problem and the behavior of the generator model
is now determined by the familiar single-mass swing equation:
𝑑2𝜃
𝑀 =T
𝑑𝑑 2

where
M = intertial constant for the mass of everything connected to the shaft
T = the net torque on the shaft including the prime mover, electromagnetic torque, and
losses
θ = angular position of the shaft relative to the power system. The first derivative of θ is
the speed of the shaft and the second derivative is the acceleration
As the simulation proceeds from time step to time step, the differential equation is solved by
numerical integration. The acceleration is integrated to obtain the new shaft speed. Then the
speed is integrated to obtain a new estimate of the power angle, which determines how much
power is produced or absorbed by the generator as it swings. The current is determined by
simply solving the Thevenin equivalent in the electrical network to which it is connected,
similarly to linear power flow.

10253878 9-2
In a dynamics program, the generator model will usually include an exciter model to control the
value of E and a governor model that controls the input torque. For some short term analyses of
DER transients, it may be acceptable to assume that neither the exciter nor the governor act
simply to see how far the generator will swing before protection acts to remove it from the grid.
This can also be an effective model to compute the fault current contribution of distributed
generators on a distribution system:
1. Solve the power flow
2. Switch to the Thevenin model and initialize for dynamics analysis.
3. Then solve one or two small time steps with the fault applied.
Simple generator models are not onerous and most power engineers have received some training
in modeling the swing equation. Data required for a minimalist simulation suitable for some
screening studies are:
1. MVA or kVA rating
2. Xd’, or transient reactance. Typical value is 0.27 pu.
a. Xd”, or subtransient reactance if interested in currents in the first 2 cycles after a fault.
Typical value is 0.20 pu.
3. Mass intertial constant for the everything on the shaft, MW-s/MVA. Typically 1..2 pu.
4. A damping constant, usually in range of 0..4 that is multiplied by the shaft velocity to damp
spurious oscillations.
While it may be difficult to come up with precise values for these quantities, there are only 4
values to obtain, or estimate. When it becomes necessary to model such things as a microgrid
operating through several disturbances for a much longer period of time, models of the exciter,
governor, droop control, etc., will be needed. This significantly complicates a dynamics study
and will require an expert in dynamics analysis to execute correctly. A simple model is certainly
not adequate for long simulations with detailed models that might have time step sizes of less
than 1 ms. It will be difficult to hold the simulation stable for more than a few seconds.

Inverter-Based DER
Many of the DER devices being installed are connected to the distribution system through
electronic power converters instead of rotating machines. The devices take the dynamics
modeling difficulty to a higher level.
One major hurdle for distribution planners considering dynamics analysis of systems with
inverter-based DER is the overwhelming amount of data required for making detailed models.
For power flow analysis, only a few parameters are needed to describe devices connected to the
circuits. QSTS analysis adds more requirements with the need for load shapes, etc., but planners
usually have some mechanisms for dealing with load data because that is their business.
Modeling DER for dynamics is generally not their business.
Most planners find it difficult to obtain sufficient data for even the simpler DER devices such as
conventional gas engine-powered generators with synchronous alternators. As presented above,
dynamics simulations require transient and subtransient reactances, inertial constants, and some

10253878 9-3
idea of exciter and governor characteristics. These data are usually available for large
transmission-connected generators, but can be scarce for distributed generators on distribution
systems.
Inverter-based DER increases the number of parameters required for detailed models by an order
of magnitude. The storage controller model for which a partial block diagram is shown in Figure
9-3 was implemented in OpenDSS for an EPRI-EDF joint research project. [77] The model
required more than one month to assemble and debug. Not only is this amount of time
unreasonable in the distribution planning environment unless it is a very large capacity device,
but the model required assembling 31 parameters representing transfer functions, time constants,
PI controller loop gains, feedback gains, capacitances, and resistances into a control model. This
a formidable number for the typical small DER project. In addition, suppliers of the controllers
consider these values and control circuit configurations to be proprietary IP and are reluctant to
publish them.

Figure 9-3
Part of the block diagram of an ultracapacitor-based distributed energy storage system (EDF) [73]

10253878 9-4
The next section deals with one approach being proposed to address this problem: have the
suppliers of the controllers provide models in a compiled binary form such as a DLL that
conceals the proprietary details. The DLL would have a function call interface built according to
an industry standard or to a commonly used dynamics analysis tool. The pros and cons of this
approach are explored in the next section. Then two common DLL interface APIs are presented.

Binary Simulation Model Interface

While several of the familiar power systems analysis program were designed to accept compiled
DLLs, this approach was formally proposed by Fortmann and Hendriks [78]. At the time it was
primarily for modeling large wind power systems. The concept can obviously be extended to
solar PV generation and to microgrid controllers. The vendor provides the models so the user of
the power systems analysis software simple plugs it in to a compatible program and executes the
typical planning studies such as power flow, short circuit, and dynamics analysis of disturbance
events.
Standard models vs. vendor-specific models
Fortmann and Hendriks list the following comparisons in the referenced source, with a wind
power emphasis (parenthetical comments added to draw analogy to solar power):
Standard models apply for
• Behavior during faults and switching actions
• Wind power plants represented as single/few scaled units
• Constant wind speed (or like constant solar irradiance)
Vendor-specific models required for
• Wind plant collector system design
• Control interaction
• Load rejection studies for islanding and reconnection
• Very weak systems
• Grid connection through HVDC
• Varying wind speed conditions. (i.e., like varying solar irradiance conditions)
Binary simulation model interface
• Models are implemented as a DLL and connected to the main program
• Multiple models and multiple instances of the same model
• Main program invokes the models by function calls, define in a standard application
programming interface (API).
• Data exchange defined in the API
Advantages
• Specific implementations not required in the main program core.
• Models are compiled black boxes.

10253878 9-5
• Interface designed with various stability simulation approaches in mind.
• Applicable to both dynamics and EMT simulation models
• Generally applicable to all sorts of dynamics models
• Supports initialization from standard models as well as custom load flow models
• Automatic extraction of parameters, limit checking and documentation.
This seems attractive, but there are also some issues to be considered. Probably the greatest is
that there are no compiled binary forms that are impervious to malicious usage. Software
developers executing the vendor-supplied DLLs from a debugger can readily decompile the
binary code. This allows a path for competitors to reverse engineer a proprietary control or
device model. Thus, many vendors are not completely on board with this idea.
Another issue is that distribution system analysis software evolves and new technologies are
developed. Who will be responsible for keeping the black box models up with the present
technology? Locking into a particular API may also have the side effect of stifling innovation if
the industry hesitates to make a change because of the ripples through all the software in use.
The next section examines two APIs for user-written models for two simulation programs
supplied as DLLs.

APIs for Binary Models

To get an idea of what is involved in providing user-written binary model, APIs for two
simulation programs are presented: EMTP-RV and EPRI’s OpenDSS. EMTP-RV is principally a
time-domain electromagnetic transients program, but also has steady-state analysis capability.
OpenDSS is strictly a frequency-domain program with capabilities of performing dynamics
simulations as well as power flow and harmonics analysis. Only the dynamics model for the
Storage model used to implement the model shown in Figure 9-3 will be presented.
EMTP-RV DLL function definitions
User-written models for EMTP-RV must support not only dynamics simulations, but EMT
simulations. Thus, there are more functions required than for a phasor-domain program that
performs only power flow and dynamics analysis.
For each request transmitted to the DLL, the expected request is based on a function definition.
The following is the list of the 35 available function definitions (reproduced from EMTPWorks
documentation [79]):
1 DLL_INITIALIZE_NEW(myname,idev,Data_Section,DLL_NAME)
1.a mandatory function
1.b myname: device name in EMTPWorks, string
1.c idev: device number in the list of all DLL devices, integer
1.d Data_Section: Array of strings, contents of ModelData after the DLL file
identification
1.e DLL_NAME: the name of this DLL file name with full path

