D8.1 PU Design and Specifications of Stable and Secure Distribution Grids - Compressed

Ref.
Ares(2016)4118986 - 04/08/2016
Title: Document Version:
D8.1 Design and specifications of Stable and secure distribution grids 1.0
Project Number: Project Acronym: Project Title:
H2020-646184 NOBEL GRID New Cost Efficient Business Models for Flexible Smart
Grids
Contractual Delivery Date: Actual Delivery Date: Deliverable Type*-Security*:
M19 (July 2016) M19 (July 2016) R-PU

*Type: P: Prototype; R: Report; D: Demonstrator; O: Other.
**Security Class: PU: Public; PP: Restricted to other programme participants (including the Commission); RE: Restricted to a
group defined by the consortium (including the Commission); CO: Confidential, only for members of the consortium (including the
Commission).
Responsible Organization Contributing WP
Jovica Milanovi UNIMAN WP8
Authors (organisation):
Huilian Liao (UNIMAN), Jelena Ponoko (UNIMAN), Xiaoqing Tang (UNIMAN), Jovica Milanovi (UNIMAN), Rafael
Peris (ETRA), Lola Alacreu (ETRA), Lucas Pons(ETRA), Jan Ringelstein(DERlab), David Nestle (IWES),Mihai Sanduleac
(ECRO), Cristian ARABOLU (ECRO), Rares POP (ECRO) and Dorin OLAR (ECRO), Mislav Findrik (AIT), Kostas Tsatsakis
(HYPERTECH), Dimitris Trakas (ICCS/NTUA), Aggelos iliopoulos (ICCS/NTUA), Rebekka Gkogkou (ICCS/NTUA), Nikos
Koutsoukis (ICCS/NTUA), Pavlos Georgilakis (ICCS/NTUA), Kelly Mavrogenous (ICCS/NTUA), Athanasios Vasilakis
(ICCS/NTUA), Aris Dimeas (ICCS/NTUA), Pietro Fragnito(ENG), Antonello Corsi(ENG), Giampaolo Fiorentino(ENG),
Arjan Aalberts (DNV), Marcel Eijgelaar(DNV),, Hans de Heer (DNV), Lambert van Vught(DNV), Joo Murta Pina
(UNINOVA), Joo Martins (UNINOVA), Vasco Delgado-Gomes (UNINOVA)
Abstract:
This deliverable describes the design of innovative functionalities and the integration of innovative and existing
technologies in smart distribution grids in order to facilitate the management and maintenance of the smart
distribution grids and provide cost efficient, secure and environmentally sustainable MV/LV distribution networks.
Keywords:
Smart distribution grids; state estimation; load flow; load forecasting; RES forecasting; demand side management
ETRA I+D ETRA INVESTIGACIN Y DESARROLLO, S.A.

Avda. Tres Forques, 147 Tel. +34 96 313 40 82 http://www.nobelgrid.eu/
46014 Valencia (Espaa) Fax +34 96 350 32 34 lalacreu.etra-id@grupoetra.com
Revision History
Revision Date Description Author (Organisation)
V0.1 15.04.2016 First draft of deliverable ToC Jovica Milanovi (UNIMAN)
V0.2 19/04/2016 Updated ToC Marcel Eijgelaar(DNV), Kostas

Tsatsakis (HYPERTECH)
v.0.3 30/06/2016 Eng contribution to USEF-NG adaptation Pietro Fragnito (ENG),

Antonello Corsi (ENG), ), Kostas
Tsatsakis (HYPERTECH)
v.0.4 4/7/2016 Input for Load Flow,RES Forecast,Static data Dimitris Trakas (ICCS/NTUA) ,
analysis, Transient Analysis Aggelos iliopoulos (ICCS/NTUA),
Rebekka Gkogkou (ICCS/NTUA),
Nikos Koutsoukis (ICCS/NTUA) ,
Pavlos Georgilakis (ICCS/NTUA),
Kelly Mavrogenous ICCS/NTUA
Nasos Vasilakis (ICCS/NTUA),
Aris Dimeas (ICCS/NTUA),
V0.5 04/07/2016 ESB and sub-optimal situations Description Rafael Peris (ETRA), Lola
Alacreu (ETRA), Lucas Pons
(ETRA)
V0.6 05/07/2016 State estimation, Load forecasting Huilian Liao (UNIMAN)
V0.7 07/07/2016 Formatting the document Version for review Huilian Liao (UNIMAN)
V0.8 28/07/2016 Final version Huilian Liao (UNIMAN)
V1.0 01/08/2016 Final version for submission
D8.1 Design and specifications of Stable and secure distribution grids 2

Index
1 INTRODUCTION ........................................................................................................ 12
1.1 Purpose of the deliverable ......................................................................................... 12
1.2 Scope of the deliverable............................................................................................. 12
1.3 Structure of the deliverable ....................................................................................... 12
2 DATA ACQUISITION AND INFORMATION EXCHANGE ................................................. 14

2.1 Introduction................................................................................................................ 14
2.2 Security requirements of real-time information exchange ....................................... 15
2.3 Development of common infrastructure for information exchange ......................... 19
3 DATA ANALYSIS ........................................................................................................ 26

3.1 Introduction................................................................................................................ 26
3.2 Analysis of Passive data.............................................................................................. 26
3.3 Energy quality monitoring service ............................................................................. 29
3.3.1 Voltage profile .............................................................................................................31
3.3.2 Voltage failures ............................................................................................................39
3.3.3 Voltage asymmetry ......................................................................................................40
3.3.4 Current asymetry .........................................................................................................40
4 FUNCTIONALITIES AND TOOL DEVELOPMENT ............................................................ 42

4.1 Introduction................................................................................................................ 42
4.2 State estimation ......................................................................................................... 42
4.2.1 SE problems .................................................................................................................42
4.2.2 SE processes ................................................................................................................43
4.2.3 Distribution System State Estimation ..........................................................................44
4.2.3.1 Three-phase state estimation ....................................................................... 45
4.2.3.2 DSSE Measurements...................................................................................... 45
4.2.3.3 Non-Linear Power Flow Equations ................................................................ 47
4.2.3.4 Uncertainty and Estimation Accuracy ........................................................... 49
4.2.4 Simulation results ........................................................................................................50
4.3 Load Flow ................................................................................................................... 53
4.3.1 Distribution Network Lines ..........................................................................................54
4.3.1.1 Overhead Distribution Lines .......................................................................... 54
4.3.1.2 Underground distribution lines ..................................................................... 56

4.3.1.2.1 Concentric neutral cable ............................................................. 56
4.3.1.2.2 Tape-Shielded cable .................................................................... 57
4.3.1.3 Distribution Line Model ................................................................................. 58
4.3.2 Three-Phase Transformers and Voltage Regulators ....................................................58
4.3.2.1 Three-Phase Transformers ............................................................................ 58
4.3.2.2 Voltage Regulators ........................................................................................ 62
4.3.3 Switches .......................................................................................................................63
4.3.4 Loads ............................................................................................................................64
4.3.5 Shunt Capacitor Banks .................................................................................................66
4.3.6 Backward/Forward Sweep Power Flow Method for Radial Distributions Networks ..67
4.3.6.1 Bus Indexing .................................................................................................. 67
4.3.6.2 Backward sweep step .................................................................................... 68
4.3.6.3 Forward sweep step ...................................................................................... 70
4.3.7 Conclusion ...................................................................................................................72
4.4 Load Forecasting ........................................................................................................ 72
4.4.1 Artificial Neural Network .............................................................................................72
4.4.2 A Committee of Machine Learning Techniques in a load forecasting Architecture....75
4.4.3 Total MW and MVar Prediction Case Study ................................................................77
4.4.3.1 Total MW Prediction in Greece ..................................................................... 77
4.4.3.2 Total MW and MVAR Prediction in UK .......................................................... 81
4.5 RES forecasting ........................................................................................................... 81
4.5.1 Methodology ...............................................................................................................82
4.5.1.1 Radial Basis Function Neural Network (RBFNN) ............................................ 83
4.5.1.2 Random Forests (RFs) .................................................................................... 84
4.5.2 Environments, Module Inputs and Outputs ................................................................87
4.5.2.1 Inputs ............................................................................................................. 87
4.5.2.2 Outputs .......................................................................................................... 87
4.5.3 High Level Forecast Service .........................................................................................87
5 INTEGRATION OF FUNCTIONALITIES IN GRID MANAGEMENT .................................... 89

5.1 Introduction................................................................................................................ 89
5.2 Standardized protocals and interfaces ...................................................................... 90
5.2.1 Standardized protocols between DSO and Aggregator - Interfaces with DRFM framework
90
5.2.1.1 Introduction to USEF framework................................................................... 90
5.2.1.2 USEF Adaptation in NOBEL GRID Framework................................................ 92
5.2.2 NOBEL GRID Demand Flexibility Framework ...............................................................98

5.2.3 Direct Interface between G3M and NOBEL GRID DER ..............................................100
5.3 Grid stability balancing support and fault analysis ............................................... 103
5.3.1 PV-inverter Management Functionalities .................................................................103
5.3.2 Voltage Control in Distribution Networks with DGs ..................................................104
5.3.2.1 The decomposition method ..................................................................... 105
5.3.2.2 Voltage control ............................................................................................ 106
5.3.2.3 Case study.................................................................................................... 107
5.3.2.4 Conclusions .................................................................................................. 113
5.3.3 Analysis of outage and the black-start process .........................................................113
5.3.3.1 Introduction ................................................................................................. 113
5.3.3.2 Modeling and Control.................................................................................. 114
5.3.3.3 Microgrid Black Start ................................................................................... 118
5.3.3.4 Black Start test scenarios............................................................................. 118
5.4 Development of Verification and certification services ........................................... 120
5.4.1 Necessity for Verification and Certification of Smart Grid Services ..........................120
5.4.2 Levels and Structure of Smart Grid Verification Services ..........................................121
5.4.3 Smart Grid Verification Services Within NOBEL GRID ...............................................122
5.4.4 Process Verification ...................................................................................................123
5.4.5 Protocol Assessment .................................................................................................124
5.5 Identification of sub-optimal situations ................................................................... 126
5.5.1 Introduction ...............................................................................................................126
5.5.2 Strategic Manager .....................................................................................................128
5.5.3 Initial Identification of Situations ..............................................................................128
5.5.4 Strategies ...................................................................................................................129
5.5.5 New Situations and Learning .....................................................................................130
5.5.6 User Interface ............................................................................................................131
5.5.7 Future Situations .......................................................................................................131
5.5.8 Complementary Visualization....................................................................................132
6 REFERENCES AND ACRONYMS................................................................................. 133

6.1 References ................................................................................................................ 133
6.2 Acronyms.................................................................................................................. 138
7 APPENDIX ............................................................................................................... 142

7.1 Appendix A ............................................................................................................... 142
7.1.1 IEEE 13-bus distribution test system .........................................................................142
7.1.2 IEEE 123-bus distribution test system .......................................................................145

7.1.3 Meltemi distribution network ...................................................................................160
7.2 APPENDIX B .............................................................................................................. 174
7.3 APPENDIX C .............................................................................................................. 175

LIST OF FIGURES
Figure 2.1 Data acquisition from SMX\SLAMs located in the distribution grid [D3.1] ........................ 14
Figure 2.2 NIST logical interfaces and connections between NOBEL GRID actors involved in D9.1 solutions
(SMX, DRFM and G3M) ........................................................................................................................ 16
Figure 2.3 NIST logical actors grouped into SLAM, G3M and DRFM domains .................................... 17
Figure 2.4 ESB overall connectivity diagram ....................................................................................... 20
Figure 2.5 Basic ESB communication schema ..................................................................................... 21
Figure 2.6 ESB subscribe ...................................................................................................................... 21
Figure 2.7 ESB publish ......................................................................................................................... 22
Figure 2.8 ESB client app with local database ..................................................................................... 22
Figure 2.9 MDM - DLMS integration.................................................................................................... 23
Figure 2.10 ESB query from measures ................................................................................................ 23
Figure 2.11 SMX inter-module communication .................................................................................. 24
Figure 2.12 SMX measures integration in ESB..................................................................................... 25
Figure 3.1 Alginet Test Site .................................................................................................................. 27
Figure 3.2 Meltemis single line diagram............................................................................................. 28
Figure 3.3 Terni s single line diagram ................................................................................................. 29
Figure 3.4 Distribution of probability (DP) graphic using a sorting method ....................................... 32
Figure 3.5 Measurements with IEC61000-4-30 equipment, URMS each 10 minutes ......................... 35
Figure 3.6 Comparison between IEC61000-4-30 and LPQAM methods over approximate 7 days, URMS each
10 minutes ........................................................................................................................................... 35
Figure 3.7 Hourly PD over three consecutive hours, calculated with voltage values recorded by SMX and an
LV connected ZMG310 meter.............................................................................................................. 36
Figure 3.8 Hourly PD over an entire day (24 hours), calculated with voltage values recorded by SMX and an
LV connected ZMG310 meter.............................................................................................................. 36
Figure 3.9 Daily voltage profile on the MV part of HV/MV transformer in Alginet (secondary values)37
Figure 3.10 Daily voltage profile on the LV part of an MV/LV transformer in Alginet ....................... 37
Figure 3.11 Voltage profile during PV production in a selected day, Terni ........................................ 38
Figure 3.12 Voltage profile on LV of a UK consumer, recorded each 10 seconds .............................. 38
Figure 3.13 Daily voltage profile, UPB, 19.02.2016 ............................................................................ 39
Figure 3.14 Daily voltage profile, UPB, 19.02.2016, detail for the interval 7:00 8:00 ..................... 39
Figure 4.1 Illustration of SE processes ................................................................................................. 43
Figure 4.2 Distribution network and actual load profile ..................................................................... 50
Figure 4.3 Simulation results on 24-bus test network ........................................................................ 50
Figure 4.4 Single line diagram of 295-bus generic distribution network .......................................... 51
Figure 4.5 Comparison between the actual and estimated voltages for 295-bus GDN ...................... 52
Figure 4.6 Heatmap indicating voltages of the network (20 real measurements) ............................. 52
Figure 4.7 Three-phase distribution line model .................................................................................. 58

Figure 4.8 LDC circuit ........................................................................................................................... 63
Figure 4.9 Load connection (a) Wye-connected load and (b) Delta-connected load.......................... 64
Figure 4.10 Flowchart of the backward/forward sweep method ....................................................... 67
Figure 4.11 Distribution network example.......................................................................................... 67
Figure 4.12 29-bus distribution network with indexed buses ............................................................. 68
Figure 4.13 Distribution network segment example ........................................................................... 69
Figure 4.14 Flowchart of the backward sweep step............................................................................ 70
Figure 4.15 Flowchart of the forward sweep step .............................................................................. 71
Figure 4.16 Structure of an ANN ......................................................................................................... 73
Figure 4.17 Total load forecasting framework .................................................................................... 75
Figure 4.18 Total Load Forecasting Training Process .......................................................................... 75
Figure 4.19 The structure of the proposed load forecasting model ................................................... 76
Figure 4.20 Load time series from May 1, 2013 until April 5, 2014 .................................................... 77
Figure 4.21 Load demand during a week in the winter....................................................................... 78
Figure 4.22 Load demand during a week in the summer .................................................................... 78
Figure 4.23 Daily consumed energy versus daily maximum temperature .......................................... 79
Figure 4.24 Load curve during a few special days ............................................................................... 79
Figure 4.25 The mean absolute error of the proposed model at the period 1 ................................... 80
Figure 4.28 Predicted demand versus actual demand: (a) real power, (b) reactive power................ 81
Figure 4.29 The structure of the proposed model .............................................................................. 83
Figure 4.30 Training of a Classification Tree [83] ................................................................................ 86
Figure 4.31 Output of the training ensemble [83] .............................................................................. 86
Figure 4.32 High level overview of the Forecast Service ..................................................................... 88
Figure 5.1 Illustration of integration of functionalities in grid management ...................................... 89
Figure 5.2 USEF Protocol Business Framework ................................................................................... 91
Figure 5.3 - Sequence Diagram Cockpit/G3M ..................................................................................... 94
Figure 5.4 - Sign interpretation ........................................................................................................... 95
Figure 5.5 - Forecasting load ............................................................................................................... 96
Figure 5.6 - Flexibility obtained applying DR by Aggregator ............................................................... 97
Figure 5.7 - Measurement by DSO after PTU is over ........................................................................... 97
Figure 5.8 Demand Flexibility as a service for DSOs ............................................................................ 98
Figure 5.9: Information chain between DER and G3M...................................................................... 100
Figure 5.10 OGEMA Resource structure representing the grid connection point metering (sample SHIC) 102
Figure 5.11 Connection of the three-phase four-leg PV inverter, intended to compensate grid unbalances.
........................................................................................................................................................... 104

Figure 5.12 Decentralized control of power system ......................................................................... 105
Figure 5.13 Single-line diagram of feeder P-25 ................................................................................. 108
Figure 5.14 Partition results of P-25 for =0.5 .................................................................................. 109
Figure 5.15 Voltage profile before and after centralized voltage control......................................... 110
Figure 5.16 Voltage profile before and after decentralized voltage control..................................... 111
Figure 5.17 Voltage profile before and after decentralized voltage control with active power adjustment
........................................................................................................................................................... 112
Figure 5.18 Overview of the synchronous generator structure ........................................................ 114
Figure 5.19 Equivalent circuit of synchronous generator in dq-frame. ............................................ 115
Figure 5.20 Grid forming inverter model.......................................................................................... 116
Figure 5.21 Control block diagram of the grid-forming VSC system [99]. ......................................... 116
Figure 5.22 Frequency vs power droop characteristic. ..................................................................... 117
Figure 5.23 Voltage vs reactive power droop characteristic ............................................................. 117
Figure 5.24 Droop control with synchronization scheme ................................................................ 117
Figure 5.25 Microgrid configuration for black start scenarios. ......................................................... 119
Figure 5.26 Black start test scenario with (right) grid forming inverter and (left) with SG. .............. 119
Figure 5.27 Active power sharing for grid forming inverter synchronisation (P/Sn). ........................ 120
Figure 5.28 Matlab/Simulink microgrid model for the black start scenarios. ................................... 120
Figure 5.29 Verification of smart grid communication layers ........................................................... 121
Figure 5.30 Layers for verifying Process and IT communication ....................................................... 122
Figure 5.31 Overview of smart meter and smart grid protocols for which verification services already exists
........................................................................................................................................................... 124
Figure 5.32 Structure of the message and process (conversation) test. The lowest row indicates which
system initiates the test .................................................................................................................... 125
Figure 5.33 Scenario selection. .......................................................................................................... 125
Figure 5.34 overview of the test scenarios for the aggregators system. ......................................... 125
Figure 5.35 example of test results (in this case the system of a BRP). ............................................ 126
Figure 5.36 Situations Management ............................................................................................... 127
Figure 5.37 G3M Strategic Manager ................................................................................................. 128
Figure 5.38 Initial Classification ....................................................................................................... 129
Figure 5.39 Cluster Creation ............................................................................................................ 129
Figure 5.40 Strategy Selection ......................................................................................................... 130
Figure 5.41 New Situation ............................................................................................................... 130
Figure 5.42 Situations Comparison .................................................................................................. 131
Figure 5.43 Evolution and Risk......................................................................................................... 132
Figure 7.1 IEEE 13-bus distribution test system [68] ......................................................................... 142
Figure 7.2 IEEE 123-bus distribution test system [68] ....................................................................... 145
Figure 7.3 Meltemi distribution network .......................................................................................... 161

Figure 7.4 Load profile (p.u.) ............................................................................................................. 161
Figure 7.5 Bus 25 voltage magnitude profile (p.u.) of the Meltemi distribution network................ 168
Figure 7.6 Bus 34 voltage magnitude profile (p.u.) of the Meltemi distribution network ................ 169
Figure 7.7 Bus 55 voltage magnitude profile (p.u.) of the Meltemi distribution network ................ 171
Figure 7.8 Bus 1 outgoing current magnitude profile (p.u.) of the Meltemi distribution network .. 172
Figure 7.9 Overhead distribution line spacing [58] ........................................................................... 174
Figure 7.10 Underground distribution line spacing [58] ................................................................... 174