10253878 9-6
2 DLL_POST_INITIALIZE_NEW(myname,idev,power_signal_nodes,n_nodes)
2.a mandatory function, called after DLL_INITIALIZE_NEW
2.b myname: device name in EMTPWorks, string
2.c idev: device number in the list of all DLL devices, integer
2.d power_signal_nodes: array of index numbers for all power signal nodes of the DLL
device
2.e n_nodes: size of the array above
3 DLL_DATA_POINTERS(idev,Pointer_simulation_data)
3.a optional function, always called first, before DLL_POST_INITIALIZE_NEW.
3.b idev: device number in the list of all DLL devices, integer
3.c Pointer_simulation_data: Pointers to the Simulation Data objects
4 DLL_INSERT_SOURCE_W_FOR_SS(idev)
4.a optional
4.b idev: device number in the list of all DLL devices, integer
5 DLL_PUT_VOLTAGE_ROW_EQUATIONS(idev,first_voltage_row)
5.a optional, called for req_put_nodes_in_Yaug request
5.b idev: device number in the list of all DLL devices, integer
5.c first_voltage_row: index of the first row in the list of voltage-defined equations
6 DLL_PUT_IN_YAUG_SS(idev)
6.a optional
6.b idev: device number in the list of all DLL devices, integer
7 DLL_PUT_IN_YN_SS(idev,w)
7.a optional
7.b idev: device number in the list of all DLL devices, integer
7.c w: the computation frequency in rad/s, double precision real number
8 DLL_PUT_IN_IAUG_SS(idev,w)
8.a optional
8.b idev: device number in the list of all DLL devices, integer
8.c w: the computation frequency in rad/s, double precision real number
9 DLL_PUT_IN_IAUG_FREQSCAN(idev,w)
9.a optional
9.b idev: device number in the list of all DLL devices, integer
9.c w: the computation frequency in rad/s, double precision real number
10 DLL_PUT_IN_YN_SS_FREQSCAN(idev,w)
10.a optional
10.b idev: device number in the list of all DLL devices, integer
10.c w: the computation frequency in rad/s, double precision real number
11 DLL_SUPERPOSE_SS_AT_W(idev,w)
11.a optional
11.b idev: device number in the list of all DLL devices, integer
11.c w: the computation frequency in rad/s, double precision real number
12 DLL_PRINT_SS(myname,idev,w,Current,Spower)
12.a optional
12.b Errors in this function may cause printed power mismatch
12.c myname: device name in EMTPWorks, string
12.d idev: device number in the list of all DLL devices, integer

10253878 9-7
 12.e w: the computation frequency in rad/s, double precision real number
12.f Current: returned double precision complex vector of currents entering each pin of
the DLL
12.g Spower: returned double precision complex vector of S powers entering each pin of
the DLL
13 DLL_INDEX_CONTROLLABLES (idev,control_valindex)
13.a Only when controlled input signals exist
13.b idev: device number in the list of all DLL devices, integer
13.c control_valindex: integer array, indexes of controlled (input) signals
14 DLL_LOAD_OBSERVABLES_T0(idev,Returned_obs_array)
14.a only when observable signals exist (control output signals)
14.b idev: device number in the list of all DLL devices, integer
14.c Returned_obs_array: double precision real array of values sent as observables
15 DLL_ZERO_INITIAL_CONDITIONS(idev)
15.a optional
15.b idev: device number in the list of all DLL devices, integer
16 DLL_INIT_AT_T0(idev)
16.a optional
16.b idev: device number in the list of all DLL devices, integer
17 DLL_EBA_INIT_AT_T0(idev)
17.a optional
17.b idev: device number in the list of all DLL devices, integer
18 DLL_PUT_IN_YAUG(idev)
18.a optional
18.b idev: device number in the list of all DLL devices, integer
19 DLL_PUT_IN_YN(idev)
19.a optional
19.b idev: device number in the list of all DLL devices, integer
20 DLL_PUT_NODES_IN_YNONLIN(idev)
20.a optional
20.b idev: device number in the list of all DLL devices, integer
21 DLL_ITER0(idev)
21.a optional
21.b idev: device number in the list of all DLL devices, integer
22 DLL_PUT_IN_IAUG(idev)
22.a mandatory
22.b idev: device number in the list of all DLL devices, integer
23 DLL_UPDATE_TOPOLOGY(idev)
23.a optional
23.b idev: device number in the list of all DLL devices, integer
24 DLL_ITER(idev,convergence_flag)
24.a optional
24.b idev: device number in the list of all DLL devices, integer
24.c convergence_flag: return true when converged, return false when did not converge
25 DLL_CONVERGENCE_MESSAGE(idev)
25.a optional

10253878 9-8
 25.b idev: device number in the list of all DLL devices, integer
25.c Allows sending a convergence problem message to the EMTPWorks progress panel
26 DLL_LOAD_OBSERVABLES(idev,Returned_obs_array)
26.a only when observable signals exist (control output signals)
26.b idev: device number in the list of all DLL devices, integer
26.c Returned_obs_array: double precision real array of values sent as observables
27 DLL_UPDATE_STATUS_AT_T(idev)
27.a optional
27.b idev: device number in the list of all DLL devices, integer
28 DLL_UPDATE_AT_T(idev)
28.a optional
28.b idev: device number in the list of all DLL devices, integer
29 DLL_UPDATE_AT_T(idev)
29.a optional
29.b idev: device number in the list of all DLL devices, integer
30 DLL_TRAPTOEBA_UPDATE_AT_T(idev)
30.a optional
30.b idev: device number in the list of all DLL devices, integer
31 DLL_EBA_UPDATE_AT_T(idev)
31.a optional
31.b idev: device number in the list of all DLL devices, integer
32 DLL_EBATOTRAP_UPDATE_AT_T(idev)
32.a optional
32.b idev: device number in the list of all DLL devices, integer
33 DLL_SAVE_ME(idev)
33.a optional
33.b idev: device number in the list of all DLL devices, integer
34 DLL_LOAD_ME(idev)
34.a optional
34.b idev: device number in the list of all DLL devices, integer
35 DLL_END(idev)
35.a optional
35.b idev: device number in the list of all DLL devices, integer

The source document provides excellent details on how EMTP-RV works internally and makes
detailed suggestions for coding the DLL. This interface has a Fortran-95 or Fortran-2003 flavor
to it and a buffer layer written in these versions of Fortran is suggested, but not required.
OpenDSS Storage Model DynaDLL Interface
The 8 key API functions in the DynaDLL interface of the Storage model in OpenDSS are listed
in Table 9-1. Note that user-written DLLs in OpenDSS are loaded by the parent device model
code rather than the main program. They are specified in the regular DSS script, in this case
during the definition of a Storage class device. There is a different API interface for each class of
device that supports DLLs. Generator models have a similar interface, but it is slightly different
than the Storage element DynaDLL.

10253878 9-9
This interface has a Delphi Pascal flavor to it, but the DLL can be written in any language
capable of producing a standard DLL. Note that Function types return a value while Procedure
types do not. All arguments are passed by reference (i.e., as a pointer). Thus, the interface can be
implemented in a wide variety of languages, but programmers must exercise discipline not to
overwrite memory. Like the EMTP-RV DLL interface, poorly written code can crash the
program, although OpenDSS has error trapping active and will generally successfully report
where the error came from before it halts.
OpenDSS comes in both 32-bit (X86) and 64-bit (X64) versions. The DLL must match the
version of OpenDSS being used. On most installations, both versions are installed. Thus, it is
advisable to develop the DLL using a compiler capable of producing both X86 and X64
executables.
A user-written DynaDLL imports variables and structures from the Storage model and OpenDSS
structures from the program’s Dynamics module. The connection to the Storage model data
structures is made through the Storage device’s public data pointer obtained from a callback
function. This allows data to be moved across the API directly through memory.
Table 9-1
The eight key functions in the OpenDSS storage model DynaDLL API

Type Function/Proc name Arguments

DynaData : TDynamicsRec;
Function New
CallBacks : TDSSCallBacks (returns: ID:Integer)
Procedure Delete ID:Integer
Function Select ID:Integer (returns ID or 0)
Procedure Init V, I: Complex Arrays
Procedure Calc V, I: Complex Arrays
Procedure Integrate
Procedure Edit Edit string or filename:Ansistring; Length of string
Procedure UpdateModel

The general usage of the API is as follows:

1. The New function creates of new instance of the custom storage model. Pointers to the
required dynamics data and callback routines are provided. An ID is returned that is
subsequently used to select this instance.
2. The Edit procedure is called to set the values of the model parameters. Since there could be
many parameters in the storage controller for dynamics, the values may be entered from a
file.
3. The Init procedure is called to initialize the state variables of the dynamics model from the
voltages and currents computed from the OpenDSS circuit solver for the initial steady-state
condition.
4. The Calc procedure computes the current given the present values of the terminal voltages
and the values of the state variables. This is called for each iteration of the solution process.

10253878 9-10
5. The Integrate function is called when the Dynamics simulation performs integration of the
state variables. This can be called twice for each iteration: the programmer must implement
both a predictor and a corrector step.
6. UpdateModel is called infrequently to synchronize model parameters with the base Storage
element model in OpenDSS.
There are, of course, many more details for DLL developers to understand that the developers
may obtain by inspecting the actual source code. The OpenDSS Dynamics algorithm is described
in the next section.

Example: The OpenDSS Dynamics Mode

This section describes how the Dynamics mode solution in the OpenDSS program works.
Examples are provided from the Generator element, which is usually the most common element
used in Dynamics mode simulations. This section also provides some information on how to
code power electronic and control blocks can be handled in DLLs for OpenDSS.
The Basic Algorithm
Each time step in the dynamics mode solution executes the following steps (extracted directly
from the code in the file SolutionAlgs.Pas):
Increment_time;

{Predictor}
IterationFlag := 0;
IntegratePCStates;
SolveSnap;

{Corrector}
IterationFlag := 1;
IntegratePCStates;
SolveSnap;

The algorithm is a simple predictor-corrector method with one step of correction. The
IterationFlag variable indicates to the integration routines whether the solution is in the predictor
step or the corrector step.
The IntegratePCStates function iterates through each Power Conversion (PC) element in the
circuit and executes its IntegrateStates function. PC elements are loads, generators, storage
devices, PV systems – devices that convert the power to and from energy sources and sinks.
IntegrateStates is a virtual function that is overridden in models that actually have something to
integrate. For example, Generator elements have state variables from the swing equation to
integrate. However, others do not and this function is simply empty.
Only PC elements can have differential equations and states that are integrated. Power Delivery
(PD) elements are constant impedance elements simply defined by a primitive Y matrix.
Going into Dynamics Mode
The state variables in the PC element models used in the dynamics simulation are initialized
from the most recent converged power flow, or other circuit solution.

10253878 9-11
In addition to Dynamics mode, the OpenDSS program goes into Dynamics solution mode for
FaultStudy and MonteFault (MF) solution modes so that contributions from Generator objects
and other active objects are more accurately captured.
The steps the program executes when going into Dynamics mode from one of the power flow
solution modes is as follows:
• Initialize state variables in all PC Elements
- For example, in a Generator object:
 Compute the voltage, E 1 , behind Xd' to match power flow as closely as possible
 Initialize the phase angle, θ, (power angle) of the equivalent voltage source
• Set derivatives of the state variables to zero so they are not moving or accelerating when the
dynamics simulation starts
- For the Generator, these variables are
 Speed (relative to synchronous frequency)
 Power Angle
• Set controlmode=time
- When running in time steps of a few seconds or less, controls that depend on the control
queue for instructions on delayed actions will be automatically sequenced when the
solution time reaches the designated time for an action to occur.
• Set a flag to preserve Node voltages in case the system Y matrix changes.
- Y matrix changes could occur if a switch were to open during the simulation, which
might re-order the bus lists. This flag prevents loss of data by reassigning the Node
voltage after the re-ordering.
Integrating the State Variables
Integration is performed within the PC element modules themselves, not in the Solution module.
Therefore, it is possible to have different integration formulae for different classes of devices.
The Generator models employs a simple trapezoidal-rule predictor-corrector integration method,
as do all the user-written models in OpenDSS that solve differential equations. This is a simple,
stable method commonly used in many power systems analysis programs.
Predictor and Corrector Steps
The predictor step offers algorithm designers the opportunity to insert a predictor formula at the
beginning of the process. It may differ from the corrector step only in that it updates the initial
value of the state variables for this time step based on derivatives at previous solution. The
following three steps are performed for both the predictor and corrector steps:
1. Compute derivatives for this time step
2. Integrate state variables
3. Solve the circuit for voltages with the latest result

10253878 9-12
For the present implementation of the Generator object, the state variables are the speed (relative
to synchronous speed) and the shaft angle, θ. The derivatives of these are computed and then
integrated in a trapezoidal formula as follows:
Derivative Calculation (Generator model):
dv Pshaft − Pterm − D v
=
dt M
dθ
=v
dt
Where,
v = shaft speed (velocity) relative to synchronous speed
θ = shaft, or power, angle (relative to synchronous reference frame)
Pshaft = shaft power
Pterm = terminal power out
D = damping constant
M = mass
Integration
Trapezoidal integration formula for θ, for example:

∆t  dθ dθ 
θn + 1 = θn +  + 
2  dt n dt n + 1

∆t = time step size

In the Predictor step, the integration routine captures the state variables and their derivatives
from the previous time step (moves the (n+1) values to the n values in the formulae above). Then
it makes a Predictor calculation of the derivatives at the new step based on the present values of
the state variables.
Based on the results of the Predictor step, a new guess at the derivatives at the present (n+1) step
is computed in the Corrector step. Then the guess at the state variables is corrected.
Only one Corrector step is executed. A circuit solution is performed between the predictor and
corrector steps and after the corrector step.
Circuit Solution in Dynamics Mode
The circuit solution proceeds more or less like it would for a normal power flow solution. The
Load elements are presently treated the same since load dynamics have not been implemented.
They are sinks or sources of power, depending on sign. However, the Generator, and any other
elements with dynamics models, such as Storage elements, will behave differently. The present
implementation of the Generator model computes its terminal currents using the Thevenin
equivalent voltage source, adjusted for the new phase angle, θ, computed by the integration of
the acceleration and speed of the angle according to the swing equation. In contrast, the currents

10253878 9-13
would be computed to meet a specified power and power factor in a power flow solution. Thus,
the power from the Generator will depend on system conditions with might be the result of a
disturbance (that’s usually why you would be simulating dynamics).
As a point of reference, in Harmonics mode, the Loadmodel option is set to Admittance and a
direct solution is performed. This is a non-iterative solution mode that treats loads and most PC
elements as constant impedances (admittances). However, in Dynamics mode, the load flow is
normally solved iteratively for each guess at the Generator voltages, maintain the nonlinearity of
Load elements.
Computing Terminal Currents in Dynamics Mode (Generator)
As an example of computing terminal currents, we will examine the default Generator object
algorithm. The electrical model for a 3-phase Generator object in Dynamics mode is illustrated
in Figure 9-4.

Phases Xd’

A Positive E1 ∠θ
Phase Sequence E1 
to
B Symmetrical
Component
Transformation
C
Negative
Xd”
Sequence

Figure 9-4
Default model of electrical part of generator object in dynamics mode

In any of the power flow modes (Snapshot, etc), the Generator object is treated as a power source
like most other power conversion elements. It is basically a negative Load element. Upon
entering Dynamics mode, the Generator is converted to a sort of Thevenin equivalent with a
positive-sequence voltage source behind a reactance of the form shown in Figure 9-4. Note that
this is a symmetrical component model with the generator part being represented by positive-
and negative-sequence networks as shown. The positive-sequence network consists of a voltage
source, E 1 ∠θ, behind the value specified for the Xdp property of the Generator model,
representing the transient reactance, Xd’. E 1 and θ are initialized so that the initial solution in
dynamics mode approximates the most recent power flow solution. Thus, simulations will be
better if the initializing power flow solution is reasonably well balanced. The generator
excitation is assumed to be only positive sequence.
The negative-sequence network consists of the value specified for the Xdpp property of the
Generator, which normally corresponds to Xd”. This value is usually closer to the actual negative
sequence reactance of a machine.

10253878 9-14
The terminal currents are computed as follows:
• The phase domain (ABC) voltages are transformed to symmetrical component (012)
voltages.
• The symmetrical component currents are computed based on the present value of E1 and θ.
• Assuming no exciter or governor action is modeled, E1 and the shaft input power remain
constant for the duration of the simulation. User-written modules could change these values.
• Finally, the symmetrical component currents are transformed back into the phase domain.
The terminal currents are used to compute the compensation currents that are used in the
circuit nodal admittance solution algorithm.

Integration for a Power Electronics Model Implementation

This section provides additional details on how to include power electronics models in user-
written DynaDLL for Storage elements. This information is primarily for software developers
and academic researchers who use OpenDSS. The algebraic equations in these models are
generally straightforward to compute; it is the control blocks that contain differential equations
that usually cause confusion so we’ll restrict the discussion here to that subject.
There are typically two types of differential equations in these kinds of models:
1. Simple integrator blocks typically used in the proportional-integral (PI) control loops, and
2. Time delay blocks.
1
Handling the 𝑠 Integrator Block
𝑑𝑑
This is the simple integrator block representing 𝑥(𝑡) = ∫ 𝑑𝑑 𝑑𝑑.