LIST OF TABLES
Table 2.1 NOBEL GRID D8.1 Actors mapped to NIST logical interfaces............................................... 15
Table 2.2 Mapping of NIST logical interfaces into logical interface categories (LIC) [D3.1] ................ 17
Table 2.3 Identified logical interfaces and categories for connections between NOBEL GRID frameworks 18
Table 2.4 NIST-IR 7628 High level security requirements for interfaces between the NOBEL GRID
frameworks.......................................................................................................................................... 18
Table 3.1 Ternis Characteristics .......................................................................................................... 29
Table 4.1 Measurement error variances (%) ....................................................................................... 51
Table 4.2 Simulation results with 10 randomly selected substations for real measurements ........... 53
Table 4.3 Simulation results based on maximum possible errors in real measurements................... 53
Table 5.1 Partition results of P-25 for different values of parameter ............................................ 108
Table 5.2 Results of centralized voltage control ............................................................................... 110
Table 5.3 Results of decentralized voltage control ........................................................................... 111
Table 5.4 Results of decentralized voltage control ........................................................................... 112
Table 5.5 division of questions among the sections in the process verification audit...................... 123
Table 5.6 a sample of results of a preliminary process audit for the process between an aggregator and a
DSO in an existing Dutch Smart grid pilot. ........................................................................................ 123
Table 7.1 Data of the IEEE 13-bus test system .................................................................................. 143
Table 7.2 Bus voltages of the IEEE 13-bus distribution system ......................................................... 144
Table 7.3 Line currents of the IEEE 13-bus distribution system ........................................................ 144
Table 7.4 Data of the IEEE 123-bus test system ................................................................................ 146
Table 7.5 Bus voltages of the IEEE 123-bus distribution system ....................................................... 154
Table 7.6 Line currents of the IEEE 123-bus distribution system ...................................................... 157
Table 7.7 Data of the Meltemi distribution network ........................................................................ 162
Table 7.8 Bus voltages of the Meltemi distribution network during peak load demand.................. 165
Table 7.9 Line currents of the Meltemi distribution network during peak load demand................. 166
Table 7.10 Bus 25 voltage profile of the Meltemi distribution network ........................................... 168
Table 7.13 Bus 1 outgoing current profile of the Meltemi distribution network ............................. 172
Table 7.14 Conductor data [58] ......................................................................................................... 174
Table 7.15 Sub-metering data as OGEMA representation ................................................................ 175
Table 7.16 Sample inverter as OGEMA representation .................................................................... 176

1 INTRODUCTION
1.1 PURPOSE OF THE DELIVERABLE
Smart Meters have the advantage to be presented at nearly each bordering point of the network, thus
acting as guardians or border equipment. With NOBEL GRID SM approach, it is envisaged a high use of all
new meter data, such that complex functionalities can be built for Smart Grid purpose without needing
additional equipment in these places. Innovative Smart Grid functionalities for the MV/LV distribution
network will be used within the project. The functionalities will be designed as independent services built
on the same distributed infrastructure of advanced smart meters. This deliverable is to provide a definition
of innovative smart distribution grid functionalities and the existing innovative technologies that can be
integrated to support the integration of local distributed renewable energy and also demand response
system, and provide cost efficient, secure and environmentally sustainable MV/LV distribution network.
The deliverable, which belongs to WP8 Stable and secure distribution grid innovative technologies, is
closely related to WP3 Flexible Architecture and Specification of Advanced and Cost-efficient Smart Grid
and Smart Metering Solutions, which defines the detailed architecture required to build the next
generation functionalities envisioned for the distribution smart distribution grid and smart metering
solutions. The functionalities and technologies can mainly be employed in WP9 Grid Management and
Maintenance Master Framework for DSOs (G3M Framework), which is to develop the Framework that will
support DSOs in order to monitor, control, manage and maintain the electricity distribution network.
1.2 SCOPE OF THE DELIVERABLE

The deliverable includes the description of:
SCADA support and DER dispatch service for Smart Grid using the Smart Meters real-time data (at
the consumer level) and direct/indirect control of DERs.
Energy quality monitoring service, especially voltage levels and voltage asymmetries, current
asymmetries in low voltage, time of voltage failures at customer levels;
Integration of Smart Meter Data in the existing Distribution Management Applications: Load and
RES Forecast, State Estimation and Load Flow. This way the operators will has accurate knowledge
of what is happening inside the Distribution Network. Perform secondary voltage control helped by
Load Flow calculation and active/reactive power control capabilities of inverters.
Avoiding network congestion in MV/LV transformer, by avoiding transformer and other critical
sections overload through signals of power reduction or increase to selected consumers. This is a
Demand-Side-Management (DSM) service for congestion avoidance in the distribution grid, as a
network service.
These functionalities are based on the extensive use of meters real-time values and events and a proper
monitoring system of the LV distribution network.
1.3 STRUCTURE OF THE DELIVERABLE

The document is further segmented into the following sections:
1. Chapter 1 provides the overview of the Deliverable focusing on the scope of the document;
2. Chapter 2 describes data acquisition;

3. Chapter 3 provides data analysis including passive data and energy quality monitoring services;
4. Chapter 4 defines the functionalities and tool development, e.g., state estimation, three-phase load
flow, load forecasting and RES forecasting;
5. Chapter 5 describes the infrastructure of integrating functionalities in grid management.

2 DATA ACQUISITION AND INFORMATION EXCHANGE
2.1 INTRODUCTION
NOBEL GRID functionalities and technologies described within this deliverable are enhancing the Grid
Management and Maintenance Master Framework (the G3M framework) that shall support DSOs to
control, manage and maintain MV/LV distribution grids with high penetration of renewable energy. These
services rely on extended smart meter functionalities (SMX) (D4.1) that allow operators of the G3M
Framework and the DRFM Framework to exchange information necessary to:
Dispatch DERs across distribution grids and at the consumer level
Monitor energy quality at customers level
Perform Load and RES forecast, as well as State Estimation and Load Flow
Avoid network congestion using demand-side management using G3M and DRGM (Aggregator
interface in order to help utilities shave peak power demand, secure power supply, etc.
Therefore, collecting data streams coming from smart meters (SMX\SLAM) located across the distribution
grid (either at customers premises or at the distribution substations) is required. The G3M control center is
exchanging measurements and control data with distributed generators, storages and electrical appliances
connected at SMXs at the prosumers premises. Information elements from DER to G3M include active and
reactive power measurements and status information. The SMX at the prosumer is also sending grid
connection point voltage and frequency to the G3M framework for energy quality monitoring services. The
data acquired from the SMXs in the field is stored in the local data repository and can later be used for
services such as load forecasting (see Figure 2.1).
DRFM for Aggregator #2
EMA APP
DRFM for Aggregator #1

G3M for DSO
Comm. Data
frontend Repository
Comm. Data
frontend Repository
SMX/SLAM
SMX/SLAM
SMX/SLAM
Distribution
grid
Figure 2.1 Data acquisition from SMX\SLAMs located in the distribution grid [D3.1]
In this chapter, security requirements for real-time information exchange between three principal entities
are derived, namely, the G3M framework, the SMX components and the DRFM framework. Moreover, the
technical infrastructure and communication drivers that enable seamless data acquisition and their usage
across various services are described.

2.2 SECURITY REQUIREMENTS OF REAL-TIME INFORMATION EXCHANGE
NOBEL GRID solutions and grid services presented in this deliverable require real-time information
exchange between the G3M framework, smart meters (SMXs), and the Demand-Response Management
Framework (Aggregators). All components therefore need to be connected via ICT networks in order to
exchange information required by the services. Interconnection of these systems over public and private
networks opens up new interfaces for potential cyber-attacks. Cyber-attacks could have significant impact
on the smart grid and endanger its operations.
In order to properly secure interfaces between different components involved in information exchange a
security by design approach to elicit the security requirements that should be taken into account during the
NOBEL GRID system design is applied. The NOBEL GRID approach for defining high level security
requirements is based on the NIST Internal Report (IR) 7628 security guidelines [1]. These requirements will
be considered during the ongoing development activities of the information exhcnage interface of the
NOBEL GRID framework.
NISTIR 7628 proposes a risk assessment process based on a conceptual smart grid model consisting of
predefined 49 actors connected with communication interfaces which are also called the logical interfaces.
The logical interfaces are further grouped into logical interface categories (LIC), whereas for each LIC a set
of high level security requirements are defined. In order to restrict the security requirements only to the
services developed in this deliverable, we reuse the NIST IR based analysis of the whole NOBEL GRID
architecture provided in the D3.1 and focus in particular on the NOBEL GRID Logical Actors which are
involved in information exchange with services developed in the deliverable D8.1. To extract the security
requirements the following steps are performed:
1. NOBEL GRID logical actors involved in necessary information exchange for services developed in
the WP8 and defined their mapping to corresponding logical actors from NIST logical
reference model (see Table 2.1).
2. Subsequently, the identified logical actors and their logical interfaces are located in the smart
grid plane (see Figure 2.2).
3. The logical actors are further grouped to respective NOBEL GRID frameworks (either to the
G3M framework, the DRFM framework or to the SMX framework see Figure 2.3).
Logical interfaces for communication between the NOBEL GRID framework are identified and mapped to
assiociated NIST logical interfaces categories, from which the security requirements are derived (Table 2.2).
Table 2.1 NOBEL GRID D8.1 Actors mapped to NIST logical interfaces
NG Logical Actor NIST 7628 Actor
SMM (8) Meter
SMX (5) CEMS; (7) HAN Gateway; (25) Distributed Generation and
Storage Management
SHIC (3) Customer Appliances and Equipment
Inverter Controller (4) DER
DRFM Cockpit (41) Aggregator/Retail Energy Provider;
DRFM (32) LMS/DRMS
Flexibility Engine (32) LMS/DRMS

G3M GUI (28) Distribution Operator
G3M (27) DMS; (29) Distribution SCADA; (30) EMS; (31) ISO/RTO
Operations;
Grid (18) Distribution Sensor; (15) RTUs or IEDs
Distribution Data Collector (12) Distribution Data Collector; (21) AMI
Actor-specific Data Repository (12) Distribution Data Collector
Actor-specific Big Data Repository (12) Distribution Data Collector; (33) MDMS
Generation Transmission Distribution DER Customer Premise
U63
Market U4
41
Aggregator /
Retail Energy
42 Billing
Provider
U15
28 Distribution
Operator U51
Enterprise
U101 2 Customer
U1 U92
U100 U39
U8
U98
27 DMS 25 Distributed
33 MDMS U125 U96
Generation and
Storage 38 Customer
U9 U2
Operation U11 Management Portal U37
29 Distribution U65
U33
U99 SCADA
32 LMS/DRMS U22 23 CIS
U2 U21
U119
12 Distribution U137
Station 21 AMI
Data Collector U106
U104 U113 U32
U7 U88
U14 U59
U25
U24
U35 U130
5 CEMS
U3 U42 U64
U111 7 HAN
Field Gateway
U117 U112 U43
15 RTUs or U41
IEDs U44 U60
U45
U127 U124
3 Customer 8 Meter
Appliances and
18 Distribution U126 Equipment U120
Process 4 DER
Sensor
9 Customer
Premise Display
Figure 2.2 NIST logical interfaces and connections between NOBEL GRID actors involved in D9.1
solutions (SMX, DRFM and G3M)

Generation Transmission Distribution DER Customer Premise
Market
41
Aggregator /
20 ISO/RTO
Retail Energy
42 Billing
Provider
Enterprise 28 Distribution 43 ESP

Operator 2 Customer
40 WMS
31 ISO/RTO 27 DMS 25 Distributed 33 MDMS

Operations 39 WAMS Generation and
Storage
Operation Management
29 Distribution
30 EMS SCADA
32 LMS/DRMS
12 Distribution
Station
Data Collector 21 AMI
16 Field Crew
Tools
Field 5 CEMS
15 RTUs or 7 HAN
IEDs Gateway
3 Customer
8 Meter
Appliances and
Equipment
SLAM
Process
18 Distribution
4 DER
Sensor G3M
9 Customer
Premise Display
DRFM
Figure 2.3 NIST logical actors grouped into SLAM, G3M and DRFM domains
Table 2.2 Mapping of NIST logical interfaces into logical interface categories (LIC) [D3.1]
Logical Interfaces
LIC 1 - 4 U3, U67, U79, U81, U82, U85, U102, U117, U135, U136, U137
LIC 5 U9, U27, U65, U66, U89
LIC 6 U7, U10, U13, U16, U56, U74, U80, U83, U87, U115, U116
LIC 7 U2, U22, U26, U31, U63, U96, U98, U110
LIC 8 U1, U6, U15, U55
LIC 9 U4, U17, U20, U51, U52, U53, U57, U58, U70, U72, U90, U93, U97
LIC 10 U12, U30, U33, U36, U59, U75, U91, U106, U113, U114, U131
LIC 11 U111
LIC 12 U108, U112
LIC 13 - 14 U8, U21, U25, U32, U95, U119, U130
LIC 15 U42, U43, U44, U45, U49, U62, U120, U124, U126, U127
LIC 16 U18, U37, U38, U39, U40, U88, U92, U100, U101, U125
LIC 17 U14, U29, U34, U35, U99, U104, U105
LIC 18 U24, U41, U46, U47, U48, U50, U54, U60, U64, U128, U129
LIC 19 U77, U78, U134
LIC 20 U11, U109
LIC 21 U5
LIC 22 U133

Table 2.3 Identified logical interfaces and categories for connections between NOBEL GRID
frameworks
Framework connections Logical Interfaces Interface Categories

1 G3M<->SMX U137,U111,U112,U117,U7,U57,U8 1, 2, 3, 4, 6, 11, 12, 13, 14, 15
2 DRFM<->SMX U92, U22, U106, U32 7, 13, 14, 16
3 G3M<->DRFM U51, U15, U90, U52, U11 9, 10
Once the logical interface categories for each framework connection have benn indentified (
Table 2.3), the associated high-level security requirements are derived from the general categories to
which the logical interfaces are assigned (Table 2.4).
Table 2.4 NIST-IR 7628 High level security requirements for interfaces between the NOBEL GRID
frameworks
Framework Connections High-level Security Requirement

GgGGG3M <-> SMX
Permitted Actions without Identification or Authentication
User Identification and Authentication
Device Identification and Authentication
Authenticator Feedback
Security Function Isolation
Denial-of-Service Protection
Boundary Protection
Communication Integrity
Voice-Over Internet Protocol
Application Partitioning
Software and Information Integrity
Non-Repudiation
Information Remnants
Confidentiality of Information at Rest
Communication Confidentiality
Session Lock
Remote Session Termination
RemoteAccess
Permitted Actions without Identification or
Authentication

GgGGG3M <-> DRFM Session Lock
Non-Repudiation
Information Remnants
Boundary Protection
GgGGG3M <-> DRFM Concurrent Session Control
Session Lock
Remote Session Termination
RemoteAccess
Boundary Protection
Voice-Over Internet Protocol
2.3 DEVELOPMENT OF COMMON INFRASTRUCTURE FOR INFORMATION EXCHANGE

There is no central data repository in NOBEL GRID that comprises all the data of all the application at the
different control centers. Instead of this, there is a middleware to let applications exchange data: the
NOBEL GRID enterprise service bus (ESB).
The ESB details are described in D9.1 G3M Framework functionalities specification and design. As a
summary, the ESB provides a way for exchanging information and send commands. Its characteristics
include:
The communication infrastructure is based on RabbitMQ
The data model of the messages is defined by the NOBEL GRID consortium and based on CIM
messages (IEC 61968)

In this context there are applications that act as servers of specific services or pieces of data. The ESB will
route the right query messages to these applications/services. They will be able to analyse and process the
queries and send back answer messages.
On the other side, there are applications that demand data or services (see Figure 2.4). These applications
send the queries to the ESB that will distribute the message accordingly. Eventually, the answer to the
queries will be received by the client:
Works/crew DB Assets DB
Web server
Electrical System GUI

calculation
(steady-state, load Workflow
flow, etc.) Asset manager ...
manager
wrapper wrapper wrapper wrapper
Nobelgrid ESB middleware
wrapper
wrapper wrapper wrapper
Demand
SCADA
Billing driver Response Topology manager
driver ...
(DLMS) driver
(MQTT)
(openADR)
CIM network
topology
DACF layer
Smart grid
Figure 2.4 ESB overall connectivity diagram
In this structure, everything is asynchronous, the clients do not get blocked waiting for the answer and the
fact that a server is processing a big query, does not prevent other queries to be dispatched (this
mechanism increases the overall system performance). Behind the scenes, the communication is based on
message queues (RabbitMQ). Message queues provide an asynchronous communications protocol,
meaning that the sender and receiver of the message do not need to interact with the message queue at
the same time. Messages placed onto the queue are stored until the recipient retrieves them. Message
queues have implicit or explicit limits on the size of data that may be transmitted in a single message and
the number of messages that may remain outstanding on the queue.

This kind of architecture is often referred as microservices architecture. In short, the microservice
architectural style is an approach to developing a single application as a suite of small services, each
running in its own process and communicating with lightweight mechanisms, often a messaging broker
platform or HTTP resource API. These services are built around business capabilities and independently
deployable by fully automated deployment machinery. Implementing this architecture requires a minimal
centralized management of services and the use of an additional component in the system, the message
broker, but it provides some interesting features:
Durability of the messages: Message is not lost even if the server is down
Purging policies, filtering and delivery policies, to better manage the data flux
Scalability by design: The queues are the basic way for distribution of task among workers. New
parallel workers can be added to leverage some complex task even without the need of restarting
the system.
The different NOBEL GRID services at the control center (G3M and DRFM) will play the role of
microservices, each one providing services required for a wider composed functionality. The microservices
may be written in different programming languages and use different data storage technologies.
Below in Figure 2.5, is the basic communication schema of the NOBEL GRID ESB. The client sends the
request messages encoded with a specific schema, the ESB routes the messages to the right server and the
answer is sent back to the answer queue where the client is listening to and waiting for the answer:
Nobelgrid ESB
requests queue
Client App
Server App
Message routing
Answer queue
Figure 2.5 Basic ESB communication schema
The NOBEL GRID ESB has also support for publish-subscribe mechanisms. In this paradigm, the client
subscribes to a given source of information (data changes, events, etc.) and then start receiving messages
asynchronously when new information is available and ready to deliver. The subscription part is given in
Figure 2.6.
Nobelgrid ESB
(2) Subscribe
requests queue
(1) subscribe
Client App
Server App
Message routing
Figure 2.6 ESB subscribe
When data changes, the clients get specific messages with the details:

Nobelgrid ESB
Client App
Server App
Answer queue
(2) publish (1) Publish
Figure 2.7 ESB publish
This mechanism is also worth for filling the client app database. The client applications may have the need
of its own database for its calculations and internal processes. This is particularly important for the services
that provide results based on data analysis.
The client application with specific database will send periodic requests of data to the ESB or alternatively
subscribe to specific data; in any case, data will eventually be available for the client application and it can
get it and use it to fill their internal database:
Local DB
(5) Data results Nobelgrid ESB

(instant/time series) (2) Data query
requests queue
(1) Data query
Client App
Server App
Message routing
Answer queue
(4) Data results (3) Data results
Figure 2.8 ESB client app with local database
In Figure 2.8 if publish/subscribe is used, the steps (1) and (2) are subscription messages which are sent
just once.
One particular usage of this is the meter data. In NOBEL GRID there is a server application called Meters
app that (among other things) receives the requests to read data from the deployed meters by using
DLMS. There is another application, the meter data management application (or MDM), that in this case is
acting as a client application of the Meters app, and periodically send requests to read the meters. When
the meter reading is available, the MDM stores it in a Real time database (see Figure 2.9).