The generic algorithm for each integrator is:

Given
𝑑𝑑
𝑑𝑑
= 𝑥̇ 𝑛+1 = 𝑓(𝑥, 𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣, 𝑠𝑠𝑠𝑠𝑠𝑠) @ n+1 time step

and values at the previous time step, n: 𝑥𝑛 , 𝑥̇ 𝑛

Discretize and integrate with trapezoidal rule formula:
∆𝑡
𝑥𝑛+1 = 𝑥𝑛 + [𝑥̇ 𝑛+1 + 𝑥̇ 𝑛 ]
2

1
Handling the 𝐴𝐴+1 Exponential Time Delay Block
This is a simple exponential time delay block that results in a time domain equation of the form:

𝑥𝑜𝑜𝑜 = 𝑥𝑖𝑖 �1 − 𝑒 −𝑡/𝐴 �

10253878 9-15
The generic algorithm for this block is derived as follows:
𝑋𝑜𝑜𝑜 1
=
𝑋𝑖𝑖 𝐴𝐴 + 1
Rearranging:
𝑋𝑖𝑖 − 𝑋𝑜𝑜𝑜
𝑠𝑋𝑜𝑜𝑜 =
𝐴
In the time domain, and discretizing at the nth time step, the derivative of x(t) is computed by:
𝑥𝑖𝑖,𝑛 − 𝑥𝑜𝑜𝑜,𝑛
𝑥̇ 𝑜𝑜𝑜,𝑛 =
𝐴
That is, the derivative of the variable of interest – the output variable – is proportional to the
difference between the present value of the input and output. Once the derivative is computed, it
is integrated in the same manner as the derivatives for the simple integrator blocks.

10253878 9-16
10
REFERENCES AND BIBLIOGRAPHIES
References (1-4)
[1] Rylander, M., Smith, J., “Stochastic Approach for Distribution Planning with Distributed
Energy Resources,” Cigre Grid of the Future Symposium, Kansas City, MO, 2012.
[2] Analysis of High-Penetration Solar PV Impacts for Distribution Planning: Stochastic and
Time-Series Methods for Determining Feeder Hosting Capacity. EPRI, Palo Alto, CA:
2012. 1026640
[3] Distributed Photovoltaic Feeder Analysis: Preliminary Findings from Hosting Capacity
Analysis of 18 Distribution Feeders. EPRI, Palo Alto, CA: 2013. 3002001245.
[4] Alternatives to the 15% Rule: Modeling and Hosting Capacity Analysis of 16 Feeders.
EPRI, Palo Alto, CA: 2015. 3002005812.
[5] A New Method for Characterizing Distribution System Hosting Capacity for DER: A
Streamlined Approach for PV. EPRI, Palo Alto, CA: 2014. 3002003278.
[6] Rylander, M., Smith, J., Sunderman, W., “Streamlined Method For Determining
Distribution System Hosting Capacity”, 23rd International Conference on Electricity
Distribution, CIRED, Lyon, France, June 2015
[7] Rylander, M., Smith, J., Sunderman, W., “Streamlined Method For Determining
Distribution System Hosting Capacity”, IEEE Rural Electric Power Conference, Asheville,
NC, April 2015
[8] The Integrated Grid: A Benefit-Cost Framework,” EPRI, Palo Alto, CA. 3002004878,
2015.
[9] Distributed Generation – Integrated Grid Research Project, www.tva.gov.
[10] Distribution Feeder Hosting Capacity: What Matters When Planning for DER?. EPRI, Palo
Alto, CA: 2015. 3002004777
[11] EPRI, 2013, OpenDSS Computer Program, Open-Source Distribution System Simulator,
Available on Sourceforge.Net, Ver 7.6.4.
[12] EPRI, 2012, Analysis of Distribution System Effects of Energy Storage Through
Simulation and Modeling, Palo Alto, CA, US. 1024285.
[13] G. Delille, G. Malarange, B. François, 2012, “Dynamic Frequency Control Support by
Energy Storage to Reduce the Impact of Wind and Solar Generation on Isolated Power
System's Inertia”, IEEE Transactions on Sustainable Energy, vol. 3, issue 4, pp. 931-939.
[14] R. Dugan, J. Taylor, G. DeLille, “Storage simulations for distribution system analysis“
CIRED 2013, Stockholm, Paper No. 1340.

10253878 10-1
[15] EPRI, 2011, EPRI Smart Grid Demonstration Initiative – Three-Year Update, Palo Alto,
CA, US. 1023411.
[16] EPRI, 2011 Understanding Energy Storage Solutions and Capabilities on Utility
Distribution Systems:Analysis of Grid Effect of Energy Storage Through Modeling and
Simulation Activities. Palo Alto, CA. 1021938.
[17] "Simulation of Arc Furnace Power Systems," R. C. Dugan, IEEE IAS Annual Conference,
Los Angeles, Oct 1977. Later published in the IEEE Transactions on Industry
Applications, Nov/Dev 1980, pp 813-818.
[18] Roger C. Dugan, Mark F. McGranaghan, Surya Santoso, H. Wayne Beaty, Electrical Power
Systems Quality, 3rd Edition, McGraw-Hill, New York, 2012.
[19] Roger Dugan, "Load Modeling in Harmonics Analysis with OpenDSS,"
http://svn.code.sf.net/p/electricdss/code/trunk/Doc/Harmonics%20Load%20Modeling.docx
, January 2015
[20] H.E. Kent and P.W. Blye, "Inductive Co-ordination with Series Sodium Highway Lighting
Circuits," AIEE Transactions, July 1939, pp.329ff.
[21] Optimal Network Reconfiguration for Electric Distribution Systems. EPRI, Palo Alto, CA:
2011. 102205.
[22] Automated Distribution Automation Switch Placement. EPRI, Palo Alto, CA: 2014.
3002003238.
[23] T. A. Short, Distribution Reliability and Power Quality, CRC Press, 2005.

References (5-8)
[24] GRIDON® Transformer, http://www.reinhausen.com/, Maschinenfabrik Reinhausen Gmb,
93059 Regensburg Germany
[25] V. Miranda and L.M. Proenca, "Why Risk Analysis Outperforms Probabilistic Choice as
the Effective Decision Support Paradigm for Power System Planning" IEEE Transactions
on Power Systems, Vol. 13, No. 2, 1998, pp. 643-648.
[26] E. Crousillat, P. Doerfner, P Alvarado, H. Merrill, “Conflicting Objectives and Risk in
Power System Planning,” IEEE Transactions on Power Systems, Vol. 8, No. 3, August
1993.
[27] D. S. Popovic, Z. N. Popovic, “A risk management procedure for supply restoration in
distribution networks”, IEEE Transactions on Power Systems, vol. 19, n. 1, p. 221–228,
Feb 2004.
[28] M. El-Hawary, Electric Power Applications of Fuzzy Systems, IEEE Press, New York,
1998.
[29] J. T. Saraiva, V. Miranda, L. M. V. G. Pinto, “Impact on some planning decisions from a
fuzzy modelling of power systems”, IEEE Transactions on Power Systems, vol. 9, n. 2, p.
819–825, May 1994.

10253878 10-2
[30] M. T. P. de Leão, M. A. Matos, “Distribution planning with fuzzy loads and independent
generation”, Electricity Distribution. Part 1: Contributions. CIRED. 14th International
Conference and Exhibition on (IEE Conf. Publ. No. 438), vol. 6, p. 12/1–12/5, 1997.
[31] V. Miranda, J. V. Ranito, L. M. Proenca, “Genetic algorithms in optimal multistage
distribution network planning”, IEEE Transactions on Power Systems, vol. 9, n. 4, p.
1927–1933, Nov 1994.
[32] S. K. Goswami, “Distribution system planning using branch exchange technique”, IEEE
Transactions on Power Systems, vol. 12, n. 2, p. 718–723, May 1997.
[33] H. Kuwabara, K. Nara, “Multi-year and multi-state distribution systems expansion
planning by multi-stage branch exchange”, IEEE Transactions on Power Delivery, vol. 12,
n. 1, p. 457–463, Jan 1997.
[34] R. Dugan and M. Waclawiak, “Using Energy as a Measure of Risk in Distribution
Planning,” CIRED 2007, Vienna, Paper 0822.
[35] R. Dugan and M. Waclawiak, “Weather Normalized Load Forecasting For Distribution
Planning,” CIRED 2007, Vienna, Paper 0722.
[36] H. L. Willis, L. A. Finley, M. J. Buri, “Forecasting electric demand of distribution system
planning in rural and sparsely populated regions”, IEEE Transactions on Power Systems,
vol. 10, n. 4, p. 2008–2013, Nov 1995.
[37] R. C. Dugan and S.K. Price, “Including Distributed Resources In Distribution Planning,”
IEEE PES Power Systems Conference and Exposition, 10-13 Oct. 2004.
[38] Roger C. Dugan, “Computing Incremental Capacity Provided by Distributed Resources for
Distribution Planning,” IEEE PES General Meeting, Conference Proceedings, Tampa, FL,
24-28 June, 2007.
[39] Rylander, M., Smith, J., “Stochastic Approach for Distribution Planning with Distributed
Energy Resources,” Cigre Grid of the Future Symposium, Kansas City, MO, 2012.
[40] PT Load Version 6.1 Users Manual: Power Transformer Loading Program, EPRI, Palo
Alto CA: 2002. 1007083
[41] Green Circuits: Distribution Efficiency Case Studies. EPRI, Palo Alto, CA: 2011. 1023518.
[42] G.T. Heydt, Computer Analysis methods for Power Systems, MacMillan, 1986, Chapter 4,
Section 2
[43] B. Stott and O. Alsac, ”Fast Decoupled Load Flow," IEEE Transactions on Power
Apparatus and Systems, Vol PAS-93, May-June 1974, pp. 859-869.
[44] W. Kersting, Distribution System Modeling and Analysis, 3rd Edition, CRC Press, 2012.
[45] IEEE Test Feeders, http://www.ewh.ieee.org/soc/pes/dsacom/testfeeders/index.html.
[46] EPRI, 2018, OpenDSS Computer Program, Open-Source Distribution System Simulator,
Available on Sourceforge.Net, Ver 7.6.5.

10253878 10-3
[47] Guideline for Reliability Assessment and Reliability Planning – Evaluation of Tools for
Reliability Planning, EPRI, Palo Alto, CA: 2006. 1012450.
[48] Development of Distribution System Reliability and Risk Analysis Models, Volume 2,
EPRI, August 1981. Report EPRI EL-2018.
[49] IEEE Guide for Electric Power Distribution Reliability Indices, IEEE Std. 1366-2012.
[50] T. A. Short, Electric Power Distribution Handbook, Second Edition, CRC Press, 2014.
[51] Reliability Benchmarking Methodology, EPRI, 1997, Report EPRI TR-107938.
[52] An Assessment of Distribution System Power Quality, Vol. 2: Statistical Summary Report.
EPRI, Palo Alto, CA: January 1996. Report EPRI TR 106294
[53] S. R. Gilligan, “A Method for Estimating the Reliability of Distribution Circuits,” IEEE
Transactions on Power Delivery, Vol. 7, No. 2, April 1992, pp 694-698.
[54] G. Kjølle and Kjell Sand, “RELRAD – An Analytical Approach for Distribution System
Reliability Assessment,” IEEE Transactions on Power Delivery, Vol 7. No. 2, April 1992,
pp. 809-814.
[55] R. E. Brown, H. V. Nguyen, and J. J. Burke, “A Systematic and Cost Effective Method to
Improve Distribution System Reliability,” 1999 IEEE PES Summer Meeting, Vol 2, pp
1037-1042.
[56] T. Gonen, Electric Power Distribution System Engineering, McGraw-Hill Book Company,
NY, 1986.
[57] R. Billinton, W., Wangdee, “Predicting bulk electricity system reliability performance
indices using sequential Monte Carlo simulation,” IEEE Transactions on Power Delivery,
Volume 21, Issue 2, April 2006, pp 909-917.
[58] Engineering Guide for Integration of Distributed Generation and Storage into Power
Distribution Systems, EPRI, Palo Alto, CA: 2000. 1000419
[59] Power Quality for Distribution Planning,EPRI, Palo Alto, CA. April 1998. EPRI Report
TR-110346.
[60] Thomas E. McDermott, Roger C. Dugan, “Distributed Generation Impact on Reliability
and Power Quality Indices,” IEEE 2002 Rural Electric Power Conference Proceedings,
IEEE Catalog No. 02CH37360, Colorado Springs, May 2002, Paper No. D3.
[61] J. H. McWhirter, et. al., "Determination of Impulse Stress Within Transformer Windings
By Computers," AIEE Transactions, pt. III, Vol 75, 1956, pp. 1267-1273.
[62] R. Dugan, “A Perspective On Transformer Modeling for Distribution System Analysis,”
IEEE PES General Meeting, 2003.
[63] R. Horton, W. Sunderman, R. Arritt, and R. Dugan, “Effect of Line Modeling Methods on
Neutral-to-Earth Voltage Analysis of Multi-Grounded Distribution Feeders.” IEEE PES
Power Systems Conference and Exposition (PSCE), 2011.

10253878 10-4
[64] A. Ballanti, R. Dugan, Cable Modeling in OpenDSS,
https://sourceforge.net/p/electricdss/code/HEAD/tree/trunk/Distrib/Doc/TechNote%20Cabl
eModelling.pdf , 2015
[65] Okonite. (2016). 15kV to 35kV Underground Primary Distribution Cable Jacketed — Red
Identification Stripes. Available: http://okonite.com/Product_Catalog/section2/section2-
pdfs/2-42.pdf
[66] Okonite. (2016). Okoguard®-Okoseal® Type MV-105 15kV Shielded Power Cable.
Available: http://okonite.com/Product_Catalog/section2/section2-pdfs/2-8.pdf
[67] Jos. Arrillaga and Neville R. Watson, Power System Harmonics, 2nd edition, Chapter 7,
John Wiley and Sons, Ltd, 2004.
[68] EPRI, 2012, Analysis of Distribution System Effects of Energy Storage Through
Simulation and Modeling, Palo Alto, CA, US. 1024285.
[69] M. Wakefield, G. Horst, J. Simmins, J. Green, B. Green, 2012, "EPRI Smart Grid
Demonstration 4 Year Update Case Study Brief", EPRI, Palo Alto, CA 2012, Product ID
1025781
[70] G. Delille, G. Malarange, B. François, 2012, “Dynamic Frequency Control Support by
Energy Storage to Reduce the Impact of Wind and Solar Generation on Isolated Power
System's Inertia”, IEEE Transactions on Sustainable Energy, vol. 3, issue 4, pp. 931-939.
[71] R. Dugan, J. Taylor, G. DeLille, “Storage simulations for distribution system analysis“
CIRED 2013, Stockholm, Paper No. 1340.
[72] Understanding Energy Storage Solutions and Capabilities on Utility Distribution
Systems:Analysis of Grid Effect of Energy Storage Through Modeling and Simulation
Activities. Palo Alto, CA: 2011.. 1021938.
[73] Analysis of Distribution System Effects of Energy Storage Through Simulation and
Modeling: Energy Storage Grid Integration Analysis with OpenDSS EPRI, Palo Alto, CA:
2012. 1024016.
[74] Roger Dugan, "Load Modeling in Harmonics Analysis with OpenDSS,"
http://svn.code.sf.net/p/electricdss/code/trunk/Doc/Harmonics%20Load%20Modeling.docx
, January 2015
[75] Optimal Network Reconfiguration for Electric Distribution Systems. EPRI, Palo Alto, CA:
2011. 102205.
[76] Automated Distribution Automation Switch Placement. EPRI, Palo Alto, CA: 2014.
3002003238.
[77] “Analysis of Distribution System Effects of Energy Storage through Simulation and
Modeling,” Electric Power Research Institute, Palo Alto, CA, EPRI Technical Report
1024285, Dec. 2012.

10253878 10-5
[78] J. Fortmann and R. Hendricks, “Towards a standardized binary simulation model
interface,” Repower Systems AG and Siemens AG, 2011.
[79] J. Mahseredjian. “DLL programming in EMTP,” EMTP-EMTPWorks, 10/24/3016,
pp 16-18.