Realtime DB Time series db
(8) RT Data changed
(7) store Nobelgrid ESB

meterReading data (2) create meterReading
Meter app queue
(1) create meterReading
MDM app
Meters app
Message routing
Answer queue
(6) Created MeterReading (5) Created MeterReading
Internet
(3) DLMS message for

(4) DLMS data
reading data
Meter
Figure 2.9 MDM - DLMS integration
The Realtime data is a MongoDB noSQL database. On this database, a special feature have been set up that
stores all the changes made in the real time data to a different time series database that holds the previous
data states tagged with a timestamp. This separation of real time and historical measures is a security
measure that prevents the huge amount of historical data to impact on the daily operations.
This MDM app itself is also acting as a server of data for the ESB (this dual nature is not the usual way, but
is possible). In that case the data offered are the meter readings, they are being stored in the databases
(see Figure 2.10).
Time series db
Local DB
(4) Meter
Readings data
(2) get
Nobelgrid ESB meterReadings
(7) save data Meter app queue
(1) get meterReadings
Client APP
MDM app
(3) get meter reading
From DB
Message routing
Answer queue
(6) reply MeterReadings (5) reply MeterReadings
Figure 2.10 ESB query from measures
The client app that demands data can optionally store the data in its own local database. For data analysis,
this will be the usual way.
This is an example of how the DLMS data is integrated in the business processes, but the SMX information
will reach the NOBEL GRID control centers (DRFM, EMA or G3M) by different means: DLMS, IEC61850,
OpenADR and MQTT. For the first three protocols, there will be specific modules that will act as server
Applications from the ESB point of view, and thus translate the data received from the SMX to the ESB

accepted data model (CIM mainly). Different services will be defined for that. The last communication
method is a bit different. MQTT is an ISO/IEEE standard for telemetry data transport. It is widely used to
send distributed sensor data to central repository. It is one of the reference M2M technologies and an
enabler of the Internet of Things (IoT). The technology behind the scenes of MQTT is similar to the NOBEL
GRID ESB technology and in fact MQTT messages has been directly included in the NOBEL GRID ESB.
At the SMX level, the information stored is managed centraly, and two different mechanisms are provided
to manage repository data from the applications at the SMX:
A RESTful Api is provided to send HTTP action on the database
A MQTT messaging framework is provided for data readings
Figure 2.11 SMX inter-module communication
In Figure 2.11, the different applications (encapsulated in docker containers) exchange data with the SMX
core by means of MQTT messaging or RESTful messages.
The strategy selected to let the data flow from SMX to the ESB (and then to the rest of the applications at
the control center), is to make the ESB subscribe to the MQTT messaging platform at the SMX (with some
security consideration). The ESB (that is common to all the control center applications) will subscribe to all
the MQTT data sources of the SMXs available and thus will start receiving the MQTT messages in real time
(Figure 2.12).

Realtime DB Time series db
(7) RT Data changed
(6) store
meterReading data
Nobelgrid ESB
MDM app (4) Created MeterReading

DACF MQTT app
MDM queue
(5) Created MeterReading Message routing
RT DACF MQTT queue
(3) MQTT data available
(1) MQTT (2) MQTT Data changed

subscribe
SMX
Figure 2.12 SMX measures integration in ESB
A special ESB client module, the MQTT app (part of the DACF), will receive and handle the MQTT messages
from the SMX, and is basically a MQTT/CIM translator
The whole process works as the following procedure:
1. The ESB gets subscribed to the SMX (this happens only once per SMX)
2. The ESB receives asynchronously data changes from the SMX.
3. The ESB routes the messages to a queue RT DACF MQTT queue , where an application is listening:
the DACF MQTT client.
4. This application makes some task of data filtering and aggregation and applies some contention to
avoid an overhead in the data storage, then encapsulates the data as a CIM object and sends a
created MeterReading message to the ESB
5. ESB will route this message to the right destination, the MDM app, that will receive the new
available measure object just like the information received by the DLMS process.
6. Data is saved in real time database
7. Automatically, data is also written in historical database.
The developed structure allows to receive from the SMX ALL the data that is flowing through the MQTT
SMX internal messaging. The DACF MQTT app is responsible to generate CIM object from raw MQTT data.
The mechanisms is flexible enouth to capture all the information that is handled at the SMX and make it
available for the rest of the services at the G3M system.

3 DATA ANALYSIS
3.1 INTRODUCTION
Data analysys is essential for being able to manage the distribution grid in a stable and secure way.
This project deploys smart meters based on the Unbundled Smart Meter (USM) architecture, which allows
that all datat available in the metrology part of the meter can be made available in real time or recorded in
detailed logs, to be used for further data analysis activities.
The usual measuremenst which are available are the active and reactive powers (p and q), the network
voltages (Ua, Ub, Uc), currents (Ia, Ib, Ic) and even the measured frequency. Not all these data are available
from each market existing or demonstrator exiting meter, as some of them can provide only some of the so
called instrumentation values. All metering points however deliver at least active power P and voltage
level U, which are essential data for grid observability.
Another important aspect of the recorded data is the time resolution of the acquired and recorded data.
One minute as highest data refresh rate was promised for the project. In most of the deployments new
data each one takes up to ten seconds, so in most cases it is 6 times better than promised. However, the
one minute records are always available in any of the metering points of the project. Moreover, as SMX is
reading the data based on its synchronized internal clock, all data are acquired synchronously, with a
synchronisation which is usually of one second. It is named as synchro-SCADA which was reported in a
peer-reviewed paper [2].
The new SLAM, to be developed in WP5, will make available all these measurements at highest rate of
refresh (each one second).
The chapter describes several data analysis applications, considering process data acquired at rates of 1
minute or less.
3.2 ANALYSIS OF PASSIVE DATA

For the analysis of passive data, data associated with the network topology and the type of cables,
switches, transformers from MV and LV are taken into consideration. Analytical data is given and also
autocad designs and cable, transformers are provided to extract and exam the majority of the required
data. The data needed for the analysis are related with distribution lines, transformers and voltage
regulators, switches, loads and capacitor banks. Below a description of data for each of three test sites
(Alginet, Meltemi and Terni) is presented.
Alginet
The electric cooperative of Alginet (Spain) is a holding form by a DSO and an energy retailer. The electric
cooperative of Alginet is one of the demonstration sites of NOBELGrid project. The aim of the project is to
reduce energy consumption in cities and neighborhoods. Cooperative supplies 46 million kilowatts annually
by means of 35 centres of transformation, with an installed power of 18,000 kW and almost 6 000 users get
benefit from the cooperative services, using smart meters. The company has recently finalized the
installation of a new electrical substation, enabling the capture of energy in a higher voltage level (132 kV.)
for further processing. Red lines in the schemas attached refer to underground MV lines, while blue lines
refer to aerial MV lines.

Figure 3.1 Alginet Test Site
Meltemi
Meltemi is a seaside holiday camp located in Rafina, a town located on the eastern coast of Attica in
Greece. Meltemi comprises 170 cottages which are fully inhabited in the summer (from May to
September) and mostly empty in winter. A typical cottage in the camp is a single floor building of 70 m2
surface. Most of the cottages are more than 30 years old and due to the small size of each cottage, its
electrical consumption is lower than an ordinary house in Greece. However, the ecological awareness of its
habitants and the electrical structure (all houses connected to the same MV/LV transformer) of the
settlement make it ideal for use as a test bed, for functions related to emergency and critical grid
situations. A number of Distributed Generators (DGs) are installed, including a 40kVA diesel generator, 4.5
kW photovoltaic panels and small residential wind turbines that can partially support the Meltemi camping
load in islanded Microgrid operation.
The network consists of a single 20kV/400V transformer with six three phase feeders of maximum length
close to 620m. Network details, like the distribution grid topology and cable type can be found in Figure
3.2. This kind of information will be important for evaluating DR and VPP network operation related KPIs.
The single line diagram of the Meltemi site is presented below .

Figure 3.2 Meltemis single line diagram
Terni
ASM TERNI S.p.A. is an Italian multi utility company, established in 1960 and fully owned by the Terni
municipality, specializing in water, gas, electricity and environmental services. It owns and operates the
local power distribution network, covering a surface of 211 km2 and delivering about 400 GWh to 65,100
customers annually. The ASM distribution network acquires electricity at HV through 3 primary substations
and distributes electricity to the end users through 60 MV lines (with 10 and 20kV ) and 700 secondary
substations . The peak power is about 70 MW and the power distribution network is 587 km MV, 1762 km
LV.

Nowadays, the ASM electric grid is characterized by a large number of distributed renewable energy
sources embedded in the medium and low voltage distribution networks: 41 plants (59% biomass-biogas,
36% photovoltaic, 5% hydro) and 690 solar photovoltaic (PV) units are currently connected directly to the
medium voltage (MV) and low-voltage (LV) distribution networks, respectively, reaching the total
installation load of 70 MW. Moreover a state-of-the-art SCADA system is currently used in remote control
of MW network.
Figure 3.3 Terni s single line diagram
Table 3.1 Ternis Characteristics

TYPE unit 2014 2015
LV grid KM 1425,378 1440,236
MV grid KM 630,291 632,359
hT grid KM 4 4
CABINE MT/BT number 597 601
STAZIONI number 3 3
AT/MT
3.3 ENERGY QUALITY MONITORING SERVICE

Smart meters are becoming the most common equipment nowadays in LV distribution networks, based on
the goal of European Union to have installed 80% smart meters in 2020, as a politically endorsed target
which shows the society engagement towards higher efficiency and sustainability of the energy sector.

Distribution networks and especially the LV networks are experiencing a dramatic change in the era of
Smart Grids deployment, through high penetration of distributed generation and through flexible
consumers and prosumers, in an environment which asks for both high network efficiency and flexible local
markets.
The new meter chips are more and more powerful and the most recent solutions are based on complex
architectures such as dual-core ARM Cortex processors with DSP functionalities, allowing intensive digital
computations for measuring the energy but also for assessing additional quantities based on voltage and
current measurements, such as Power Quality (PQ) indexes.
The main PQ areas of interest for DSOs as well as for the energy end-user are:
a) Voltage level, which is currently described by EN 50160 [3], giving legal limits in a statistical approach
based on P95% levels (EN 50160, IEC 61000-4-7);
b) Voltage harmonic level, through the synthetic THD and through individual harmonics up to the 50th, with
the same statistical approach of P95% (IEC 61000-4-30, IEC 61000-4-7) and with specific normatives for the
statistical level
c) Current harmonic level, analysed with the same algorithms (IEC 61000-4-30, IEC 61000-4-7), but without
a legally required limit [4, 5]
d) Non-symmetries of voltage and current, calculated for the fundamental component or for the rms value,
within a statistical approach [4];
e) Flicker effect, meaning low frequency deviations of voltage rms values, with the same statistical method
(IEC 61000-4-15) [6];
f) Voltage dips and swells, as short-time voltage depreciations, to be classified e.g. according to Unipede
rules;
In addition, the quality of supply is quantified by number and time-lengths of outages, to be used for
calculating SAIDI (System Average Interruption Duration Index) and SAIFI (System Average Interruption
Frequency Index).
Particularly, PQ issues a), b) and e) have standards with severe requirements in terms of computation,
which ask for high cost specialized equipment.
In this context, and based on existing smart meter functionalities, a novel and simplified (light) solution for
performing the PQ analysis, which is tailored to be adapted for the latest Smart meters design, thus
allowing highly informative PQ monitoring in every Smart Meter. In the project are considered and
proposed agile methods for a) and b), with possible extensions also for d) issues.
In this project, energy quality monitoring service is approached especially in terms of voltage levels (a) in
low voltage networks, with possible application on their harmonics (b), as well with possible extension for
voltage asymmetries, current asymmetries and time of voltage failures at customer levels; this feature will
be obtained through the extensive use of meters real-time values (especially voltage values) and its related
events (e.g. power failure)..
The Unbundled Smart Meter uses extensively the instrumentation values, and relies on part of them to be
used for assessing energy quality. The following specificities can be used for this purpose:
Instrumentation vales can be usually obtained from the meter each 1 second to 10 seconds,
depending on the meter type and on the used protocols. Most of the meters can provide meter
data in 1 to 10 seconds. Shorter time periods can be obtained with RS232 or RS485 communication
interface and with DLMS protocol
Instrumentation values are the only data which are used for any energy / power quality (PQ)
assessment.

Some meters provide also events for power-up / shut-down, which can be used for voltage failure
analysis
All information needed for PQ assessment is read from SMM by SMX and is recorded on disk in
daily files which can be processed off-line. PQ assessment can be made therefore offline with
special applications which reads recorded daily files, after the day is finished, and calculates the
different PQ indexes, to be further described..
Some existing meters do not have all available instrumentation data to allow calculation of some
PQ aspects; for instance, voltage and current asymmetries require not only voltage and current
levels at each phase, but also angles between voltages (U1-U2, U1-U3, U2-U3) and between
voltages and currents (U1-I1, U2-I2, U3-I3), which are not always available to be read by SMX from
the metrology meter (SMM).
3.3.1 Voltage profile

In order to use available data from the smart metrology meter (SMM), in particular the voltage values
obtained by SMX each 1 to 10 seconds, in the project is used a simplifed method which is named Light PQ
Assessemnt Method (LPQAM).
Comapring with this light method (LPQAM), todays PQ standards ask for continuous measurements on
series of 10xT0 or 12xT0 time windows (T0=20ms for the grid frequency of 50 Hz), which are further
quadratically mediated in 10 minutes values; this approach is equivalent to a continuous RMS calculation
on a 10 minutes time period of RMS voltage and its harmonics, with frequency synchronization which is
possible, at the best, on each next 10T0/12T0 period. Finally, a statistical approach is applied on the 10
minutes periods results, over one week, using up to 1008 intervals to assess the 95% statistical levels. This
statistical results are supported by several norms or network operation codes.
The LPQAM uses the meter real-time data (instrumentation values) and is able to address a small but
essential subset of PQ aspects, by proposing simple rules to be considered within the PQ assessment:
A) Use of probability distribution (PD) instead of RMS mediation
The RMS (Root Mean Square) mediation of the 10T0 values is equivalent to a continuous integration over
10 minutes of the RMS values corresponding to voltage waveform. The result is a compression of all 10T0
values over the 10 minutes interval. For sensible loads the true effect of voltage variation is hidden by the
long-time averaging process over 10 minutes.
For better addressing this issue, instead of keeping track of the 10 minutes RMS values and make further
statistical computations on larger intervals, such as one week, a full linear statistics is considered in our
approach, without any quadratisation of interim results, based at least on one second measurement (or
less), to be statistically analysed on each one hour interval, thus being tailored also on commercial intervals
for the energy trading.
It is proposed, as a practical method to obtain probability distributions (PD), the sorting method presented
in Figure 3.4 below, which is also suitable as a simple computation approach easily to be implemented in
Smart Meters processing units.

Figure 3.4 Distribution of probability (DP) graphic using a sorting method
The figure shows the sorting method in a graphic and intuitive way. The P95% limit means that 95% of the
measurements (showed as red balls) are placed in its left part. Each ball is introduced in a certain slot
based on the following algorithm:
- It is considered a certain voltage band for which the PD wil be constructed, e.g. between UMIN=175 and
UMAX=275 V. It means that the total band is UBAND_PD = UMIN-UMAX = 275-175 = 100 V
- A number of slots are chosen, in ordedr to cover this voltage band, e.g. NSLOTS=250, as in figure above. It
means that each slot will correspond to a voltage small band of 100/250 = 0.4 V. For better resolution,
the number of slots may be increased, e.g. 400, which will bring smaller quantas for each band,
meaning 0.25 V.
- A certain voltage measurement will be classified (sorted) by a ball in a certain slot, with the formula:
K(t) = INT [ (U(t) - UMIN)* NSLOTS / UBAND_PD] (3.1)
Where K(t) = the slot number where a new ball will be placed; U(t) = voltage value to be sorted
To be noted that K is the intreger part of the value obtained with this formula, ronded at lower integer.
As an example, if SMX gets from SMM a value U(t) = 232 V, the slot number K(t) will be:
K(t) = INT [ (232 - 175)* 250 / 100 ] = INT [142.5] = 142 (3.2)
If K(t)<0 , then K(t) = 0, meaning that all voltage values below 175 V are sorted as being of 175 V. It
means that the first slot means in fact 175 V or less.
If K(t)>250 , then K(t) = 250, meaning that all voltage values above 275 V are sorted as being of 275 V. It
means that the last slot means in fact 275 V or more.
- After all voltage values have been sorted, the PD is ready to be stored and for further analysis. For one
voltage level measurement per second, there is a total number of 3600 balls each one hour; for a
measurement each 5 seconds, there are 3600/5=720 balls, a number still big enough to bring accurate
statistics for calculating P95%.
The following first rule is considered for the light PQ assessment method (LPQAM):
Rule 1: The PQ assessment can be based on calculated probability distributions (PD) of voltage fundamental
values which are measured at small and fixed intervals - for example each 1 second, which are recorded
each 1 hour - as usual energy commercial interval

There are many advantages of recording this linear PD sets as raw data, particularly because several PQ
aspects can be inferred from it. The major advantage consists in the full linear behaviour of the method: by
recording PDs for 1h (each hour), one can get, by simple PD summation (of PD values on each subinterval),
the values for PD_2h and PD_1w (each 2 hours or one week).
B) Sampled time windows instead of continuous measurements
Today's standards ask for continuous measurement of voltage waveforms, such that in a second one
measures all 5 windows of 10T0 (5 x 200 ms at 50 Hz) and in a 10 minutes interval all 3000 x 10T0 intervals.
In [7] it was shown that measuring only one window, from all 5 windows during each second, brings similar
results as considering all windows, with an error less than 0.2% for voltage fundamental and for harmonics,
due to the phenomenons statistical behaviour. It means that it is not necessarly needed a continuous
measurement, and that a number of voltage samples which do not cover continuous time but are in a high
number during the commercial interval, still give accurate enough results for voltage P95% level.
Considering this, it will be used the following second rule for LPQAM.
Rule 2: The PQ assessment can use meter sampled voltage measurements in a high enough number to allow
statistics over a commercial interval, for example a meter measurement each 1 to 5 seconds.
If the meter can give a new voltage measurement each 1 second, there are 3600 measurements per hour,
thus a good enough PD can be contructed, in order to extract the P95% voltage level.
Some project existing meters can support this reporting rate of the voltage measurement, e.g. ZMD405
meters in Alginet and Terni, the Circutor meters in Alginet and some meters connected to pre-deployed
SMX in various places for testing purposes (e.g. University Politehnica Bucharest, University Lucian Blaga,
Electrica and Transelectriva metering points, Imperioal College in London).
The new SLAM, under development, will be able to give also voltage values each second.
Some other meters, especially those which can communicate on IEC62056-21 protocol (ex. IEC1107), have
reporting rates of 5 to 10 seconds, e.g. the E230 meter which will be deployed together with SMX in the UK
demonstrator.
The Flemish demonstrator will use some meetring points where meters have a so called P1 interface, which
sends new data each 10 seconds.
In all cases, a number of 3600 down to 360 sampled volatges over one hour can be sued for calculating the
hourly PD. It is recommended for a more accurate P95% calculation the 1 second sampling, but acceptable
statistical results can be obtained also with 5 and 10 seconds reporting periods.
To be noted that the new Romanian Metering Code (valid stating with July 2016) asks for the meters to
record voltage values each 15 minutes, for a backup PQ assessment over long periods (there are 2880
voltage samples per each month, thus allowing some PQ assessment even based on such long reporting
period, of 900 seconds).
C) Compatibility with EN 50160
The LPQAM method simplifies the sampling procedure and brings a new PQ assessments statistical
approach, by using instrumentation data from meter, obtained at reporting rates of one to 5 or 10 seconds.
With these rules, a new methodology for voltage monitoring can be outlined, able to be deployed into the
new SM generation in LV networks, by using the LPQAM.
However, a compatibility bridge with EN50160 is still needed, and thus Rule 3 and equation (3.3) below
address this issue.
Rule 3: LPQAM assessment can be compared with existing PQ results through the following equation:
10_ = =1..(_ ())2 / (3.3)

where the UMETER_AVAIL(k) samples, available from the meter over 10 minutes (assuming N values, with k
from 1 to N), are used in a quadratic formula to get the RMS value of U over 10 minutes. For a one second
reporting rate of voltage measurement the number of values will be N = 60 x 10 = 600.
The difference from the EN50160 formula relays on the fact that in LPQAM are used only samplings of U
according to Rule 2 (high enough number of values to allow statistics over a commercial interval) and not a
continuous and complete calculation of all 10T0 windows, as EN50160 requires.
With these simplifications, which fit within meters' complexity, allow small errors under 0.5% in all 10
minutes intervals over one week period, for the measured situation,which wil be further shown.
With these rules, a new methodology for voltage monitoring is implemented, able to be deployed into the
new SM's generation in LV networks, using the LPQAM.
The Light PQ Assessment Method, including the proposed three rules for PQ assessment in the new
generation of Smart Meters for LV/MV networks, has also been discussed in the CIGRE WG C4.24 Power
quality and EMC issues associated with future electricity networks, which is a joint working group of CIGRE
and CIRED, operating in close cooperation with an IEEE working group covering the same subject. C4.24
obtained its mandate in 2013 and is expected to deliver its final report by the end of 2016. The scope of
C4.24 includes, among others, the study of positive and negative impact of new smart distribution
applications on power quality in the distribution grid.
The work of Joint Working Group C4.24, is focused on several important subjects [8] one of them being the
new measurement techniques. In future networks this issue will play a crucial role and the balance
between traditional PQ monitoring, with dedicated PQ instruments, and non-traditional devices, such as
meters with PQ functions, must be achieved [9]. The foreseen solution relies on a modular concept,
allowing stepwise increase of PQ monitoring functionality, for smart meters with increased flexibility and
reduced costs.
The accuracy of new LPQAM has been proven by measuring in parallel the supply voltage RMS with a PQ
portable monitoring instrument [10] and with an Unbundled Smart Meter (USM) equipment which records
the voltage each 1 second.
Taking into account the traceability for both instruments, their major specifications are:
1) The portable monitoring instrument equipped with GPS receiver is compliant with the Power Standards
Lab USA, IEC 61000-4-30, edition 2, Power Quality Measurement Methods Compliance Report, operating at
230V, 50/60Hz, Power Quality parameters power frequency and magnitude of the supply voltage
conform to class A requirements.
2) The USM equipment contains a full Linux machine on the Smart Meter Extension (SMX) part, allowing
meter data acquisition synchronized by Network Time Protocol (NTP) through internet connection, with
less than 100 ms. time synchronization. All internal instrumentation values URMS, IRMS, P, Q and frequency
f have a maximal relative error less than the meters accuracy of 0.5. For this, the instrumentation values
read from USM equipment have been previously compared with the values recorded by the high accuracy
three-phase Portable Reference Standard class 0.02 [10], at the laboratory of Romanian National Institute
of Metrology
Both units [10, 11] have been installed in parallel, without transducers, in the same common coupling point
belonging to an actual consumer connected to the low voltage distribution grid, without transducers.
Measurements were performed on three-phase supply system, where voltages were measured between
phase conductors and neutral (line-to-neutral) for all three phases.
A nearly complete 7 days, and several additional one day measurements in two different points of an LV
network of Sibiu town (Romania) have been made with both equipment.

The supply voltage variation during the first period is presented in Figure 3.5 and shows many situations
with high level of variation of the voltage level.
Figure 3.5 Measurements with IEC61000-4-30 equipment, URMS each 10 minutes
In Figure 3.6 below the voltage of phase 3 is compared for both types of measurements: using the
IEC61000-4-30 compliant to class A requirements [10] and in the new LPQAM method.
Figure 3.6 Comparison between IEC61000-4-30 and LPQAM methods over approximate 7 days, URMS
each 10 minutes
It can be seen that the two types of measurement give very similar values over one week.
The advantage of having probability distributions (PDs) based on e.g. one second voltage values rely in the
fact that PD can be calculated each hour. If PD is calculated each hour, P95% can be also calculated each
hour, thus being able to associate it with the commercial interval for energy and allowing new business
models where energy is coupled with the light PQ assessment. Figure 3.7 below shows a superposition of
PDs over three consecutive hours of the same day, showing how the voltage level depreciates during the
evening peak hours (19-20 and 20-21 hourly intervals).

Figure 3.7 Hourly PD over three consecutive hours, calculated with voltage values recorded by SMX
and an LV connected ZMG310 meter
The Figure 3.8 gives as an example a 3D image showing an entire day having PDs calculated each hour.
Figure 3.8 Hourly PD over an entire day (24 hours), calculated with voltage values recorded by SMX
and an LV connected ZMG310 meter
Voltage level and continuity of supply (next chapter) are essential PQ criteria, as can be also seen in the
Italian latest requirements for PQ in Smart meters[12].
Some further examples are given (Figure 3.9-Figure 3.14) of daily voltage profiles obtained with exiting
meters and with pre-deployed SMX, recorded in various places (in the demonstrators and in other testing
sites).

Figure 3.9 Daily voltage profile on the MV part of HV/MV transformer in Alginet (secondary values)
Observation: pre-deployed SMX acquired data from an existing ZMD405 meter from Landis+Gyr, acting as
SMM
Figure 3.10 Daily voltage profile on the LV part of an MV/LV transformer in Alginet
Observation: pre-deployed SMX acquired data from an existing Cirwatt meter from Circutor, acting as SMM

Figure 3.11 Voltage profile during PV production in a selected day, Terni
Observation: pre-deployed SMX acquired data from an existing ZMD405 meter from Landis+Gyr, acting as
SMM. The meter is supplied only during PV production, thus only during the day is possible to measure the
voltage level. This is a particular type of connection in Terni.
Figure 3.12 Voltage profile on LV of a UK consumer, recorded each 10 seconds
Observation: pre-deployed SMX acquired data from a market existing E230 meter from Landis+Gyr, acting
as SMM. Acquisition made with IEC62056-21 (ex. IEC1107) protocol, each 10 seconds, with an RS485
connection.

Figure 3.13 Daily voltage profile, UPB, 19.02.2016
Observation: pre-deployed SMX acquired data from an laboratory based ZMD310 meter (Landis+Gyr),
acting as SMM of the Unbundled Smart Meter.
Figure 3.14 Daily voltage profile, UPB, 19.02.2016, detail for the interval 7:00 8:00
Observation: pre-deployed SMX acquired data from a ZMD4310 meter from Landis+Gyr, acting as SMM.
Acquisition made with DLMS protocol, each one second.
3.3.2 Voltage failures

Voltage failures are addressed in USM by using as basis for the assessment on events related to power-up
and power down, meaning that the situations which can be observed are in fact failures in power supply.
These events are usually provided by the meter itself (the SMM part of USM), so external processing of
event log of the meter can calculate the number of voltage failures (three phased and / or mono-phased).
As a less accurate offline alternative, the same archived log of values can be used for voltage failure
calculation, but this is acceptably accurate only if the meters can be read at the highest rate, e.g. 1 second.
For 5 or 10 seconds readings, only larger voltage (and power supply) failures can be recorded.
In order to obtain voltage failure information, a special procedure will process in SMX each archived daily
file, after the day is finished and no records are anymore appended to the file. Specific data will be
obtained, based on this processing:
- number of failures and
- total period of failures (in seconds),
as basis to be able to calculate separately, over longer periods such as a year, the Momentary Average
Interruption Frequency Index (MAIFI) and the Customer Average Interruption Duration Index (CAIDI) for the
given customer.
CAIDI gives the average outage duration that any given customer would experience and can also be viewed
as the average restoration time, which is usually measured in minutes or hours. In order to be able to
compute these values based on archived values, SMX need to be supplied even during an outage situation,
which can be usually found in substations, where alternative power supply may exist. For this reason, the
most secure source for calculating these values remain the SMM itself.
3.3.3 Voltage asymmetry

Voltage asymmetry require both voltage values (modules) and the angles between them. Only some of the
meters provide this complete set of values, e.g. some of the Landis+Gyr meters.
For a three phased system with voltages Va, Vb and Vc, the symmetrical values are V1 (direct component),
V2 (inverse component) and V0 (homopolar component), the relation between then being:
0 1 2 1 1 1 0
,, = [0 ] + [ 1 ] + [ 2 ] = [1 2
2
] [1 ] (3.4)
0 1 2 2 1 2 2
2
where = 3 .
The symmetrical values can be obtained by using the reverse matrix:
0,1,2 = 1 ,, (3.5)
where
1 1 1
1
1 = 3 [1 2 ] (3.6)
1 2
which requires calculations in the complex domain (voltages are phasors, having modules as well as angles).
Both voltages and their angles can be acquired in the ZMD meter mounted in UPB (having a pre-deployed
SMX). None of existing meters which could be integrated in the demonstrators have both set of values.
However, SLAM will allow the SMX readout of both voltages and their angles, thus being able to calculate
voltage asymmetries according to the formulas above.
3.3.4 Current asymetry

Similar with the previous values, current asymmetry require both currents modules and angles between
them on all phases. Even fewer of the existing meters provide such values.
For a three phased system with currents Ia, Ib and Ic, the symmetrical values are I1 (direct component), I2
(inverse component) and I0 (homopolar component), the relation between then being:
0 1 2 1 1 1 0
,, 2
= [0 ] + [ 1 ] + [ 2 ] = [1 2 ] [1 ] (3.7)
0 1 2 2 1 2 2
2
where = 3
The symmetrical values can be obtained by using the reverse matrix:
0,1,2 = 1 ,, (3.8)
where A-1 has the same meaning as for voltage asymmetry:
1 1 1
1
1 = 3 [1 2 ] (3.9)
2
1
and requires also calculations in the complex domain (voltages are phasors, having both module and angle).
Both currents modules and their angles have been acquired from the same ZMD meter mounted in UPB
(using the pre-deployed SMX). Similar with voltages, none of existing meters which could be integrated in
the demonstrators have both set of values. However, SLAM will allow the SMX readout of both currents
and their angles, thus being able, when available, to calculate current asymmetries according to the
formulas above.

4 FUNCTIONALITIES AND TOOL DEVELOPMENT
4.1 INTRODUCTION
Secure operation of power systems requires a comprehensive understanding of the system operating
status, which helps in identifying potential critical operating conditions and determining the necessary
preventative measures. Power system load flow analysis and state estimation (SE) play an important role in
secure operation of power systems [13]. It provides an optimal estimate/forecasting of the system states
based on continuous monitored measurements, pseudo-measurements, load estimation and forecasted
load demands etc, helping the operators identify current network states or potential network states. With
the increasing penetration of distributed renewable energy resources, the connection of medium-sized
distributed generation (DG) and more and more electronic interfaced devices (e.g., different types of load
storage and electric vehicles) connected in the distribution networks, it is required to have an improved
observability of the distribution networks in order to ensure secure operation of distribution systems. This
led to the intensive research and application of three-phase load flow and state estimation at distribution
levels.
4.2 STATE ESTIMATION

State estimation (SE) is particularly relevant to DNOs wishing to gain greater visibility of their network as
part of a future smart grid. SE provides an optimal estimate of the system states based on both continuous
monitored measurements and pseudo-measurements. The concern of state estimation was first addressed
in 1970s [14], and later with the increasing capability of SCADA system computers and establishment of
Energy Management Systems (EMS), this functionality was widely integrated in EMS and used for operation
and management of transmission systems [15]. Different from meshed transmission networks, most
distribution networks are radial, often with high R/X ratios. Besides, application of real time measurement
is very limited in distribution networks. The SE approaches developed for transmission networks such as
decoupled methods and DC approximations cannot be applied to distribution system state estimation
(DSSE) directly [16]. Furthermore, ill conditioned matrices and large number of nodes in distribution
networks imposes great difficulty to DSSE. Different from transmission system SE which is to improve the
reliability of the measured values, the usual purpose of DSSE is to extend the observability of the network.
With the rigorous requirement of observability in distribution networks, the development of DSSE has
attracted great attention recently. Various techniques have been investigated for DSSE in literature, such as
Weighted Least Squares (WLS) based SE [13], fuzzy SE [17] and branch-based SE [18], and so on. Besides,
various types of data in distribution networks including Advanced Metering Infrastructure data have been
extensively explored for the purpose of DSSE [16, 19-25].
4.2.1 SE problems
The state estimation problem is defined by:
= () (4.1)
where is a vector of measurements, () is a nonlinear set of equations that describes the true state of
the power system with state variable , and is a vector of errors between the observed measurements and
true state of the system. ~(0, ), where is the covariance matrix of the measurement errors ().

4.2.2 SE processes
SE processes are illustrated in Figure 4.1. First, topology processor verifies the up-to-date network
parameters and configures the one-line diagram of the network; observability analysis tool identifies the
un-observed branches in the network and establishes the required but missing information, often referred
to as pseudo measurements; given all available input data including topology, measurements and pseudo
measurements, the state estimator searches for an optimal estimate of the system state; bad data
processor identifies and eliminates the data affected by gross errors and noise. To continuously update and
improve the pseudo measurements, the estimated results can also be used as feedback to adjust the
methodologies of load estimation and load forecasting, and form a close-loop information flow.
P, Q
Pseudo
Load Measurement; Calculation
estimation/ Virtual
forecasting measurement V,
State
Real time estimator
measurement Bad data
V, P, Q, I, processor
Breaker
positions Topology Observability
processor Analysis
Figure 4.1 Illustration of SE processes
The most advanced techniques and applications of DSSE in literature.

1) Load Estimation
In distribution networks, the lack of real measurements is compensated by the use of information from the
loads. Thus a high proportion of the measurements can be pseudo-measurements. Load estimation (or
pseudo-measurements) plays an important role in DSSE. Computation intelligence methods have been
investigated for DSSE. For instance, artificial neural network has been adopted to generate pseudo
measurements based on a few real measurements in conjunction with typical load profiles [26]. In [27],
machine learning methodologies have been investigated to provide reliable input information to a robust
state estimation algorithm. Real time estimation of loads can also be obtained based on iterating between
WLS estimator and load flow [28].
2) Use of Measurements
More comprehensive and accurate knowledge of the system is required in order to make efficient and
reliable control actions. In particular, attention should be paid to phase angles estimation which can be
used to avoid critical situations. In this context, the use of phasor measurement units (PMU), which
provides synchronized local measurement with global time stamp, is promising. PMUs measure not only
voltage phasors but also current phasors through all incident buses. With the increasing deployment of
PMUs in the distribution grids, state-of-the-art SE are implemented using data from PMUs [19, 20]. The

benefits of the use of PMU data and the approach of integrating PMU data in DSSE have been extensively
investigated in literature [29, 30].
The roll-out of smart meters means that an unprecedented amount of detailed historical data on user loads
is becoming available. This data can be used to better understand and model the behaviors of distribution
network loads, allowing to improve load estimation techniques, and ultimately, DSSE accuracy [16]. Studies
have been made into the incorporation of smart meter data in DSSE to estimate flows, voltages, and losses
in low voltage distribution networks [21-24].
3) State Estimator
Various advanced computation intelligence techniques have been extensively explored for DSSE. Auto-
associative neural networks (autoencoders), which only require historical database and few quasi-real-time
measurements to perform an effective SE, are used to learn the behavior of the grid and avoid the need of
characterizing the grid parameters and topology [31]. Hybrid particle swarm optimization was adopted to
estimate bus voltage magnitudes and angles, and discrete tap values of on-load tap changers [32]. In [27],
machine learning methodology collaborating with SE forms a closed-loop information flow in order to
continuously update and improve the performance of the state estimator.
Since distribution systems are typically very large and dense, to improve the calculation efficiency,
distributed SE (also named multi-area) is investigated for processing separate sub-systems in parallel, in
conjunction with a certain level of coordination or agent communication [19, 33, 34]. In order to
accomplish very large-scale monitoring of interconnected power systems, multi-level, or hierarchical SEs,
are also developed to integrate existing SEs that are designed to function at different levels of modelling
hierarchy [35]. These distributed SEs can greatly enhance the computational performance and improve the
reliability of SE algorithms.
To allow the system operators to have more time in making control decisions, FASE approaches are
attracting more and more attention in DSSE, particularly when high-resolution data from synchronised
metering devices such as PMUs are available. To make a more accurate mathematical description of the
time evolution of the state, attempts have been made to generate a more accurate mathematical
description of the time evolution of the state using Kalman filter and other techniques [19, 36, 37].
4.2.3 Distribution System State Estimation

Distribution system state estimation (DSSE) can be considered as an extension of the standard load flow
equations which makes them solvable within the context of a practical power system. The objective of DSSE
is to establish the current state of the network by minimising the least square error in a set of
measurements and estimates taken from a network. The output from a DSSE algorithm is a set of voltages
and angles which define the state of the network.
There are several important differences between DSSE and load flow. In DSSE, the number of equations
which describe the system may be more than the number of state variables. This is described as an over-
determined system, and it has the result of making the Jacobian matrix rectangular. Unlike load flow, the
set of equations for DSSE may also include line flows, voltages, and currents. A DSSE Jacobian matrix
therefore also includes derivatives of real and reactive line flow, voltage or current. The DSSE equations can
also be weighted by the accuracy of the measurements, whilst load flow implicitly assumes that each
power and reactive power measurement is equally accurate.

4.2.3.1 Three-phase state estimation
In this study, the the weighted least squares (WLS) and generalized least squares (GLS) [38] state estimation
technique was used. The objective function of the GLS / WLS estimation problem is shown as
[ ()] [ ()] (4.2)

The state estimation equations can be solved iteratively, using the Newton-Raphson method. The update
equations are defined in equation

+ = + ( ) [ ()] (4.3)
()
() = (4.4)

Where + is the estimate for the state variables at the (k+1)th iteration. is the Jacobian matrix.
= () (4.5)
The DSSE equations are formulated to solve for the state vector . The state vector fully defines the state
of the system. It typically contains a set of voltages for all buses in the network; angles and magnitudes or
real and imaginary components. The measurement vector can be configured to include any measurement
such as three phase active and reactive power demands, three phase active and reactive power flows, as
well as voltage magnitudes, and currents.
4.2.3.2 DSSE Measurements

The accuracy of a DSSE formulation is dependent on the quality of the measurements available in the
power system. Both real measurements and estimated measurements can be incorporated into a DSSE
formulation. Estimated measurements can be subdivided into two further groups known as virtual
measurements and pseudomeasurements. To work accurately, a DSSE formulation must be fed with
accurate estimates of the expected error distributions of real virtual pseudo-measurements
1) Real measurements: Real measurements cover all measurements taken by monitoring devices. This can
include real and reactive power consumption, voltage and current. The standard deviation of a real
measurement is dependent on the accuracy of a monitoring device and can be modelled. The
measurement error of a monitor has a very low variance. Each real measurement was modelled with a
standard deviation of 0.2% of the mean (or scheduled) value of the measurement. Measurement errors
were assumed to be independent from one another and other types of measurements.
2) Virtual Measurements: Virtual measurements are used when the value of a measurement is known with
almost near certainty. Virtual measurements are used when the value of the real and reactive power
injection at a busbar is defined. An example of a virtual measurement is an unloaded bus, which is defined
with zero real and reactive power injection. The virtual mea surements standard deviation should be small
enough to so that the weighting (and, hence, importance) assigned to virtual measurements is large
relative to those assigned to real PM, but large enough to avoid problems inverting the gain matrix

( ). The standard deviation used in this research was 2 107 and is similar to that used in
[39](1 107 ). Virtual measurement errors were assumed to be independent from one another and
other types of measurements.
3) Pseudo-Measurements (PMs): A PM should add information to the state estimator to enable full system
observability. Pseudo-measurements are estimates for the value of a parameter based on some model of
system behaviour. PMs can be constructed for any measured quantity of the power system (e.g., voltage,
current, power, etc.). In this research, PMs are limited to estimates of the real and reactive power injected
at a busbar. For example, it may also be possible to predict the load at certain busbars by building a model
based on historical analysis. The electrical load profile follows a fairly predictable pattern which can be
estimated through knowledge of external factors such as ambient temperature, weather conditions, time
of year, day of the week, and the types of customers connected to a busbar. Voltages can also be predicted
to within certain ranges if voltage control equipment is installed at a busbar.
A good pseudo-measurement should have a number of properties. A pseudomeasurement should add
information to the state estimator to enable full system observability. Pseudo-measurements should have a
low standard deviation, and a standard deviation which can be quantified. The error of a pseudo-
measurement should (ideally) be uncorrelated with other pseudo-measurement errors, and follow a
normal distribution (as the WLS / GLS procedure assumes that measurement errors are normally
distributed). It is also convenient if the pseudo-measurement can be easily estimated using a statistical
model, or engineering judgement.
Scheduled Power (SP) - The most basic form of a pseudo-measurement is to assume a value based on a
historical average at that point in the network. Scheduled power (SP) has the advantage of being easy to
estimate. However, the variances of the pseudo-measurement errors are high, and there is a high level of
correlation between pseudo-measurements at different busbars. SP estimates can be used for three phase
state estimation by assuming power consumption in each phase is balanced, and equal to a third of the
overall three phase power. The power factor is also assumed constant across all phases.
Load Estimation (LE) - LE concerns estimating the nodal load level at a bus. Many factors will influence the
load profile at a bus including ambient weather conditions, time of day, season, and the types of customers
(for example residential) connected to a bus. Utilities store several pieces of information which can be used
to estimate the load at a busbar including peak load data, customer billing data and load curves for
different customer types.
Mixed Model (MM) Pseudo Measurements - In some situations, a limited amount of monitoring may be
available at a site. For example, monitors may be installed which measure power consumption in phase A
only. In this scenario, a mixed model (MM) can be created using a mixture of real and estimated
measurements. Knowledge of the power consumption in one phase can provide significant insight into the
estimated power consumption in the other two. A distribution system that supplies only three phase loads
will show a high level of correlation between real and reactive power in each phase. A system which
supplies single phase loads will show lower levels of correlation between the phases. A real distribution
system contains a mixture of single and three phase loads. The expected level of correlation between real
and reactive power can be estimated by considering the correlation between phases at a typical bus. The
aim is to construct a multi-variate normal distribution that encompasses dependencies of the variation of
each of these variables. Using this distribution, the active and reactive power consumption in phase i (P(i)
and Q(i) respectively) can be determined.