Bibliography

Distribution Planning
Roger C. Dugan, Mark F. McGranaghan, Surya Santoso, H. Wayne Beaty, Electrical Power
Systems Quality, 3rd Edition, McGraw-Hill, New York, 2012.
T. A. Short, Distribution Reliability and Power Quality, CRC Press, 2005.
Analysis of High-Penetration Solar PV Impacts for Distribution Planning: Stochastic and Time-
Series Methods for Determining Feeder Hosting Capacity. EPRI, Palo Alto, CA: 2012. 1026640
Distributed Photovoltaic Feeder Analysis: Preliminary Findings from Hosting Capacity Analysis
of 18 Distribution Feeders. EPRI, Palo Alto, CA: 2013. 3002001245.
Alternatives to the 15% Rule: Modeling and Hosting Capacity Analysis of 16 Feeders. EPRI,
Palo Alto, CA: 2015. 3002005812.
A New Method for Characterizing Distribution System Hosting Capacity for DER: A
Streamlined Approach for PV. EPRI, Palo Alto, CA: 2014. 3002003278.
Rylander, M., Smith, J., Sunderman, W., “Streamlined Method For Determining Distribution
System Hosting Capacity”, 23rd International Conference on Electricity Distribution, CIRED,
Lyon, France, June 2015
Rylander, M., Smith, J., Sunderman, W., “Streamlined Method For Determining Distribution
System Hosting Capacity”, IEEE Rural Electric Power Conference, Asheville, NC, April 2015
The Integrated Grid: A Benefit-Cost Framework,” EPRI, Palo Alto, CA. 3002004878, 2015.
Distribution Feeder Hosting Capacity: What Matters When Planning for DER?. EPRI, Palo Alto,
CA: 2015. 3002004777
M. Blanchard, et al, “Experience with optimization software for distribution system planning”,
IEEE Transactions on Power Systems, vol. 11, n. 4, p. 1891–1898, Nov 1996.
G. Brauner, M. Zobel, “Knowledge based planning of distribution networks”, IEEE Transactions
on Power Systems, vol. 9, n. 2, p. 942–948, May 1994.
J. G. Dalton, D. L. Garrison, C.M. Fallon, “Value-based reliability transmission planning”, IEEE
Transactions on Power Systems, vol. 11, n. 3, p. 1400–1408, Aug 1996.
R. C. Dugan, et al, “Using voltage sag and interruption indices in distribution planning”, IEEE
Power Engineering Society 1999 Winter Meeting, vol. 2, p. 1164–1169, 1999.
L. A. F. M. Ferreira, P. M. S. Carvalho, L. M. F. Barruncho, “An evolutionary approach to
decision-making in distribution planning”, Electricity Distribution. Part 1: Contributions.

10253878 10-6
CIRED. 14th International Conference and Exhibition on (IEE Conf. Publ. No. 438), vol. 6, p.
6/1–6/5, 1997.
Y. Fukuyama, H.-D. Chiang, “A parallel genetic algorithm for generation expansion planning”,
IEEE Transactions on Power Systems, v. 11, n. 2, p. 955–961, May 1996.
L. Goel, R, Billinton, “Determination of reliability worth for distribution system planning”, IEEE
Transactions on Power Delivery, vol. 9, n. 3, p. 1577–1583, Jul 1994.
B. F. Hobbs, H. B. Rouse, D. T. Hoog, “Measuring the economic value of demand-side and
supply resources in integrated resource planning models”, IEEE Transactions on Power Systems,
vol. 8, n. 3, p. 979–987, Aug 1993.
S. K. Khator, L. C. Leung, “Power distribution planning: a review of models and issues”, IEEE
Transactions on Power Systems, vol. 12, n. 3, p. 1151–1159, Aug 1997.
G. Latorre-Bayona, I. J. Perez-Arriaga, “Chopin, a heuristic model for long term transmission
expansion planning”, IEEE Transactions on Power Systems, vol. 9, n. 4, p. 1886–1894, Nov
1994.
C. N. Macqueen, et al, “An integrated software environment for distribution system planning”,
Power Industry Computer Application Conference, Conference Proceedings, p. 162–168, 1993.
N. S. Rau; Y-H. Wan, “Optimum location of resources in distributed planning”, IEEE
Transactions on Power Systems, vol. 9, n. 4, p. 2014–2020, Nov 1994.
Tang, “Power distribution system planning with reliability modeling and optimization”, IEEE
Transactions on Power Systems, vol. 11, n. 1, p. 181–189, Feb 1996.
P. Paliwal, N. Patidar, R. Nema, “Planning of grid integrated distributed generators: A review of
technology, objectives and techniques”, Renewable and Sustainable Energy Reviews, vol. 40,
n.1, pp.557-570, 2014.
C. Y. Teo, S. Feng, “Visually integrated modeling for distribution network operation and
planning”, IEEE Power Engineering Society Winter Meeting, vol. 2, p. 937–941, 2000.
M. Yehia, et al, “A global planning methodology for uncertain environments: application to the
Lebanese power system”, IEEE Transactions on Power Systems, vol. 10, n. 1, p. 332–338, Feb
1995.
W.-S. Tan, M. Y. Hassan, M. S. Majid, H. A. Rahman, “Optimal distributed renewable
generation planning: A review of different approaches”, Renewable and Sustainable Energy
Reviews, vol. 18, p. 626 – 645, 2013.
J. Millan, R. A. Campo, G. A. Sanchez-Sierra, “A modular system for decision-making support
in generation expansion planning (SUPER)”, IEEE Transactions on Power Systems, vol. 13, n. 2,
p. 667–671, May 1998.
M. E. Baran, F. F. WU, “Optimal sizing of capacitors placed on a radial distribution system”,
IEEE Transactions on Power Delivery, vol. 4, n. 1, p. 735–743, Jan 1989.

10253878 10-7
T. -H. Chen, J.-T. Cherng, “Optimal phase arrangement of distribution transformers connected to
a primary feeder for system unbalance improvement and loss reduction using a genetic
algorithm”, IEEE Transactions on Power Systems, vol. 15, n. 3, p. 994–1000, Aug 2000.
P. Chiradeja, R. Ramakumar, “An approach to quantify the technical benefits of distributed
generation”, IEEE Transactions on Energy Conversion, vol. 19, n. 4, p. 764–773, Dec 2004.
N. Hadjsaid, J. F. Canard, F. Dumas, “Dispersed generation impact on distribution networks”,
IEEE Computer Applications in Power, vol. 12, n. 2, p.22–28, Apr 1999.
J. P. Lopes, N. Hatziargyriou, J. Mutale, P. Djapic, N. Jenkins, “Integrating distributed
generation into electric power systems: A review of drivers, challenges and opportunities”,
Electric Power Systems Research, vol. 77, n. 9, p. 1189 – 1203, 2007.
P. R. Macgregor, H. B. Puttgen, “The integration of nonutility generation and spot prices within
utility generation scheduling”, IEEE Transactions on Power Systems, vol. 9, n. 3, p. 1302–1308,
Aug 1994.
E. G. Neudorf, et al, “Cost-benefit analysis of power system reliability: two utility case studies”,
IEEE Transactions on Power Systems, vol. 10, n. 3, p. 1667–1675, Aug 1995.
R.-H. Liang, Y.-S. Wang, “Fuzzy-based reactive power and voltage control in a distribution
system”, IEEE Transactions on Power Delivery, vol. 18, n. 2, p. 610–618, April 2003.
M. L. Baughman, S. N. Siddiqi, J. W. Zarnikau, “Integrating transmission into irp part i:
analytical approach”, IEEE Transactions on Power Systems, vol. 10, n. 3, p. 1652–1659, Aug
1995.
R. Orans, C.-K. Woo, B. K. Horii, “Case study: Targeting demand-side management for
electricity transmission and distribution benefits”, Managerial and Decision Economics, John
Wiley & Sons, Ltd., vol. 15, n. 2, p. 169–175, 1994.
G. J. Peponis, M. P. Papadopulos, N. D. Hatziargyriou, “Optimal operation of distribution
networks”, IEEE Transactions on Power Systems, vol. 11, n. 1, p. 59–67, Feb 1996.
T. M. Peng, N. F. Hubele, G. G. Karady. “An adaptive neural network approach to one-week
ahead load forecasting”, IEEE Transactions on Power Systems, vol. 8, n. 3, p. 1195–1203, Aug
1993.
D. I. H. Sun, et al, “Calculation of energy losses in a distribution system”, IEEE Transactions on
Power Apparatus and Systems, PAS-99, n. 4, p. 1347–1356, July 1980.
T. Ackermann, G. Anderson, L. Söder, “Distributed Generation: a definition”, Electric Power
Systems Research, vol. 57, pp. 195-204, 2001.
S. J. Wang, S. M. Shahidehpour, N.-D. Xiang, “Power systems marginal cost curve and its
applications”, IEEE Transactions on Power Systems, vol. 10, n. 3, p. 1321–1328, Aug 1995.
S. M. Rios, V. P. Vidal, D. L. Kiguel, “Bus-based reliability indices and associated costs in the
bulk power system”, IEEE Transactions on Power Systems, vol. 13, n. 3, p. 719–724, Aug 1998.