4.2.3.3 Non-Linear Power Flow Equations
The non-linear power flow equations describe the set of equations in ().

There are 5 different types of measurements: real ( ) and reactive power ( ), voltage magnitude ( ),

real power flow ( ) and reactive power flow ( ) in a line. Each of these measurements can be
formulated in terms of magnitude and phase of the voltage at all buses within the network. The equations
which relate voltages at each bus for the pth phase for power injections are shown
(4.6)
(4.7)

where the pth or the mth phase is a (1st index), b(2nd index), or c(3rd index), is the real i, kth element

of the three-phase admittance matrix Ybus, and is the imaginary i, kth element of Ybus.
The following two equations describe the line flow equations, where , and , are elements of
() () () () () ()
36 line admittance matrices GL and GL relating voltage ( = [ , , , , , ] )
() () ()
and current ( = [ , , , , , ] ) between buses i and j in the form Iij=( GL +j GL)Vij, where Iij is
a 31 vector representing the current flows in phases a, b and c from bus i to j and Vij is a 61
vector representing the voltages at bus i in all three phases and the voltages in bus j.
(4.8)
(4.9)
To build the Jacobian matrix (), the derivatives of the measurement equations are evaluated with
respect to each state of the system in exactly the same way as for load flow. The equations are not shown
for brevity, but can be derived by differentiation of (4.6), (4.7), (4.8), and (4.9).
Admittance Matrix: Formulation of the three-phase admittance matrix Ybus is not covered in this paper for
brevity. The reader is referred to [40] and [41] for further details on how to form the three-phase
admittance matrix. The Ybus admittance matrix was assumed to be constant ((in between tap change
operations of the onload tap-changing transformers). Where admittance subcomponent matrices
were modeled initially in the sequence domain, their subcomponent matrices (Ybus) were converted

into the phase domain by applying the Fortescue transformation [42] to Ybus to obtain a phase
domain subcomponent matrix.
= 1 (4.10)
where is the transformation matrix to convert sequence components into the phase domain.
Transformer models: The network studied in this research contains several transformers. These
devices were modeled as onload tap-changing transformers (OLTCs). Transformer taps were assumed to
move the voltage on the secondary side of the transformer as close as possible to the voltaje setpoint. All
transformers were assumed to be solidly grounded on the Y side of the transformer. The three-phase
admittance matrix for the transformer was formed by taking into account the positive-, negative-, and zero-
sequence admittances of the transformer, and described in detail in [43].
Three-Phase Components: Three-phase components (such as cables and lines) were modelled by forming a
nodal admittance matrix from their positive, negative-, and zero-sequence impedances. The positive- and
negative-sequence impedances were assumed to be equal. Coupling between zero-, positive-, and
negative-sequence impedances was assumed to be zero within each cable and line subcomponent model.
Building the Covariance Matrix R: The covariance matrix is different depending on whether the WLS or GLS
estimation procedure is used. The covariance matrix contains the information relating to the accuracy of
measurements used in a DSSE formulation. The covariance matrix for WLS is diagonal. The diagonal
elements of the matrix R are described in
= 2 (4.11)
Where is the ith diagonal element of R and is the standard deviation of the ith measurement.
With WLS, all measurement errors are assumed to be independent. To build the covariance matrix for the
GLS formulation, the correlation between pseudo-measurement errors must be taken into account. Real
measurements are always assumed to be uncorrelated from one another. The correlative nature of
pseudomeasurements can be explored through historical analysis of measurement data. The measurement
error models can be incorporated into the state estimation equations by altering the measurement error
covariance matrix R. The errors in the pseudo-measurement models and the mixedpseudo measurement
model developed in this chapter are all highly correlated. The correlation can be conceptualised by
considering a simple example. Consider a scheduled power (SP) pseudo-measurement made at two busbars
in an arbitrary network. At times of peak load, the SP pseudo-measurement is likely to under-estimate both
loads, and at times of light load, the SP pseudo-measurement is likely to over-estimate both loads. Thus,
the errors in the pseudo-measurement error will be strongly correlated. It is therefore important to take
into account the covariance of both inter and intra busbar measurement errors [44], as well as
understanding their typical variance, and whether or not the errors are normally distributed.
The output of the distribution system state estimator is a set of voltages and angles at all of the buses in
the network. A distribution for the voltage and angle at a busbar can be obtained if the error properties of
the measurements are known. The state estimator can be run repeatedly using a set of measurements
which are taken from within the distribution of the measurement errors. This generates a set of solutions
to the state estimator which can be plotted as a probability distribution. Care must be taken to ensure that
any correlation between pseudo-measurement errors is also taken into account.

4.2.3.4 Uncertainty and Estimation Accuracy
The deviation/uncertainty of input data/information impacts the accuracy of DSSE to a certain extent.
1) Measurement uncertainty: DSSE works with a limited number of real measurements. Measurement error
and bias exist in each measurement. The measurement can be characterized by its own range of
measurement errors which is primarily determined by the corresponding measurement devices [45]. Due
to the lack of real measurements, pseudo-measurements play an important part in DSSE. The deviation of
both real and pseudo- measurements can greatly affect the accuracy of system state estimation. A number
of dedicated papers provide the analyses regarding the impact of measurement uncertainty (pseudo-
measurements and heterogeneous measurements) on SE accuracy [46-48]. Reference [49] analyses the
relationship between the measurement errors existing in both load pseudo measurements and real-time
measurements and the deviations of the system state estimation, presenting how increasing the load
measurement accuracy will improve the estimated deviations of bus voltages. In [48], the impact of
different types of measurements and the accuracies of the measurement devices on accuracy of DSSE is
analysed. In [50], the classic WLS based SE is analysed to point out under which circumstance and to what
extent SE results are affected by measurement uncertainty when a minimum number of measurements
which is just enough to assure full observability is used. It provides a straightforward criterion to predict the
overall worst-case SE accuracy or, dually, to establish the maximum measurement uncertainty that is able
to keep the average or worst-case estimation accuracy within given boundaries.
2) Uncertainty of operating conditions: The impact of the uncertainty of the network behavior (such as
distributed voltage control and demand side management) on state estimation is analysed in [48], which
demonstrates that investment in the improvement of the knowledge of the distribution grid behavior is
inevitable in order to improve the accuracy of DSSE.
3) Network parameters uncertainty: Network parameter uncertainty would also impact the accuracy of
DSSE. The impact of network parameters uncertainty on state estimators bias is analysed in [51], and the
tolerance of the value of the network parameters is analysed by means of suitable Monte Carlo procedures
in [52].
4) Topology uncertainty: In distribution networks, the network topology in operation can change very
frequently. For instance, switching is usually used as a resource to improve system operation or to reduce
the impact of outages. The frequent topology changes in distribution networks together with the reduced
number of installed measurement devices can lead to situations in which the topology is not known beyond
any doubt even after running the topology processor. The impact of topology uncertainty on estimation
accuracy is analysed in [53]. In [54], the network configuration changes are identified using recursive
Bayesian approach by utilizing the outputs of state estimation function.
5) Input data correlation: Correlation usually exists among different input data. For instance, a degree of
correlation can often be assumed among power consumptions or generations of some particular nodes.
The impact of parameters correlation on the quality of the estimation is analysed in [51], and the influence
of input data correlation on DSSE is investigated in [29, 55].

4.2.4 Simulation results
DSSE is tested on a meshed 24-bus section of real UK distribution network [44], as given in Figure 4.2 (a).
The actual load profiles of the network shown in Figure 4.2 (b) are taken from a week in August 2012 in UK,
and applied in the study. The unbalance is generated by unevenly distributing the load demand among
three phases at the representative loads located at buses 15, 17, 19 and 21 respectively [44]. The load
distribution is guided by, i.e., it is made proportional to, the total load demand at the buses. In total, there
are 1008 operating points each representing the operation during a ten minute period.
3 6 50
Total load demand

14 20 21 9
45
15 2 10
24 4 12 22 40
16 13 11
5 8 23 1 35
18 7 19 ~
30
17 0 100 200 300
Operating points
(a) 24-bus distribution network (b) Actual load profile
Figure 4.2 Distribution network and actual load profile
DSSE is tested for 1008 operating points on the 24-bus distribution network with four monitors at 2,5,9 and
11 respectively. The mean and standard deviation of the errors as a percentage of the nominal voltages are
0.19% and 0.38% respectively. The simulation results at the peak time is presented in Figure 4.3, which
provides the actual and estimated voltage magnitudes obtained at phase A. It can be seen that the
estimated results are very closed to the actual voltage magnitude except for bus 15 which has a relatively
larger error tan other buses.
Actual
1.15 Estimated
1.1
Voltage magnitude (V)
1.05
0.95
0.9
0.85
0.8
0 5 10 15 20 25
Bus index
Figure 4.3 Simulation results on 24-bus test network

DSSE is also tested in a large-scale distribution network, 295-bus generic distribution network (GDN), as
shown in Figure 4.4 [56, 57]. It comprises 275 kV transmission in-feeds, 132 kV and 33 kV predominantly
meshed sub-transmission networks, and 11 kV predominantly radial distribution network. The network
consists of 276 lines including overhead lines and underground cables, 37 transformers with various
winding connections, 297 loads (including 10 unbalance loads) representing industrial, commercial and
domestic loads. The default errors of the measurement errors are given in Table 4.1.
300
400kV 293 292 299 298
275kV 294 295 297 296
258 259 257 256 254 255 233 234
132kV 279 280
278 277
271 276 275 253 274 270 272 262 263 264 273
N O
241 246 245 244 243 242 249 250 240 248 247
251 237
261 252
260
33kV
235 236
238 268 269 269 267 266 77
A B C D E H I J K L
289 229 228 231 230 226 227 232
50 54 223 22 78 87 57 58 60 77
221 52 20 79 56 62 59 225
51 76 53 21 19 80 55 61
74 15 81 2 1 4 65 216
75 14 18 88 82 3 5 64 215 211 217
12 16 17 291 89 83 179 72 6 66 63 210 212 218 171
13 290 84 25 73 70 7 193 213 219 172
23 91 85 27 68 71 8 188 194 214 220 173
24 26 92 86 29 67 69 9 10 166 189 197 195 174
28 93 94 30 11 165 167 190 191 198 196 175
45 46 95 102 32 161 169 168 192 200 199 176
44 48 47 96 103 104 31 33 224 160 162 170 201 177
222 49 97 98 105 35 34 151 157 163 180 202 178

11kV 43 101 99 106 107 36 152 149 155 158 164 181 203 204
42 100 108 150 147 153 156 159 182 183 205
41 109 148 146 154 185 184 206
40 112 110 145 186 187 207
39 113 111 143 209 208
38 114 115 144 141
37 116 142 139
117 121 131 140
118 122 132 129
119 120 124 123 133 130
125 134
126 135
Zone-1 127 Zone-2 137 136 Zone-3

F G 128 138
3.3kV 288 287 286 285
Figure 4.4 Single line diagram of 295-bus generic distribution network
Table 4.1 Measurement error variances (%)

measurement type (a) (b) (c)
Real measurements 0.2 0.2 0.2
Virtual measurements 0.01 0.01 0.01
Pseudo-measurements 7 7 7
To reflect the PQ performance accurately, the variation of load profiles and network parameters are taken
into account. In this study, annual hourly loading curves were extracted from 2010 survey of different types
of loads (including commercial, industrial and residential loads), and 8760 operating points are obtained.
Since there exist similar patterns of load demand variation among loads of the same types (e.g., industrial,
commercial and domestic loads) and similar variation trends of the outputs of certain DGs (i.e., PV) in terms
of day and season, similar operating condition re-occurs throughout the whole year. In the study, Cluster
Evaluation of Statistics Toolbox in Matlab is used to find the representative operating condition. Various
clustering approaches (K-means, fuzzy c-means, agglomerative clustering algorithm and Gaussian mixture
distribution algorithm) and clustering criteria were tested, and the approach yielding the best results during
evaluation is adopted here. The appropriateness of the obtained clusters is validated using the method of
Silhouette. It was found that the K-means with the clustering criterions of Calinski-Harabasz, which defines
the ratio between the overall between-cluster variance and the overall within-cluster variance, yields the
best results. Using this approach, the representative operating points which has the largest cluster is used

to test DSSE. The mean and standard deivation of the errors (in percentage) are 0.31% and 0.2%
respectively. The comparison between the actual and estimated voltage magnitudes is provided in Figure
4.5, which shows the accuracy of the estimated voltages compared with the actual voltages. To present the
estimated results visually, the heatmap of voltages obtained with 20 real measurements are plotted in
Figure 4.6. It can be seen that the critical area marked in red is exposed to relatively high voltage.
1,08
1,07
1,06
1,05
1,04
digsilent
1,03
1,02
SE (Sqrt of Covariance
1,01 Matrix)
1
0,99
0,98
0,97
0 100 200 300 400
Figure 4.5 Comparison between the actual and estimated voltages for 295-bus GDN
Figure 4.6 Heatmap indicating voltages of the network (20 real measurements)
The influence of the selection of real measurements on the estimation accuracy is tested by randomly
selecting 10 substations for real measurements. Ten runs are performed and the results are given in Table
4.2, in which the row of mean provides the mean of the errors between the actual and estimated
voltages among all 295 buses; the row of std and max provides the standard deviation and maximum of
errors among all buses. It can be seen that the variation due to the measurement locations is 0.1082%.

Table 4.2 Simulation results with 10 randomly selected substations for real measurements
error (%) 1 2 3 4 5 6 7 8 9 10
mean 0.439 0.699 0.4 0.45 0.5133 0.659 0.4054 0.606 0.599 0.480
std 0.365 0.525 0.4 0. 42 0. 5509 0.542 0.3603 0.493 0.455 0.365
max 1.969 1.991 2 1.95 2.1807 2.083 1.9288 1.971 2.014 1.950
The influence of the accuracy of the real measurement on the estimation accuracy is tested by the
following three cases:
1. All real measurements have fixed +3.5% deviation
2. All real measurements have fixed -3.5% deviation
3. Real measurements have the mix of +3.5% and -3.5% deviation
Among these three cases, 3.5% is given as this is the maximum possible error accumulated by transducers
error (3% based on IEC60044-2) and measurement error (shall not exceed 0.5% for class B performance).
The results of simulation (three cases) are given in Table 4.3. It can be seen that case 1 presents the
maximum possible error of estimation which reaches 2.56% of the nominal voltages.
Table 4.3 Simulation results based on maximum possible errors in real measurements
Error (%) 3.50% -3.50% mix 3.5%
mean 0.026 0.020 0.012
std 0.013 0.014 0.009
max 0.041 0.038 0.028
4.3 LOAD FLOW

The analysis of a power distribution network typically consists of a study of the network under normal
steady-state operating conditions. The power flow analysis, also called load flow analysis, of the
distribution network is similar to that of an interconnected transmission system [58]. The loading of a
distribution feeder is inherently unbalanced due to a large number of unequal single-phase loads and the
non symmetrical conductor spacing of three-phase underground and overhead line segments. However,
the conventional power flow programs used for transmission systems assume a perfectly balanced system
so that a single-phase equivalent can be used. As a result, these conventional power flow analysis tools do
not provide the actual behavior of the network. Furthermore, they do not show good convergence
properties for distribution systems due to their radial structure and their high R/X ratio. The power flow
analysis is the core tool for the operation and planning of the distribution networks. The unbalanced
loading of a distribution network results to power losses increase and protection malfunction. Thus, a three
phase power flow analysis is necessary to identify these issues and to obtain the actual behavior of the
system. This report is part of a project dealing with the three-phase power flow analysis of electric power
distribution networks. More specifically, this report presents the modeling of all the major components of a
distribution system and the development of a three-phase power flow tool. Afterwards, the developed
three-phase power flow method is validated using IEEE distribution test systems and it is applied to the
real-world Meltemi distribution network.
In Section 4.3.1 the modeling of the distribution lines are presented. Section 4.3.2 presents the modeling of
the transformers and the voltage regulators (VRs). Section 4.3.3 presents the needed data for the modeling

of the switches. Sections 4.3.4 and 4.3.5 present the needed input data for the modeling of the loads and
capacitor banks (CBs), respectively. Section 4.3.6 presents the three-phase power flow method based on
the backward/forward sweep method. Section 7 presents the results of the developed three-phase power
flow tool on the IEEE 13-bus distribution test system, the IEEE 123-bus distribution test system and the
Meltemi distribution network. The conclusions are drawn in Section 4.3.7.
The developed unbalanced three-phase power flow method of a distribution network determines the
following by phase and total (three-phase) [58]:
Voltage magnitudes and angles at all nodes of the distribution network
Power flow in each line section specified in kW and kvar, amps and degrees or amps and
power factor
Power loss in each line section
Total feeder input kW and kvar.
4.3.1 Distribution Network Lines

The determination of the series impedance matrix Z abc and the shunt admittance matrix y abc
for distribution lines is a critical step before the execution of the unbalanced three-phase power
flow analysis of the distribution network. These two matrices have the following form:
z aa z ab z ac

Z abc z ba z bb z bc (4.12)
km
z ca z cb z cc
y aa y ab y ac
S
y abc y ba y bb y bc (4.13)
km
y ca y cb y cc
The distribution lines can be overhead or underground. The computation of the matrices Z abc and y abc
is automatically done by the software tool using the data of Sections 4.3.1.1 and 0, which are mainly
related with the conductor type and the geometry of the distribution lines.
4.3.1.1 Overhead Distribution Lines

In order to model the series impedance and the shunt admittance of an overhead distribution line
the following data are needed:
1. ri : resistance of conductor i (/km)
2. Dij : distance between conductor i and j (m)
3. Di g : distance between conductor i and ground (m)
4. GMRi : Geometric Mean Radius of conductor i (m)