10253878 10-8
C. M. K. Porter, M. Weisburger, “Review of International Experience Integrating Variable
Renewable Energy Generation”, 2007.
P. P. Barker, R. W. D. Mello, “Determining the impact of distributed generation on power
systems. I. radial distribution systems”, IEEE Power Engineering Society Summer Meeting, vol.
3, p. 1645–1656, 2000.
I. J. Ramirez-Rosado, J. L. Bernal-Agustin, “Genetic algorithms applied to the design of large
power distribution systems”, IEEE Transactions on Power Systems, vol. 13, n. 2, p. 696–703,
May 1998.
G. B. Sheble, K. Brittig, “Refined genetic algorithm-economic dispatch example”, IEEE
Transactions on Power Systems, vol. 10, n. 1, p. 117–124, Feb 1995.
“Proposed Definitions of Terms for Reporting and Analyzing Outages of Electrical Transmission
and Distribution Facilities and Interruptions”, IEEE Transactions on Power Apparatus and
Systems, PAS-87, n. 5, p. 1318–1323, May 1968.
X. Wang, N. N. Schulz, S. Neumann, “CIM extensions to electrical distribution and CIM CML
for the IEEE radial test feeders”, IEEE Transactions on Power Systems, vol. 18, n. 3, p. 1021–
1028, Aug 2003.
American National Standard for Electric Power-Systems and Equipment Voltage Ratings (60
Hertz), ANSI C84.1-1995, National Electric Manufacturers Association, Rosslyn, Virginia,
1996.

Power Flow
D. R. R. Penido, L. R. de Araujo, S. Carneiro, “Three-phase power flow based on four-conductor
current injection method for unbalanced distribution networks”, IEEE Transactions on Power
Systems, vol. 23, n. 2, p. 494–503, May 2008.
Maya K. N., Jasmin E.A., “A three phase power flow algorithm for distribution network
incorporating the impact of distributed generation models”, Procedia Technology, vol. 21, p. 326
– 331, 2015.
J. Yuntao, W. Wenchuan, Z. Boming, S. Hongbin, “An extension of fbs three-phase power flow
for handling pv nodes in active distribution networks”, IEEE Transactions on Smart Grid, vol. 5,
n. 4, p.1547–1555, July 2014.
P. A. N. Garcia, J. L. R. Pereira, S. Carneiro, V. M. da Costa, N. Martins, “Three-phase power
flow calculations using the current injection method”, IEEE Transactions on Power Systems, vol.
15, n. 2, p. 508–514, May 2000.
P. A. N. Garcia, J. L. R. Pereira, S. Carneiro, “Voltage control devices models for distribution
power flow analysis”, IEEE Transactions on Power Systems, vol. 16, n. 4, p. 586–594, Nov
2001.
H. W. Dommel, W. F. Tinney, “Optimal power flow solutions”, IEEE Transactions on Power
Apparatus and Systems, PAS-87, n. 10, p. 1866–1876, Oct 1968.

10253878 10-9
R. M. Ciric, A. P. Feltrin, L. F. Ocha, “Power flow in four-wire distribution networks-general
approach”, IEEE Transactions on Power Systems, vol. 18, n. 4, p. 1283–1290, Nov 2003.
V. M. da Costa, N. Martins, J. L. R. Pereira, “Developments in the newton raphson power flow
formulation based on current injections”, IEEE Transactions on Power Systems, vol. 14, n. 4, p.
1320 – 1326, Nov 1999.
C. S. Cheng, D. Shirmohammadi, “A three-phase power flow method for real-time distribution
system analysis”, IEEE Transactions on Power Systems, vol. 10, n. 2, p. 671–679, May 1995.
T. -H. Chen, M. -S. Chen, K. -J. Hwang, P. Kotas, E. A. Chebli, “Distribution system power
flow analysis-a rigid approach”, IEEE Transactions on Power Delivery, vol. 6, n. 3, p. 1146–
1152, Jul 1991.
T.-H Chen, Y.-L Chang, “ Integrated models of distribution transformers and their loads for
three-phase power flow analyses”, IEEE Transactions on Power Delivery, vol. 11, n. 1, p. 507–
513, Jan 1996.
S. Carneiro, J. L. R., P. A. N. Garcia. “Unbalanced distribution system power flow using the
current injection method”, IEEE Power Engineering Society Winter Meeting, vol. 2, p. 946–950,
2000.
X.-P. Zhang, “Continuation power flow in distribution system analysis”, IEEE PES Power
Systems Conference and Exposition, p. 613–617, 2006.
X.-P. Zhang, P. Ju, E. Handschin, “Continuation three-phase power flow: A tool for voltage
stability analysis of unbalanced three-phase power systems”, IEEE Transactions on Power
Systems, vol. 20, n. 3, p. 1320–1329, Aug 2005.
R. D. Zimmerman, H.-D. Chiang, “Fast decoupled power flow for unbalanced radial distribution
systems”, IEEE Transactions on Power Systems, vol. 10, n. 4, p.2045–2052, Nov 1995.
Z. Wang, F. Chen, J. Li, “Implementing transformer nodal admittance matrices into
backward/forward sweep-based power flow analysis for unbalanced radial distribution systems”,
IEEE Transactions on Power Systems, vol. 19, n. 4, p. 1831–1836, Nov 2004.
W. F. Tinney, C. E. Hart, “Power flow solution by newton’s method”, IEEE Transactions on
Power Apparatus and Systems, PAS-86, n. 11, p. 1449–1460, Nov 1967.
Tong, S.; Miu, K. N. “A network-based distributed slack bus model for DGs in unbalanced
power flow studies”, IEEE Transactions on Power Systems, vol. 20, n. 2, p. 835–842, May 2005.
J.-H. Teng, “A Network-Topology-based Three-Phase Power Flow for Distribution Systems”,
Proceedings of National Science Council ROC(A), vol.24, no.4, pp.259-264, 2000.
H. Sun, D. C. Yu, Y. Xie, “Application of fuzzy set theory to power flow analysis with uncertain
power injections”, Power Engineering Society Winter Meeting, v. 2, p. 1191–1196 vol. 2, 2000.
D. I. Sun, et al, “Optimal power flow by newton approach”, IEEE Power Engineering Review,
PER-4, n. 10, p. 39–39, Oct 1984.

10253878 10-10
D. Shirmohammadi, H. W. Hong, A. Semlyen, G. X. Luo, “A compensation-based power flow
method for weakly meshed distribution and transmission networks”, IEEE Transactions on
Power Systems, vol. 3, n. 2, p. 753–762, May 1988.
W.-M. Lin, Y.-S. Su, H.-C. Chin, J.-H. Teng, “Three-phase unbalanced distribution power flow
solutions with minimum data preparation”, IEEE Transactions on Power Systems, vol. 14, n. 3,
p. 1178–1183, Aug 1999.
G. X. Luo, A. Semlyen, “Efficient load flow for large weakly meshed networks”, IEEE
Transactions on Power Systems, vol. 5, n. 4, p. 1309–1316, Nov 1990.
Y. P. Dusonchet, et al, “Load flows using a combination of point jacobi and newton’s methods”,
IEEE Transactions on Power Apparatus and Systems, PAS-90, n. 3, p. 941–949, May 1971.
R. Cicoria, et al, “Load flow calculations on distribution networks by using a statistical
approach”, Electricity Distribution. Part 1: Contributions. CIRED. 14th International Conference
and Exhibition on (IEE Conf. Publ. No. 438), vol. 6, p. 13/1–13/5, 1997.
D. Das, D. P. Kothari, A. Kalam, “Simple and efficient method for load flow solution of radial
distribution networks”, International Journal of Electrical Power & Energy Systems, vol. 17, n.
5, p. 335 – 346, 1995.
R. Berg, E. S. Hawkins, W. W. Pleines, “ Mechanized calculation of unbalanced load flow on
radial distribution circuits”, IEEE Transactions on Power Apparatus and Systems, PAS-86, n. 4,
p. 415–421, April 1967.
T. L. Baldwin, S. A. Lewis, “Distribution load flow methods for shipboard power systems”,
IEEE Transactions on Industry Applications, vol. 40, n. 5, p.1183–1190, Sept 2004.
J.-H. Teng, C.-Y Chang, “A novel and fast three-phase load flow for unbalanced radial
distribution systems”, IEEE Transactions on Power Systems, vol. 17, n. 4, p. 1238–1244, Nov
2002.
J.-H. Teng, “A direct approach for distribution system load flow solutions”, IEEE Transactions
on Power Delivery, vol. 18, n. 3, p. 882–887, July 2003.
B. Stott, O. Alsac, “Fast decoupled load flow”, IEEE Transactions on Power Apparatus and
Systems, PAS-93, n. 3, p. 859–869, May 1974.
M. S. Srinivas, “Distribution load flows: a brief review”, IEEE Power Engineering Society
Winter Meeting, vol. 2, p. 942–945, 2000.
J. Nanda, M. Kothari, M. S. Srinivas, “On some aspects of distribution load flow”, IEEE Region
10 International Conference on Global Connectivity in Energy, Computer, Communication and
Control, vol. 2, p. 510–513, 1998.
Y.-H. Moon, et al., “Fast and reliable distribution system load flow algorithm based on the Ybus
formulation”, IEEE Power Engineering Society Summer Meeting, vol. 1, p. 238–242, 1999.

10253878 10-11
M. Abdel-Akher, K. M. Nor, A. H. A Rashid, “Improved three-phase power-flow methods using
sequence components”, IEEE Transactions on Power Systems, vol. 20, n. 3, p. 1389–1397, Aug
2005.
D. M. Anderson, B. F. Wollenberg, “Solving for three phase conductively isolated bus bar
voltages using phase component analysis”, IEEE Transactions on Power Systems, vol. 10, n. 1,
p. 98–108, Feb 1995.
B. A. Carre, “Solution of load-flow problems by partitioning systems into trees”, IEEE
Transactions on Power Apparatus and Systems, PAS-87, n. 11, p. 1931–1938, Nov 1968.
R. G. Cespedes, “New method for the analysis of distribution networks”, IEEE Transactions on
Power Delivery, vol. 5, n. 1, p. 391–396, Jan 1990.
R. C. Dugan, T. E. Mcdermott, “An open source platform for collaborating on smart grid
research”, 2011 IEEE Power and Energy Society General Meeting, p. 1–7, 2011.
L. L. Freris, A. M. Sasson, “Investigation of the load-flow problem”, Electrical Engineers,
Proceedings of the Institution of, 1968, vol 115, n. 10, p. 1459–1470, October 1968.
W. H. Kersting, “The simulation of loop flow in radial distribution analysis programs”, IEEE
Transactions on Industry Applications, vol. 51, n. 2, p. 1928–1932, March 2015.
K. M. Nor, H. Mokhlis, T. A. Gani, “Reusability techniques in load-flow analysis computer
program”, IEEE Transactions on Power Systems, vol. 19, n. 4, p.1754–1762, Nov 2004.
Y. Wallach, “Gradient methods for load-flow problems”, IEEE Transactions on Power
Apparatus and Systems, PAS-87, n. 5, p. 1314–1318, May 1968.
P. Yan, A. Sekar, “Analysis of radial distribution systems with embedded series facts devices
using a fast line flow-based algorithm”, IEEE Transactions on Power Systems, vol. 20, n. 4, p.
1775–1782, Nov 2005.
D. A. Wiegmann, G. R. Essenberg, T. J. Overbye, Y. Sun, “Human factor aspects of power
system flow animation”, IEEE Transactions on Power Systems, vol. 20, n. 3, p. 1233–1240, Aug
2005.
M. E. Baran, I. EI-Markaby, “ Fault analysis on distribution feeders with distributed generators”,
IEEE Power Engineering Society General Meeting, p. 1, 2006.

Modeling
Y.-J. Wang, M.-J. Yang, “Probabilistic modeling of three-phase voltage unbalance caused by
load fluctuations”, IEEE Power Engineering Society Winter Meeting, vol. 4, p. 2588–2593,
2000.
J. W. Smith, R. Dugan, W. Sunderman, “Distribution modeling and analysis of high penetration
PV”, 2011 IEEE Power and Energy Society General Meeting, p. 1–7, 2011.
R. E. Brown, T. M. Taylor, “Modeling the impact of substations on distribution reliability”,
IEEE Power Engineering Society 1999 Winter Meeting, vol. 2, p. 889, 1999.

10253878 10-12
A. Capasso, et al, “A bottom-up approach to residential load modelling”, IEEE Transactions on
Power Systems, vol. 9, n. 2, p. 957–964, May 1994.
M. E. Baran, E. A. Staton, “Distribution transformer models for branch current based feeder
analysis”. IEEE Transactions on Power Systems, vol. 12, n. 2, p. 698–703, May 1997.
T. -H. Chen, M. -S. Chen, T. Inoue, P. Kotas, E. A. Chebli, “Three-phase cogenerator and
transformer models for distribution system analysis”, IEEE Transactions on Power Delivery, vol.
6, n. 4, p. 1671–1681, Oct 1991.
W. H. Dickinson, “A new method of electric load data calculations using a digital computer”,
IEEE Transactions on Industry and General Applications, IGA-2, n. 5, p. 393–401, Sept 1966.
T. Frantz, et al, “Load behavior observed in lilco and rg e systems”, IEEE Transactions on
Power Apparatus and Systems, PAS-103, n. 4, p. 819–831, April 1984.
C. W. Gellings, R. W. Taylor, “Electric load curve synthesis - a computer simulation of an
electric utility load shape”, IEEE Transactions on Power Apparatus and Systems, PAS-100, n. 1,
p. 60–65, Jan 1981.
W. H. Kersting, “Radial distribution test feeders”, IEEE Transactions on Power Systems, vol. 6,
n. 3, p. 975–985, Aug 1991.
W. H. Kersting, W. H. Phillips, “Distribution feeder line models”, IEEE Transactions on
Industry Applications, 1995. vol. 31, n. 4, p. 715–720, Jul 1995.
L. M. Popovic, “Practical method for evaluating ground fault current distribution in station,
towers and ground wire”, IEEE Transactions on Power Delivery, vol. 13, n. 1, p. 123–128, Jan
1998.
D. S. Popovic, E. Varga, Z. Perlic, “Extension of the common information model with a catalog
of topologies”, IEEE Transactions on Power Systems, vol. 22, n. 2, p. 770–777, May 2007.
A. M. Stankovic, B. C. Lesieutre, “Parametric variations in dynamic models of induction
machine clusters”, IEEE Transactions on Power Systems, vol. 12, n. 4, p. 1549–1554, Nov 1997.

10253878 10-13
10253878
10253878
Export Control Restrictions The Electric Power Research Institute, Inc.
Access to and use of EPRI Intellectual Property is granted (EPRI, www.epri.com) conducts research and
with the specific understanding and requirement that development relating to the generation, delivery
responsibility for ensuring full compliance with all applicable and use of electricity for the benefit of the public.
U.S. and foreign export laws and regulations is being
An independent, nonprofit organization, EPRI
undertaken by you and your company. This includes an
brings together its scientists and engineers as well
obligation to ensure that any individual receiving access
hereunder who is not a U.S. citizen or permanent U.S. as experts from academia and industry to help
resident is permitted access under applicable U.S. and address challenges in electricity, including
foreign export laws and regulations. In the event you are reliability, efficiency, affordability, health, safety and
uncertain whether you or your company may lawfully obtain the environment. EPRI members represent 90% of
access to this EPRI Intellectual Property, you acknowledge
the electric utility revenue in the United States with
that it is your obligation to consult with your company’s legal
international participation in 35 countries. EPRI’s
counsel to determine whether this access is lawful. Although
EPRI may make available on a case-by-case basis an principal offices and laboratories are located in
informal assessment of the applicable U.S. export Palo Alto, Calif.; Charlotte, N.C.; Knoxville, Tenn.;
classification for specific EPRI Intellectual Property, you and and Lenox, Mass.
your company acknowledge that this assessment is solely
for informational purposes and not for reliance purposes. Together…Shaping the Future of Electricity
You and your company acknowledge that it is still the
obligation of you and your company to make your own
assessment of the applicable U.S. export classification and
ensure compliance accordingly. You and your company
understand and acknowledge your obligations to make a
prompt report to EPRI and the appropriate authorities
regarding any access to or use of EPRI Intellectual Property
hereunder that may be in violation of applicable U.S. or
foreign export laws or regulations.

© 2017 Electric Power Research Institute (EPRI), Inc. All rights reserved.
Electric Power Research Institute, EPRI, and TOGETHER…SHAPING THE
FUTURE OF ELECTRICITY are registered service marks of the Electric
Power Research Institute, Inc.
3002011007

Electric Power Research Institute

3420 Hillview Avenue, Palo Alto, California 94304-1338 • PO Box 10412, Palo Alto, California 94303-0813 • USA
10253878800.313.3774 • 650.855.2121 • askepri@epri.com • www.epri.com