5. RDi : radius of conductor i (m)
6. : resistivity of earth (m)
7. f : system frequency (Hz)
8. l : lentgth of the distribution line (km)
9. The phase sequence.
The resistance, the diameter and the GMR of a conductor are associated with its type and it is a common
industrial practice these data to be provided in a datasheet form. Such a datasheet example for various
conductors is presented in Appendix A. The spacing distances between the phase conductors and the
neutral conductor depends on the geometry of the utility pole. Since in a real distribution network, a few
types of distribution lines are used, the different types of distribution lines are defined in the library of the
unbalanced three-phase power flow software tool by giving their relevant technical and geometrical
characteristics. Such an approach is done only once and is very convenient for multiple power flow studies
of various parts of the distribution network of a certain electric power distribution utility.
The modified Carsons equations (4.14) and (4.15) are used to calculate the elements of the primitive series
impedance matrix Z [58] The dimension primitive impedance matrix Z is N
prim prim cond Ncond, where
Ncond is the number of the conductors.
0.098696
zii ri f
100
(4.14)
1
j
0.125664
f ln
1
7.6786 ln / km
100 3.28 GMRi 2 f
0.098696
zij f
100
(4.15)

j
0.125664
f ln
1 7.6786 1 ln / km
100 3.28 Dij

2 f
In case, there is no neutral conductor (delta connection), i.e., the distribution line is consisted of 3
conductors (one per phase), Zabc Z prim. If there is a neutral conductor, i.e., the distribution line is
consisted from four conductors, the Kron reduction technique is applied to calculate Z abc , as follows:
Z abc zij zin znn 1 z jn / km (4.16)
The self and mutual potential coefficients of the permittivity are defined by (4.17) and (4.18) and the
primitive coefficient matrix P prim can be constructed.
S
Pii 17.98746 ln ii km / F (4.17)
RDi
Sij
Pij 17.98746 ln km / F (4.18)
Dij

Paa Pab Pac
P
prim Pba Pbb Pbc
km
F
(4.19)
Pca Pcb Pcc
where S ii : distance from conductor i to its image i (m)
S ij : distance from conductor i to the image of conductor j (m)
In case, there is no neutral conductor (delta connection), i.e., the distribution line is consisted of 3
conductors (one per phase), Pabc Pprim . If there is a neutral conductor, i.e., the distribution line is
consisted from four conductors, the Kron reduction technique is applied to calculate Pabc , as follows:
Pabc Pij Pin Pnn km / F

1
P jn (4.20)
The capacitance matrix is defined by (4.11) and the shunt admittance matrix is defined by (4.22).
Cabc Pabc 1 (4.21)
y abc j 2 f Cabc S / km (4.22)
4.3.1.2 Underground distribution lines

The two most popular types of underground cables are:
i. The concentric neutral cable and
ii. The tape-shielded cable.
The needed data for the modeling of both types of underground cables are presented in Sections 4.3.1.2.1
and 4.3.1.2.2.
4.3.1.2.1 Concentric neutral cable
In order to model the concentric neutral cable the following data are needed:
1. d c : phase conductor diameter (m)
2. d od : nominal diameter over the concentric neutrals of the cable (m)
3. d s : diameter of the concentric neutral strand (m)
4. Dnm : center to center distance between phase conductors (km)
5. GMRc : Geometric Mean Radius of the phase conductor (m)
6. GMRs : Geometric Mean Radius of a neutral strand (m)
7. rc : resistance of a phase conductor (/km)
8. rs : resistance of a solid neutral strand (/km)

9. k : number of concentric neutral strands
10. l : length of the conductor (km)
It is a common industrial practice the data for a phase conductor and neutral strands to be provided in
datasheet forms. Such a datasheet example for various conductors and concentric neutral cables is
presented in Appendix A. The data for the different types of the concentric neutral cables are defined in the
library of the unbalanced three-phase power flow software tool by giving their technical and geometrical
characteristics.
The series admittance matrix is calculated by the modified Carsons equations (4.14), (4.15) and the Kron
reduction technique (4.16) [58]. The shunt admittance matrix is given by (4.22). The equivalent geometric
mean radius of the concentric neutral ( GMRcn ) is calculated by (4.23) and (4.24). The equivalent resistance
of the concentric neutral ( rcn ) is given by (4.25).
GMRcn k GMRs k R k 1 (4.23)
d od d s
R (4.24)
24
rs
rcn ( / km) (4.25)
k
4.3.1.2.2 Tape-Shielded cable

In order to model the tape-shielded cable the following data are needed:
1. d c : phase conductor diameter (m)
2. d od : outside diameter over the jacket (m)
3. d s : outside diameter of the tape shield (m)
4. Dnm : center to center distance between phase conductors (km)
5. GMRc : Geometric Mean Radius of the phase conductor (m)
6. rc : resistance of a phase conductor (/m)
7. T : thickness of copper shield tape shield (mm)

8. l : length of the conductor (km)
It is a common industrial practice the data for a phase conductor and neutral strands to be provided in
datasheet forms. Such a datasheet example for various conductors and tape-shielded cables is presented in
Appendix A. The data for the different types of the tape-shielded cables are defined in the library of the
unbalanced three-phase power flow software tool by giving their technical and geometrical characteristics.
The series admittance matrix is calculated by the modified Carsons equations (4.14), (4.15) and the Kron
reduction technique (4.16) [58]. The shunt admittance matrix is given by (4.22). The GMR of the tape shield
( GMRshield ) is given by (4.26) and the resistance of the tape shield is calculated by (4.27).

T
ds
GMRshield 1000 (4.26)
24

rshield 1.2304 1012 ( / km) (4.27)
ds T
The resistivity () must be expressed in /m at 50o C.

4.3.1.3 Distribution Line Model
The model of an overhead or underground distribution line is shown in Figure 4.7. It should be noted that
the series impedance and shunt admittance matrices are 33 (Sections 4.3.1.1 and 0). If any phase of the
line section does not exist, the corresponding row and column in the series impedance and shunt
admittance matrices contain all zero entries.
Ia,i i j I
a, j
Va,i Va, j
Ib,i Ib, j
Zabc Vb, j
Vb ,i Ic,i Ic, j
Vc,i Vc, j
1 1
yabc yabc
2 2
Figure 4.7 Three-phase distribution line model
Applying the Kirchhoffs laws, the voltages and currents at bus i in terms of the voltages and currents of bus
j are calculated by (4.28)(4.31), as follows:
1
[Vabc ] j [Vabc ]i Z abc yabc [Vabc ]i [ I abc ]i (4.28)
2
yabc [Vabc ]i [Vabc ] j [ I abc ]i
1
[ I abc ] j (4.29)
2
1
[Vabc ]i [Vabc ] j Z abc yabc [Vabc ] j [ I abc ] j (4.30)
2
1
[ I abc ]i yabc ([Vabc ]i [Vabc ] j ) [ I abc ] j (4.31)
2
4.3.2 Three-Phase Transformers and Voltage Regulators

The data needed for the modeling of the three-phase transformers and the voltage regulators (VRs) are
presented in this section.
4.3.2.1 Three-Phase Transformers

In order to model a three-phase transformer the following data are needed:
1. The position in the distribution network

2. The general winding configuration
3. The nominal power capacity (kVA)
4. The nominal voltage of both primary and secondary windings (kV)
5. The series impedance ( Z t ) (%).
The general equations that calculate the voltages and currents in the primary and secondary side are given
by (4.32)(4.34). The capital letters A, B and C denote the primary side and the lower case letters a, b and c
denote the secondary side.
[VLN ABC ] [at ] [VLN abc ] [bt ] [ I abc ] (4.32)
[ I ABC ] [ct ] [VLN abc ] [d t ] [ I abc ] (4.33)
[VLN ABC ] [ At ] [VLN abc ] [ Bt ] [ I abc ] (4.34)
In (4.32)(4.34), [VLN ABC ] and [VLN abc ] are the line-to-neutral voltages for ungrounded wye, the
equivalent line-to-neutral voltages for delta connections or the line-to-ground voltages for grounded wye
connections.
For every different winding configuration the matrices in (4.32)(4.34) should be defined. It is assumed that
in all winding configurations the current running through the magnetizing branch of the transformer is
equal to zero, hence all elements of [ct ] are equal to zero. The winding configurations that were
modelled are the following:
DeltaGrounded wye step-down transformer:
Vnpr
nt , pol
(4.35)
Vnsec
, ph
0 2 1
[at ] 1 0 2
nt
(4.36)
3
2 1 0
0 2 Z t ,b Z t ,c
nt
[bt ] Z t ,a 0 2 Z t ,c (4.37)
3
2 Z t ,a Z t ,b 0
1 1 0
[d t ] 0 1 1
1
(4.38)
nt
1 0 1
1 0 1
[ At ] 1 1 0
1
(4.39)
nt
0 1 1

Z t ,a 0 0

[ Bt ] 0 Z t ,b 0 (4.40)
0 0 Z t ,c
pr sec
where, Vn , pol is the nominal polar voltage of the primary winding; Vn , ph is the nominal phase voltage of the
secondary winding; Z t ,a , Z t ,b and Z t ,c are the per phase impedances expressed in the secondary side.
Grounded wyeGrounded wye step-down transformer:
Vnpr
nt , ph
(4.41)
Vnsec
, ph
1 0 0
[at ] nt 0 1 0 (4.42)
0 0 1
Z t ,a 0 0

[bt ] nt 0 Z t ,b 0 (4.43)
0 0 Z t ,c
1 0 0
[d t ] 0 1 0
1
(4.44)
nt
0 0 1
1 0 0
[ At ] 0 1 0
1
(4.45)
nt
0 0 1
Z t ,a 0 0

[ Bt ] 0 Z t ,b 0 (4.46)
0 0 Z t ,c
pr
where, Vn, ph is the nominal phase voltage of the primary winding.
DeltaDelta transformer:
Vnpr
nt , pol
(4.47)
Vnsec
, pol
2 1 1
[at ] 1 2 1
nt
(4.48)
3
1 1 2

Z t ,ca Z t ,bc 0
1
[G1] Z t ,ca Z t ,ab Z t ,bc 0 (4.49)
Z t ,ab Z t ,bc Z t ,ca
Z t ,ab Z t ,bc Z t ,bc 0
2 1 0 Z t ,ab 0 0

[bt ] 0 2 1 0
nt
Z t ,bc 0 [G1] (4.50)
3
1 0 2 0 0 Z t ,ca
1 0 0
[d t ] 0 1 0
1
(4.51)
nt
0 0 1
2 1 1
1
[ At ] 1 2 1 (4.52)
3 nt
1 2 1
2 1 0 Z t ,ab 0 0

[ Bt ] 0 2 1 0
1
Z t ,bc 0 [G1] (4.53)
3
1 0 2 0 0 Z t ,ca
sec
where, Vn , pol is the nominal polar voltage of the secondary winding.
Open wyeOpen delta transformer:
Vnpr, ph
nt (4.54)
Vnsec
, pol
nt nt 0
[at ] 0 nt nt (4.55)
0 0 0
nt Z t ,ab 0 0

[bt ] 0 0 nt Z t ,bc (4.56)
0 0 0
1 0 0
[d t ] 0 0 1
1
(4.57)
nt
0 0 0
2 1 0
1
[ At ] 1 1 0 (4.58)
3 nt
1 2 0

2 Z t ,ab 0 Z t ,bc
1
[ Bt ] Z t ,ab 0 Z t ,bc (4.59)
3
Z t ,ab 0 2 Z t ,bc
4.3.2.2 Voltage Regulators

In order to model a voltage regulator the following data are needed:
1. The position in the distribution network
2. The general winding configuration. The following configurations are used:
Wye-connected regulators
Delta-connected regulators.
3. The phase(s) in which the VR is installed
4. BWreg : The bandwidth (V)
5. Vn,reg : The nominal voltage of the regulator (V)
6. The voltage transformer ratio (Npt)
7. The primary current transformer rating (CTp) and the secondary current transformer rating (CTs)
(A)
8. The R and X settings of the compensator (V)
9. Vrelay : The voltage level (V)
The VR consists of an autotransformer and a load tap changing mechanism. The position of the tap is
determined by a control circuit, which is called line drop compensator (LDC). Figure 4.8 shows a simplified
sketch of the LDC circuit.
The general equations that model the operation of the voltage regulator are given by (4.60)(4.62).
[VLN ABC ] [areg ] [VLN abc ] [breg ] [ I abc ] (4.60)
[ I ABC ] [creg ] [VLN abc ] [d reg ] [ I abc ] (4.61)
[VLN ABC ] [ Areg ] [VLN abc ] [ Breg ] [ I abc ] (4.62)

I line CTp : CTs Rline j X line
R j X Load
Center
1:1
Vdrop

Voltage
Npt :1 Vreg Relay

Figure 4.8 LDC circuit
where
aR _ ph 1 0.00625 Tap ph (4.63)
Vn,reg 0.5 BWreg Vrelay

Tap ph (4.64)
0.00625 Vn,reg
a R _ a 0 0

[areg ] 0 aR _ b 0 (4.65)
0 0 aR _ c

0 0 0
[breg ] 0 0 0 (4.66)
0 0 0
0 0 0
[creg ] 0 0 0 (4.67)
0 0 0
a R1_ a 0 0
1
[d reg ] 0 a R_b 0 (4.68)
0 0 a R1_ c

[ Areg ] [areg ]1 (4.69)
[ Breg ] [areg ]1 [breg ] (4.70)
4.3.3 Switches
The modeling of the switches does not depend on their technical characteristics and their material. For
power flow calculations, the only information needed is the status of the switch:
1. Open The switch is modeled as an open-circuit

2. Closed The switch is modeled as short-circuit.
Considering that between buses i and j of Figure 4.8 there is a switch, the general equations that model the
operation of the voltage regulator are given by (4.71) and (4.72).
[Vabc ]i [Vabc ] j (4.71)
[ I abc ]i [ I abc ] j (4.72)
4.3.4 Loads
In order to properly model the loads in the three-phase power flow analysis of the distribution network,
four features should be provided. The required features of the loads are:
1. The way of their connection to the network. The loads can be:
Wye-connected, as shown in Figure 4.9 (a), or
Delta-connected, as shown in Figure 4.9 (b).
a Va a Va
Vb Vb
b b
c Vc c Vc
Ia Ib Ic Ia Ib Ic
Zb
Zab Zbc
Za Zc
Zac
a) b)
Figure 4.9 Load connection (a) Wye-connected load and (b) Delta-connected load
2. The model type. The loads can be:
Constant Power (constant PQ). In this case, the active (kW) and reactive power (kVar) of the
load per phase should be given.
Constant Current. In this case, the active (kW) and reactive power (kVar) of the load per
phase should be given in the nominal voltage of the system.
Constant Impedance. In this case, the active (kW) and reactive power (kVar) the load per
phase should be given in the nominal voltage of the system.
3. The position in the network. The loads can be:

Concentrated Load. In this case, the node of the network in which the load is connected
should be provided.
Distributed Load. In this case, the total length of the distribution line segment and the total
load that is served by this distribution line should be given. The distributed load is divided
into two concentrated fictitious loads. The first fictitious load is equal to the 2/3 of the total
distributed load and it is connected to a fictitious bus placed at the one-fourth way from the
source end. The second fictitious load is equal to the 1/3 of the total load and it is placed at
the end of the line.
4. The number of the phases. The loads can be:
Three Phase Load.
Two Phase Load. In this case, the load of the missing phase is set equal to zero.
Single Phase Load. In this case, the load of the two missing phases is set equal to zero.
In order to perform the three-phase power flow analysis, the injected currents from the load demand have
to be calculated. Thus, for a given connection and type load the injected current ( I L _ inj ) is calculated, as
follows:
Constant PQWye connected:
~ S~n, a V~a
*
I a _ inj
~ ~ ~
I L _ inj I b _ inj S n,b . / Vb (4.73)
I~c _ inj ~ ~
Sn,c
Vc
~ ~ ~
where I a _ inj , S n ,a and Va are the injected current, the nominal apparent power and the phase
voltage of phase a, respectively.
Constant PQDelta connected:
~ ~ ~ *
I ab _ inj 1 0 1 Sn, ab Vab
~ ~ ~
I L _ inj I bc _ inj 1 1 0 Sn,bc . / Vbc (4.74)
I~ca _ inj ~ ~
0 1 1 Sn, ca Vca

~ ~
where S n, ab and Vab are the nominal apparent power and the polar voltage between phase and
phase b, respectively.
Constant CurrentWye connected:
~
I a _ inj
~

S n, a V~a exp j arg( I~a _ inj )
I L _ inj
~ ~ ~
~
I b _ inj S n,b . / Vb exp j arg( I b _ inj ) (4.75)
I~c _ inj

S~n, c V~c exp j arg( I~c _ inj )


Constant CurrentDelta connected:
~ ~ ~ ~

I ab _ inj 1 0 1 Sn, ab Vab exp j arg( I ab _ inj )
I L _ inj
~ ~ ~ ~

I bc _ inj 1 1 0 Sn,bc . / Vbc exp j arg( I bc _ inj ) (4.76)

I~ca _ inj 0 1 1 S~n, ca V~ca exp j arg( I~ca _ inj )

Constant ImpedanceWye connected:
~ ~ * ~
Ya S n,a Va2
~ ~ ~
YL Yb S n ,b . / Vb2 (4.77)
Y~c S~n ,c V~c2

~ ~ ~
I a _ inj Ya Va ,t
~ ~ ~
I L _ inj I b _ inj Yb Vb ,t (4.78)
I~c _ inj Y~c V~c ,t

~
where Va ,t is the value of the voltage of phase a at iteration t of the power flow algorithm.
Constant PQDelta connected:
~ ~ * ~
Yab S n,ab Vab2
~ ~ ~
YL Ybc S n ,bc . / Vbc2 (4.79)
Y~ca S~n,ca V~ca2

~ ~ ~
I ab _ inj 1 0 1 Yab 1 1 0 Va ,t
~ ~ ~
I L _ inj I bc _ inj 1 1 0 Ybc 0 1 1 Vb,t (4.80)
I~ca _ inj 0 1 1 Y~ca 1 0 1 V~c ,t

4.3.5 Shunt Capacitor Banks
Shunt CBs are commonly used in distribution systems for voltage regulation and reactive power support.
The features that are needed in order to model the CBs in the power flow analysis are:
1. The node that the CB is connected
2. The way of their connection to the network. The CBs can be:
Wye-connected or
Delta-connected.
3. The reactive power (kVar) per phase of the CBs in the nominal voltage. Similar to the loads the
reactive power of the missing phases is set equal to zero for single phase and two phase CBs.

The CBs are modeled as constant impedance loads for the three-phase power flow analysis, as shown in
Section 4.3.4. The nominal active power of the CB is set equal to zero and the reactive power of the CB is
negative.
4.3.6 Backward/Forward Sweep Power Flow Method for Radial Distributions Networks
Several distribution power flow methods have been proposed in the literature. These methods are
categorized into three main categories:
Network reduction methods [59]
Backward/forward sweep methods [60-63]
Fast decoupled methods [64-66]
The developed three-phase power flow analysis software tool was based on the backward/forward sweep
method [67]. The radial structure of the distribution networks under study allows the modeling of the
network as a tree. The indexing of the buses is performed using the reverse breadth first (RBF) and breadth
first (BF) sorting methods [67]. The backward/forward sweep method consists of three basic steps and its
general architecture is presented in Figure 4.10. The method is an iterative algorithm and it is terminated
until a convergence criterion is achieved. Figure 4.11 shows a simple distribution network example, which
will be helpful to describe the backward and forward sweep steps. In Figure 4.11, bus zero is the source
bus and bus N is the end bus.
Backward Sweep Forward Sweep

Initialize all bus Calculate currents or Is convergence No Calculate voltage drops
Voltages power flows and achieved? and update currents/
update voltages power flows
Yes
End
Figure 4.10 Flowchart of the backward/forward sweep method
~ ~
V0 VN
...
0 1 2 i N-2 N-1 N
Figure 4.11 Distribution network example
4.3.6.1 Bus Indexing

A radial distribution network is consisted of a main feeder with laterals. These laterals may also have sub-
laterals, which may also have their sub-laterals etc. Thus, bus indexing is required to traverse the
distribution network in a specific order in the steps of the backward/forward sweep method [67]. A triple
bus indexing (L, M, N) is assigned to each bus. The indices are defined as follows:

Index L denotes the level of the lateral that bus i is placed. For example, the main feeder would be
level 1, its sub-laterals would be level 2, their sub-laterals would be level 3, etc.
Index M denotes the laterals within the same level L according to the order seen during a breadth-
first traversal. For example, the main feeder will be level 1 ( L 1 ) and its laterals will be level 2 (
L 2 ). The buses in the first level 2 lateral will be indexed with M 1, the buses in the second
level 2 lateral will be indexed with M 2 , etc.
Index N denotes the buses that have the same L and M indices, which means that they belong to
the same lateral. The source bus always has N 0 , while the first buses of the rest laterals always
have N 1 .
Figure 4.11 presents an example of the indexing scheme on a 29-bus distribution network. The RBF order is
determined by sorting the indices in descending order, first by index L, then by index M and finally by index
N. The RBF order is typically used for the backward sweep step. If the distribution network in Figure 4.12 is
traversed using the RBF order, the traversal order starts from the lateral that has as an end bus 26, then the
laterals that has as an end bus 13, bus 27, bus 22, bus 15, bus 6 and bus 29. If the indices are sorted in
ascending order, the BF order is determined, which is used for the forward sweep step.
(2,4,3) 27
(2,1,4) 6
(3,2,1)
(2,1,3) 5 (2,4,2) 25
(2,1,2) 4
(2,4,1) 24
(2,1,1) 3 26
(1,1,0) (1,1,1) (1,1,2) (1,1,3) (1,1,4) (1,1,5) (1,1,6) (1,1,7)
1 2 16 17 18 23 28 29
(2,3,1) 19
(2,2,1) 7
(2,3,2) 20
(2,2,2) 8 (2,3,3) 21
(2,3,4) 22
(2,2,3) 9
(3,1,1) (3,1,2) (3,1,3)
(2,2,4) 10
(2,2,5) 14
11 12 13
(2,2,6) 15
Figure 4.12 29-bus distribution network with indexed buses
4.3.6.2 Backward sweep step

The method starts by initializing all the bus voltages. The selected initial values of the bus voltages are
usually equal to their nominal values. Afterwards, the distribution network is traversed processing the
buses and the laterals in the RBF order (see Section 4.3.6.1). For example, in Figure 4.11, the distribution
network will be traversed from the end bus, i.e, bus N, towards the first bus, i.e., bus zero.

In every bus, the injected currents by the load or the shunt capacitor are calculated based on the bus
voltage of the current iteration, as shown in Sections 4.3.4 and 4.3.5. Then, the Kirchhoff's current law (KCL)
is applied in order to determine the incoming injected current ( I inj,i ) in every bus. Figure 4.12 illustrates
the KCL application at bus i and KCL is solved for I inj,i , as follows:
I inj,i I L _ inj,i I Cap _ inj,i I out, j I out,i (4.81)
j Ai
where, I L _ inj,i and I Cap_ inj,i are the injected currents by the connected load and shunt capacitor at bus i,
respectively; I out, j is the injected current from a sub-lateral branching off from bus i; Ai is the set of the
sub-laterals branching off from bus i; I out, j is the outgoing current from bus i to the next bus on the same
lateral.
V i 1 Vi I
j Ai
out , j
I out,i 1 Distribution line, I inj ,i I out ,i

Transformer,
Voltage regulator,
or Switch
i 1 i
ICap _ inj ,i I L _ inj ,i
Load
Figure 4.13 Distribution network segment example
After computing the incoming injected current ( I inj,i ) at bus i, the voltage of bus i 1 ( Vi 1 ) and the
outgoing current bus ( I out,i1 ) from bus i 1 are calculated using the appropriate equations developed in
Sections 4.3.14.3.5. More specifically:
In case a distribution line is placed between buses i 1 and i, Vi 1 and I out,i1 are calculated
according to (4.28) and (4.29), respectively.
In case transformer is placed between buses i 1 and i, Vi 1 and I out,i1 are calculated according
to (4.32) and (4.33), respectively.
In case a voltage regulator is placed between buses i 1 and i, Vi 1 and I out,i1 are calculated
according to (4.60) and (4.61), respectively.

In case a switch is placed between buses i 1 and i, Vi 1 and I out,i1 are calculated according to
(4.71) and (4.72), respectively.
Considering the aforementioned equations, all bus voltages and line currents are determined in the RBF
order and moving the process towards the source bus. Thus, the last examined lateral, which is in fact the
main feeder, is processed and the bus voltage of the source bus i is determined completing the backward
sweep step. However, the voltage of the source bus is given and the error between the calculated value of
the backward sweep step and the given value of source bus voltage is computed. If the error is smaller than
a given threshold the power flow analysis is terminated. Otherwise, the forward sweep step is executed.
The selection of the error threshold value is crucial because a relative high value may lead to inaccurate
results, while a relative small value may lead to infinite iterations. A reasonable threshold value is
considered to be equal to 10e-5. Figure 4.14 presents the flowchart of the backward sweep step.
Initialize all bus
voltages
Set
i=1
Find the ith bus in

the RBF order
Application of
KCL of bus i
Is there a Is there a
No No Is there a VR No Is there a switch
distribution line transformer
between buses i-1 between buses i-1
between buses i-1 between buses i-1
and i ? and i ?
and i ? and i ?
Yes Yes Yes Yes
Calculate the Calculate the Calculate the Calculate the

bus voltage and bus voltage and bus voltage and bus voltage and
line current by line current by line current by line current by
(17) and (18) (21) and (22) (49) and (50) (60) and (61)
Examine the
No Yes Calculate the No
next bus in the Is bus i-1 the Is convergence Forward sweep
source bus
RBF order: source bus? achieved? step
voltage error
i=i-1
Yes
End
Figure 4.14 Flowchart of the backward sweep step
4.3.6.3 Forward sweep step

If the error has not a value lower than the defined threshold, then the execution of the forward sweep step

is necessary. In this step, the currents that were calculated in the backward sweep step are taken as
granted and the bus voltages of the network are calculated considering that the source bus voltage is equal
to its given value. The distribution network is, now, traversed processing the buses and the laterals in the
BF order (see Section 4.3.6.1). For example, in Figure 4.11, the distribution network will be traversed from
the first bus, i.e., bus zero, towards the end bus, i.e, bus N.
Starting from the source bus, the equations that are used to calculate the voltage of bus i based on the
voltage and the outgoing current from bus i 1 are as follows:
In case a distribution line is placed between buses i 1 and i, Vi 1 is calculated according to

(4.30).
In case transformer is placed between buses i 1 and i, Vi 1 is calculated according to (4.34).
In case a voltage regulator is placed between buses i 1 and i, Vi 1 is calculated according to

(4.62).
In case a switch is placed between buses i 1 and i, Vi 1 is calculated according to (4.71).
In this way, the bus voltages are updated and the algorithm moves to the backward sweep step. Figure
4.15 presents the flowchart of the forward sweep step.
Backward
sweep step
Set
i=1
Find the ith bus in

the BF order
Is there a Is there a
No No Is there a VR No Is there a switch
distribution line transformer
between buses i between buses i
between buses i between buses i
and i+1 ? and i+1 ?
and i+1 ? and i+1 ?
Yes Yes Yes Yes
Calculate the Calculate the Calculate the Calculate the

voltage of bus voltage of bus voltage of bus voltage of bus
i+1 by (19) i+1 by (23) i+1 by (51) i+1 by (60)
Examine the
next bus in the No Is bus i+1 the end
BF order: bus?
i = i+1
Yes
Figure 4.15 Flowchart of the forward sweep step

4.3.7 Conclusion
The objective of this report is to develop a three-phase power flow analysis tool considering the
comprehensive formulation of all distribution network elements. More specifically, the detailed and
extensive modeling of distribution lines, transformers, voltage regulators, switches, loads and capacitor
banks is performed. Thus, the developed software tool provides the actual behavior of the distribution
network, which is necessary for the distribution network automation of real world power systems. The
power flow method is based on the backward/forward sweep method. The developed three-phase power
flow tool was validated using the IEEE 13-bus and 123-bus distribution test systems and the obtained
results were almost identical to [68]. Furthermore, it was applied to the real-world Meltemi distribution
network and its 24h steady-state operating conditions (voltage and current phasors) were obtained.
Detailed analysis of the results is included in Appendix A.
4.4 LOAD FORECASTING

Load forecasting is a technique used by power or energy-providing companies to predict the power/energy
needed to meet the demand and supply equilibrium. The accuracy of forecasting is of great significance for
the operational and managerial loading of a utility company. It also provides a great saving potential for
power system planning and investment activities. In [69, 70], a variety of load forecasting approaches are
introduced in detail, including multiple regression and stochastic time-series, fuzzy logics and neural
network etc. This section introduces techniques used for load forecasting in the study. Load forecasting
based on artificial intelligence (AI) techniques received significant attention in the past and it is rapidly
developing because of its high accuracy. Some of the AI based methodologies for load forecasting have
already been adopted and widely used by the industry. From the literature review, it was found that on the
one hand, among the nine most frequently used approaches in the past, artificial neural network (ANN) is
the most efficient one with the highest accuracy.
4.4.1 Artificial Neural Network

An artificial neural network (ANN) is a family of models inspired by biological neural networks (the
central nervous systems of animals, in particular the brain) which are used to estimate or approximate
functions that can depend on a large number of inputs and are generally unknown. ANN is an
interconnected assembly of simple processing elements, units or nodes. The processing ability of the
network is stored in the inter-unit connection weights, obtained by a process of adaptation to a set of
training patterns. It is inspired by biological nervous system and now widely applied to quantities of
areas such as prediction, curve fitting and clustering etc. The ANN can be used to learn the relationship
among past, current and future temperatures and loads.
Data Collection is essential for both training neural networks and implement forecasting. The inputs for the
total demand forecaster include: (i) P or Q demand of today in MW or MVAR; (ii) weather parameters
(including temperature, humidity and wind speed respectively) of today and tomorrow; (iii) day type
(i.e. working days, holidays) of today and tomorrow; (iv) the time tag which represents the period of
time in a day. Demand data used in this study are collected based on certain resolution requirements. The
general time resolutions is 30 minutes, i.e, 48 samples per day from real local distribution network. Real
time weather data, with the same 30 min resolution, including temperature (C), humidity (p.u), wind
speed (km/h) etc. are obtained from Weather History Website [71]. The input of the ANN contains 10 rows.

Apart from the 9 rows given in (4.84), an extra row, representing the time tag in one day, is also required
for both inputs of the nueral networks. For instance, if there are 48 samples a day, then the time tags
would be from 1 to 48 in sequel starting from midnight.
PN,0:00 PN,0:30 PN,23:00 PN,23:30

T TN,0:30 TN,23:00 TN,23:30
N,0:00
H N,0:00 H N,0:30 H N,23:00 H N,23:30

WS N,0:00 WS N,0:30 WS N,23:00 WS N,23:30
PTRN BLF TN+1,0:00 TN+1,0:30 TN+1,23:00 TN+1,23:30 (4.82)

H N+1,0:00 H N+1,0:30 H N+1,23:00 H N+1,23:30
WS WS N+1,0:30 WS N+1,23:00 WS N+1,23:30
N+1,0:00
DTN,0:00 DTN,0:30 DTN,23:00 DTN,23:30

DTN+1,0:00 DTN+1,0:30 DTN+1,23:00 DTN+1,23:30
1) ANN Structure: Two-layer ANNs are adopted in this analysis as they can potentially represent almost
all input-output relationships with a finite number of discontinuities as long as an appropriate number
of neurons is assigned to the hidden layer. Figure 4.16 illustrates the structure of a two-layer feed-
forward ANN including a hidden layer and an output layer. Either layer consists of an input vector p, a
weight matrix W, a bias vector b, a sum operator, a transfer function (TF) f and an output vector a. The
input elements are weightied by the matrix weighs, while the bias vector biases the weighed inputs. The
sum operator gathers the weighed inputs and the biases to produce an intermediate variable for the TF,
and the TF generates the final output of the layer. The hidden layer and output layer are connected in
serquence, and the output of the hidden layer is the input of the output layer. The full description of this
type of ANN and its parameter setting rules can be found in [70].
Hidden Layer Output Layer

Input Output
p W + f W + f a
b b
Figure 4.16 Structure of an ANN
The relationship of the input and the output in either layer can be represented by
a f W T p b (4.83)

2) Setting of Main Parameters: The main parameters of an ANN to be configured include the type of
ANN, the hidden layer size, transfer functions and the training algorithm. ANN with various types can be
selected as the tool for load forecasting. Either the feed-forward ANN or the cascade-forward ANN can
be selected as the tool for load forecasting. In MATLAB ANN Toolbox, FFANN and CFANN are used and
called by matlab commands newff and newcf respectively. In [72], a novel approach is proposed to optimize
the number of neurons in the hidden layer to avoid over-fitting, and the mathematical approximation for
N/d>>30 is shown as
N
n (4.84)
d ln N
where n is the number of neurons in the hidden layer, N is the number of the training sets, and d is the
input dimension. If N/d is smaller or close to 30, optimal n most frequently occurs on its maximum. While,
in [73] and [74], signal-to-noise-ratio figure (SNRF) and genetic algorithm is proposed to optimized the size
of hidden layers respectively, which prove that the hidden layer size is always finally determined by trial
and error with different number of neurons around the estimation from (4.2). But in most cases, the
nearest integral to the estimation of (4.2) works well.
The transfer function (TF) can be any differentiable function. The most commonly used TFs are log-sigmoid
(logsig), tan-sigmoid (tansig) and linear transfer function (purelin). The output ranges of the three TFs are
respectively [0,1], [-1,1] and [- ,+ ] [75].
Backpropagation (BP) [76] was originally introduced in the 1970s, but its importance wasn't fully
appreciated until 1982. Backpropagation works far faster than earlier approaches to learning, making it
possible to use neural nets to solve problems which had previously been insoluble. Today, the
backpropagation algorithm is the workhorse of learning in neural networks and it is widely used in ANN
training. It is composed of different variations, such as most commonly used gradient descent
backpropagation (GDBP), Levenberg-Marquardt backpropagation (LMBP), Bayesian Regulation
backpropagation (BRBP). In MATLAB ANN Toolbox, they are represented by traingd, trainlm and trainbr
respectively. GDBP is also called the basic BP, in which the weights are moved in the direction of the
negative gradient. LMBP is always suggested as the 1st selection in the ANN training due to its high speed
and high performance accuracy [75]. However, fast speed of LMBP induces relatively large uncertainties. As
a result, BRBP is employed sometimes to increase the robustness of the ANN. BRBP updates the weight and
bias based on Levenberg-Marquardt optimization. It minimizes a combination of squared errors and
weights and determines the correct combinations in order to produce a network that could generalize
better [75].
3) Framework: Figure 4.17 shows the framework for total load forecaster. It consists of basic load
forecaster (BLF), change load forecaster (CLF) and the adjuster. BLF is trained to forecast the regular load of
the next day, and CLF is trained to forecast the change of the load (from working day to holiday and vice-
versa) at the same time on following day. The adjuster, which can be either a trained ANN or a least square
algorithm box, is installed after BLF and CLF to take the advantage of the both. And the output is the day-
ahead forecasted demand in MW or MVAR.

Total Load Forecaster
Base Load
1. Demand of Day1 Forecaster
2. Forecasted Weather of Day1 Forecasted
and Day2 Adjuster Demand of
3. Day Type of Day 1 and Day 2
Change Load
Day2
Forecaster
Figure 4.17 Total load forecasting framework
4) Training: Figure 4.3 presents the training process. Before training, the weights are initialised to small
random values. The training process needs a set of examples of appropriately selected network
behaviours, inputs and targets. The differences between the created outputs and the target values are
recorded and propagated backwards through the network, and the weights and biases of the network
are adjusted to minimise the difference. This process is repeated until either the difference is within a
predefined range or the maximum epoch is reached. Bayesian Regulation Backpropagation (BRBP) is
selected as the training algorithm for its advantage in stable performance. BRBP is widely used in ANN
training. The basic backpropagation training algorithm, which is also called gradient descent
backpropagation (GDBP), adjusts the network weights and biases in the direction of the negative of
gradient, in which the performance function decreases most rapidly. The training process is
implemented in MATLAB.
Inputs of the Training Process
Targets of the Training
1. Demand of Day1 (1 row) Process
2. Forecasted Weather of Day1 and
Day2 (6 rows)
3. Day Type of Day 1 and Day2 (2 Demand of Day2 (1 row)
rows)
Training
Trained ANN
Figure 4.18 Total Load Forecasting Training Process
5) Validation: After the training, the trained neural networks are tested with another group of data.
forecasted results (i.e., the forecasted demand in MW or MVAR) is based on the new data are compared
against the actual measured demand. Mean absolute percentage error (MAPE) is adopted for error
analysis.
4.4.2 A Committee of Machine Learning Techniques in a load forecasting Architecture

In smart grids, load time series dynamics are influenced by customer groups. Load forecasting models
require short development time, moderate computational task and a small amount of employed data. A

combined application of four modeling schemes in a load forecasting tool is investigated in this study. The
proposed implementation employs four Prediction Units accompanied by a Combination Unit, shown in
Figure 4.19. The Prediction Units accommodate multiple predictors designed to capture the entire
dynamics of all possible load time series and accurately perform within a 24-hour forecasting window. Four
methodologies, Generalized Linear Models (GLM), Gaussian Processes (GP), Multilayer Perceptrons (MLP)
and Random Forests (RF), are applied to the proposed model predictors and their performance in accuracy
and in computational time is investigated. The data employed originate from a MV/LV substation located at
the Zografou area in Athens.
Prediction Unit
Short-term
Normal days: 24 models
Special days: 2 models
Prediction Unit
Short-term
Combination Unit
Prediction Unit
Long-term
Prediction Unit V
Long-term
Figure 4.19 The structure of the proposed load forecasting model
In specific, Prediction Units I and III consist of 24 predictors that correspond to each hour of the day and are
applied in the regular days. For the special days, two predictors are applied; one at the high consumption
hours of the day (from 09:00 to 15:00 and from 18:00 to 22:00) and one at the low consumption hours
(from 16:00 to 17:00 and from 23:00 to 08:00).
In Prediction Units II and IV, there are also 24 predictors applied to predict the load in regular days, but
they correspond by two to a different month. In particular, the regular days are divided in two time
periods: the low demand time period (from 23:00 to 08:00 and from 16:00 to 17:00) and the high demand
time period (from 09:00 to 15:00 and from 18:00 to 22:00. Regarding special days, there is one predictor
trained with the special days occurred in high temperature year period (from April to September) and also
one predictor activated in special days occurred in low temperature year period (from October to March).
Prediction Units I and II receive the most recent load data and perform better in short term horizons, while
Prediction Units III and IV receive the load values of the last seven days that correspond to the similar hours
with the prediction hour and perform better in longer horizons than 6 hours ahead.
Finally, the Combination unit receives the Prediction Units predictions. It applies 24 MLPs which one is
activated to a specific hourly forecasting horizon.

The proposed model requires the training of the 104 predictors of four prediction units, as well as 24 MLPs
in combination unit. In order to be applicable in a smart grid environment, the computational time of its
training procedure should not exceed the order of minutes.
The RF parameters are the number of trees and the number of variables used to build a tree. The decision
was simple: the trees number was set to 1000 and the variables number was set by the Breimans rule, one
third of the total number of input variables. The proposed model at this case was trained in 5.65 min.
When GP are implemented as predictor inside the proposed model, the training time outreaches 18 min.
The design of GP requires the selection of the mean, covariance and likelihood functions. Testing several
functions to a single predictor case, the optimal performance was achieved by constant and equal to the
mean of response vector as mean function, linear as covariance function and Gaussian as likelihood
function with zero mean and standard deviation equal to the standard deviation of response divided by 12.
For GLM, the Gaussian joint distribution function was used for the input variables and as linkage function it
applied the identity function. Finally, for MLPs the number of hidden layer neurons was set to 24. The
computational training time of these two methods was 2.37 min and 25.79 min respectively.
4.4.3 Total MW and MVar Prediction Case Study

4.4.3.1 Total MW Prediction in Greece
1) Time Series Characteristics: Combined unit, a modelling approach different from those used for
traditional load forecasting is proposed according to the time series characteristics. The proposed
methodology uses an actual load time series from May 1, 2013 until April 5, 2014 along with the
corresponding daily maximum predicted temperature values. The load time series has been recorded with
a 15 minute time step at an MV/LV substation feeding an urban neighborhood in Athens. A large number of
students (a consumer group following non-ordinary and highly stochastic electricity consumption patterns)
living in the area but holding permanent residency outside Athens, influences load consumption in the
following ways. It is customary for university students to leave Athens during summer holidays (July and
August) as well as for Christmas and Easter breaks. The consumption at the beginning and the end of the
summer holiday is unknown due to the variability of the university exam periods. Figure 4.20 presents the
recorded load time series where it appears that students use air conditioners for heating during the winter,
but rarely use them for cooling during the summer.
140
120
100
LOAD (MW)
80
60
40
20
0
01/05/13 26/05/13 25/06/13 25/07/13 24/08/13 23/09/13 23/10/13 22/11/13 22/12/13 21/01/14 20/02/14 22/03/14
Date
Figure 4.20 Load time series from May 1, 2013 until April 5, 2014

Figure 4.20 depicts the different periodicities of the studied load pattern as compared with load time series
obtained by a transmission system. In the studied series there are no significant differences observed in the
consumption patterns during spring, summer and autumn. Minimum load occurred in spring represents
just 12% of the annual maximum load, namely the consumption is ranged in a high values space. In winter
months, high fluctuations are observed during the day, with the daily peak being roughly three times the
minimum demand (also shown in Figure 4.21). Besides, the maximum demand observed during the winter
is 65% higher than that recorded throughout the summer. Figure 4.22 shows the demand during a week in
the winter. With the considerable variability of the daily load curves observed in Figure 4.21 and Figure
4.22, the applied case study could be considered as a challenge for the existing load forecasting models.
A week in winter period
130
120 Sunday Monday Tuesday Wednesday Thirsday Friday Saturday
110
100
90
Load (MW)
80
70
60
50
40
30
15/12/13 16/12/13 17/12/13 18/12/13 19/12/13 20/12/13 21/12/13 22/12/13
Figure 4.21 Load demand during a week in the winter
A week in summer period

60
Sunday Monday Tuesday Wednesday Thirsday Friday Saturday

55
50
45
Load (MW)
40
35
30
25
20
22/09/13 23/09/13 24/09/13 25/09/13 26/09/13 27/09/13 28/09/13
Figure 4.22 Load demand during a week in the summer

In load time series modeling, it is customary to consider Sunday as a special day because the demand is
lower than other days. However, in the application discussed, the days with the lowest energy consumption
during the week are Saturdays (shown in Figure 4.21 and Figure 4.22) while Sundays appear to have
behaviors/performances similar to those of weekdays. In fact, application of the Kruskal-Wallis hypothesis
test shows that only Mondays, Wednesdays and Thursdays exhibit similar behaviors while Tuesdays,
Fridays, Saturdays and Sundays form separate populations. Regardless of the Kruskal-Wallis test results,
only Saturdays are considered as special days in this application for they possess similar attributes to
celebration days. In both of cases shown in Figure 4.21 and Figure 4.22, the daily load profiles appear to
change randomly during the weekly period while large random fluctuations occur during different hours
each day. The level of uncertainty in the recorded time series proves that this case-study is suitable for
testing the proposed methodology.
8000
7000
Daily consumption (MWh)
6000
5000
4000
3000
2000
5 10 15 20 25 30 35 40
Temperature (oC)
Figure 4.23 Daily consumed energy versus daily maximum temperature
There is a well-documented correlation between load demand and atmospheric temperature. More
specifically in Greece, the load demand is strongly correlated with high temperatures during the summer.
However, the student summer holidays reduce the high temperature influence to load demand at the area
of interest, while high temperature/load correlation appears when temperatures drop below 20o C. Figure
4.23 illustrates the maximum daily temperature against the daily energy consumption, clearly showing the
strong effect of low temperature to consumption. It should be taken into consideration that the winter this
year in Athens was warmer than usual.
Special days load curves
100
90 Holy Saturday
Easter Sunday
80 National celebration
Christmas Day
70 New Years Eve
Load (MW)
60
50
40
30
20
10
0 5 10 15 20 25
Hour
Figure 4.24 Load curve during a few special days

Considering that special days include Greek national holidays along with religious celebrations, it is
customary to extract different load profiles, shown as Figure 4.24. For example, during Christmas, the
Christmas Day and New Years Eve present a different profile when compared with the remaining days
which possess similar profiles during that period. The load profile of Holy Saturday is different from that of
Easter Sunday. The load time series recorded during Greek national holidays appear to be similar, though
different from other special days. Employment of the Kruskal-Wallis test groups the special days according
to their load profile similarity, with an index predefined for each group.
2) Prediction Result: From the training period covered from May 1, 2013 until April 5, 2014, three weeks
were removed and used as the evaluation period. These weeks correspond to the periods from 23/08/2013
to 31/08/2013, from 22/11/2013 to 30/11/2013 and from 22/02/2014 to 28/02/2014. The Figure 4.25-
Figure 4.27 shows the mean absolute error of the four versions of the proposed model to these three
periods.
Performance in summer testing period (23/08/2013-31/08/2013)
8
7.5
6.5
% Mean Absolute Error
5.5
MLP
5
GP
4.5 RF
GLM
4
3.5
3
0 5 10 15 20 25
Look ahead hours
Figure 4.25 The mean absolute error of the proposed model at the period 1
Performance in Automn testing period (22/11/2013-30/11/2013)

12
11
10
8
MLP
7 GP
RF
6 GLM
4
0 5 10 15 20 25
Look ahead hours

Performance in winter testing period (22/02/2014-28/02/2014)
11
10
9
7 MLP
GP
6 RF
GLM
4
0 5 10 15 20 25
Look ahead hours
4.4.3.2 Total MW and MVAR Prediction in UK

Similar approach can be applied for reactive power prediction. Figure 4.28 shows the predicted demand (in
MW and MVAR) at a distribution level bus. The prediction is 24 hour ahead, because it will be further used
for demand side management (DSM) actions. It is shown that the predicted load curves for both P and Q
are close to the actual curves. MAPE of P and Q are 4.27% and 17.49%, respectively.
(a) (b)
6.5 2
6 Actual
1.75
5.5 Predicted
Q(MVAR)
5 1.5
P(MW)
4.5 1.25
4 1
3.5
3 0.75
2.5 0.5
0 3 6 9 12 15 18 21 24 0 3 6 9 12 15 18 21 24
Hour Hour
Figure 4.28 Predicted demand versus actual demand: (a) real power, (b) reactive power
4.5 RES FORECASTING

In this chapter, initially a brief description of the usage of forecast system is provided, along with technical
details and a documentation of the applied methodology helpful for this software to be more
understandable. Also, the required programming environment, inputs and expected outputs, as well as the
modules procedure of training and operating are presented in detail. The aforementioned information is
useful because it shows the importance of the forecast module integration in different sites and is

interaction with other applications such as Data Analytics and VPP Platforms, DRMS etc.ICCS will develop
forecasting modules for all sites except Meltemi. Regarding Meltemi Forecasting modules are under
development in a different Research Project (SmarterEMC2).
In order to estimate the RES production in medium time scales (8 hours to 48 hours) various non-linear
statistical methods are utilized along with numerical weather predictions. These forecasting algorithms
focus on probabilistic forecasting, namely on improving the accuracy as well as the estimation of expected
error (http://www.anemos-plus.eu/). Hence, the information send in the operator is comprised from point
forecasts and confidence intervals, which can be used for risk evaluation of the systems operations.
The role of forecasting algorithms concerns virtual power plants (VPP) and demand response (DR). In
particular, VPP will make use of forecasting modules for market participation and more specifically, for the
intraday market while providing voltage control services. On the other hand, forecasting models are also
utilized from DR aggregators both for market participation and supporting technically the distribution grid.
Therefore, Demand Response Management System (DRMS) will be able to optimally operate, designating
efficiently load curtailment/enhancement targets to DR events and participant resources.
4.5.1 Methodology
In the following, an overview of the applied algorithms will be presented which are based on ensemble
learning, on radial basis function neural networks and on Random Forests (RFs) and that will be used in the
project.
Wind power forecasting is a non-linear approximation problem with different non-linearity subspaces.
Multiple forecasters can be used operating in overlapping subspaces of the problem. In particular, each
part of the problem is described by different forecasters creating an ensemble of individual hypotheses.
These individual forecasters create an ensemble forecast which is a set of predictions and are combined
providing the final estimation. These estimates are often much more accurate than individual predictions as
Dietterich in [77] has proven.
For the developing of multiple forecasters, RBFs with different width scales are very useful. In particular,
RBF width is inversely related to a subspace local non-linearity, which is efficiently described by an
adequate number of RBFs with specific centers ci,j and different widths bi,j (for every variable i and for every
RBF j). Therefore, the centers and the widths of the required RBFs should be selected according to the non-
linearity of subspaces [78].
In this project, the recommended algorithm consists of an adaptive neural structure RBFNN associated with
RFs, as shown in Figure 4.29, supporting both real-time and off-line in batch mode operations. In particular,
the RBFNN first classifies the input data to overlapping areas, determined by clusters each of them is
connected with a RF. Therefore, the number of the RBF clusters equals the number of the RFs. The
activated clusters perform the corresponding RFs creating an ensemble prediction. The ensemble is
combined at the output of the proposed model using recursive least squares or weighted average
algorithms.

INPUT
DATA
Cluster 1
Cluster 2
Cluster 3
Cluster M
RBFNN ..
RF KM
RF K1
RF K2
RF K3
Radom Forests
RF 1
RF 2
RF 1
RF 2
RF 1
RF 2
RF 1
RF 2
..
.. .. .. ..
Weighted average
Figure 4.29 The structure of the proposed model
4.5.1.1 Radial Basis Function Neural Network (RBFNN)
A RBFNN consists of two neuron layers [79, 80], a hidden and an output layer. The input space d is
mapped by non-linear kernel functions at the hidden layer, while the kernel outputs are translated to the
desired result by linear functions in the output layer. In particular, the input vector xi,n is associated with
centers ci,j of the kernel functions fj, depending on their width bi,j in such a way that the kernel outputs oj ,
are as linearly dependent as possible, with the desired outputs yn. Here i is the number of the input
variable, j is the number of the kernel function in the hidden layer, d is the dimension of the input space
and n the number of the input pattern.
In RBFNNs, the kernel function that is very popular and efficient, is the Gaussian function. In this case, the
outputs of the hidden layer, called here as RBFs, are given by
o j f j ( xi , ci , j , bi , j ) (4.85)
with
2
xi ci , j

bi , j
fj e
(4.86)
The first layers RBFs have a different width bi1 for each input variable i. The output of a first layers RBF is
given by

2
xi c1i ,m
1

bi
f m1 e
(4.87)
1
where xi is the input vector and ci , m is the center i of mth RBF. In general, the initial setting of the widths
1
for each input variable, the number M and the positions ci , m of the RBFs, can be controlled. For example,
in most cases, the RBF widths bi,j are chosen constant for all the RBF centers ci,j and estimated either by
1
trial-and-error or by an optimization algorithm [79, 80]. However, the selection of the widths bi is also
based on the relative importance of each variable to the problem and can be set also as the ratio or a
multiple of the variable range. In addition, the widths set can be amplified with a coefficient which controls
how the knowledge is distributed in RFs, namely the number of the RBF clusters activated by an input
vector.
The first layer is trained sequential, receiving the input vectors one by one. This is done irrespectively if the
network is trained in batch mode or online. The first RBF of the RBFNN is created always with the first input
vector of the training set, while the next RBFs using a simple rule. Initially, the outputs f1 from the existing
RBFs are computed for every input inserted to the layer. When all the activations f1 of the existing first
layers RBFs are less than a constant f, which is usually set between 0.25 and 0.5, then a new RBF cluster is
added. However, it is possible more than one RBF clusters to be activated by an input. This can be achieved,
when the respective output f 1 is greater than zero and using the above RBF construction rule (
f m1 f ' m {1..M} ). Thus, an input vector belongs to multiple input areas defined by the RBF clusters
and is used for the training of multiple RFs.
4.5.1.2 Random Forests (RFs)
In classification and regression problems a learning method that is used commonly is Random Forests (RFs)
[81, 82]. This method is a modified version of the bagging method and it is based on the generation of tree
decisions in a sample of the whole data set during the calibration phase of the model. Then, for the
determination of the final output, all trees are taking into account. In RFs, each tree is constructed using a
different bootstrap data sample and each node is split using the best among a subset of predictors
randomly chosen at that node. Finally, the algorithm of RFs has two significant characteristics: it does not
overfit the data, even with the increase of the number of trees B [3.5] and there is no need for cross
validation.
More analytically, the algorithm of random forests for the regression analysis can be determined as follows
[82]: Let there be a set of data {(1 , 1 ), (2 , 2 ), , ( , )} with input variables (predictors) with
p-characteristics and output variables (responses). In addition, let there be the set of the total trees
which are generated.
1. For b=1 till B:
i. A random sub-set Z of magnitude is selected from the training set (bootstrapping).

ii. In the sample, a random tree Tb is developed and for every terminal node of the tree the
following steps are repeated retrospectively, till a minimum number of nodes nmin is
reached.
i. From p-characteristics, m random characteristics are selected.
ii. The best discrimination variable/point from m, is chosen.
iii. The division of a node into two nodes-children, takes place.

2. The output of the mean of the tree ensemble { b }1 .
The prediction of a new point x is given by

1

() = (; ) (4.88)

=1
The random forest reaches the expectation
() = [(; )] = lim

() (4.89)

with the mean of trees B in the distribution . At the output of the model the combination of the ensemble
predictions pm of RFs provides the final prediction, while the ensemble predictions are weighted aggregated
with the cluster activations of the RBFNN as
p f m1,n pm (4.90)
In the following and in Figure 4.30 and Figure 4.31, examples of training an classification tree as well as the
output training ensemble are presented [83]. In particular, in Figure 4.30 an example of random
classification tree training is depicted. First, candidates for classification are chosen corresponding to data
randomly derived from the whole training set. Then, a split point is selected from this data set. This point
corresponds to the maximum gain of information as well as to the data in which the discrimination will take
place. This discrimination generates two nodes-children, SL and SR, and this process is repeated till all data
are fully separated. The evolution of the separation increases the confidence concerning the distribution of
the data in classes. In the example presented in Figure 4.30, even though the data v in the initial tree root
have the same probability distribution, it can be seen that if the route from the tree root to leaf C is
followed, the data v has a very high probability of belonging in the green class.

Figure 4.30 Training of a Classification Tree [83]
Figure 4.30 provides presents an example referring to the creation of an output corresponding to data v. In
particular, classification predictions for data v are generated from each random tree, which were created
from sequential divisions. The trees are trained separately, while each tree gives different predictions.
Specifically in all trees, the point v, starting from the root, proceeds forward till it reaches the appropriate
leaf. There, it is related to estimation. All this takes place during the phase of the models testing. In the
example presented in Figure 4.31, the most confident prediction is generated in the tree t = 2, a result that
indicates that the point will belong to the class with the green color. On the contrary, the most uncertain
prediction corresponds to the tree t = 3, which has a more homogeneous probability distribution of all
classes. The final prediction for the point is derived from the whole ensemble of the random trees and is
the mean of the predictions for this point in every leaf pt (c|v), namely:
T
1
p(c|v) = pt (c|v) (4.91)
T
t
Figure 4.31 Output of the training ensemble [83]

4.5.2 Environments, Module Inputs and Outputs
For the implementation of the forecasting algorithms the hosting machine needs to have the following
software and hardware characteristics:
MathWorks Matlab software
Microsoft Windows operating system (7, Server 2008 and above)
Quad Core 64 bit processor (Intel or AMD)
8 GB RAM
300 GB of storage space (for the weather predictions)
Connection to the internet
For various forecast target objects (e.g. Wind Farms, PVs and Load) an executable file is created which
includes the necessary code and structures for model training and forecasting. It must be noted that the
models will be initially untrained so as to be able to be trained for different sites and data sets.
4.5.2.1 Inputs
In order to train and produce reliable forecast data, first of all numerical weather predictions (NWP) must
be considered with a time window of at least 48 hours ahead. The NWP along with actual measurements
are utilized to train and run the forecast models. In addition, a set of configuration data must be available
corresponding to the specifications of the site considered (GPS coordinates of the wind farms, nominal
power etc). Finally, for the forecast modules a continuous flow of past production measurements is
needed. The format of these measurements should be in comma separated file (CSV) including the date
and the actual production of each of the sites.
4.5.2.2 Outputs
The outputs of the models consist of CSV files including the anticipated hourly production (or load) for a
time period of 48 hours ahead for each point of interest for each of the sites, the appropriate quantiles
where applicable. In particular, the CSV file includes a column with the RES ID, a column with the date the
prediction was made (date/time) and a column with the date/time the prediction is for. The actual forecast
data follow after these columns in the form of 10 columns containing the actual forecast, standard
deviation and quantile information.
4.5.3 High Level Forecast Service

In the following a brief description of the Forecast service is provided (a schematic overview is shown in
Figure 4.32). It is developed to schedule and monitor the running of all the forecast models. Therefore, the
Forecast Service is developed as a multithreaded host application. This means that the Forecast Service can
utilize all the available underlying processor threads, multiple forecasting cases can be run simultaneously
and the modules themselves use parallel processing. The Forecast Services primary target is the provision
of all the necessary data to the modules by interacting with an already existing measurements database. If
the data provided are present and valid, as described in previous paragraph, the calibration process is
automatic, and when the training ends, the model (an executable file) is able to initiate forecasts. However,
the completion time strongly depends on the time period that is used.
As it can be seen in Figure 4.32, the Forecast Service is separated from the database (which is in a separate
network Database server) hosted on any machine with characteristics described previously. In order for the

Forecast Service to read and write to the database, a common Database Layer is used as an intermediate.
The Forecast Service also has Data Transformation Service (D.T.S) module. This module is responsible for
the appropriate data format needs, handling the exchange between the service and the forecast modules.
Figure 4.32 High level overview of the Forecast Service
Another important feature of the Forecast Service is the Scheduler. This module first acquires the real-time
and sites configuration data from the Database Layer and drives them to D.T.S. The result of the DTS
process is reported back to Scheduler. Then the Forecast Service initiates the forecast process by calling
the proper forecast module in a pre-determined time interval (e.g. hours) that lies on a service
configuration file. Then, the Scheduler runs the forecast module and monitors its progress.
In addition, the forecast module is connected to the Numerical Weather Forecasts Service acquiring
numerical predictions. As mentioned before, the numerical weather predictions along with configuration
and input files are needed for the initiating of the training and forecasting processes. When this task is
completed the Scheduler is notified for the results of the forecast modules operation, which uses this
result invoking the D.T.S module. If the forecast is successful, the Scheduler uses the Database Layer again
to store the output to the database, whereas a failure is only logged for error tracing.

5 INTEGRATION OF FUNCTIONALITIES IN GRID MANAGEMENT
5.1 INTRODUCTION
With increasing demand ( EVs, heat pumps) and local generation (both DER), the Blind but happy
operation of DSOs is no longer sufficient:
- Time: Changes in demand and generation happen too fast. Traditional solutions (investments in the
grid) require long term planning and no longer can keep up.
- Finance: it will become too expensive to increase the grid capacity to accommodate 2 way flow
(Solar!) and the possible large peak caused by EVs and electric climate control (airco and heat
pumps)
- Opportunity:
o Smart devices and smart metering make disclosure of DER feasible
o Other business cases require disclosure of DER as well, especially balancing renewables
(solar and wind) on a (inter-)national scale.
The last point is both a opportunity as a threat for DSOs. It might induce a lot of synchronous behaviour
(imaging EVs all reacting on a dynamic price or TOU price from a retailer or offering FRR to a TSO).
On the other hand, it is an opportunity because the required infrastructure for demand response (including
remuneration etc.) can also be used for grid capacity management.
The alternative for the blind but happy is using the local DER (smart flexible demand and storage, local
generation etc.) to solve the problems. However this requires that the DSO has a smart system to predict
possible problems, evaluate possible solutions and implement the most suitable one.
The core of such a system is described in chapter 4: State estimation, RES prediction and load flow.
Network risks
(load flow)
Activities to
reduce risks
Grid Operations Network status

(DMS, switching, Data Acquisition Strategic threats
(state estimation
disturbances, events, (Measurements) and scenarios
workforce deloyment) and RES prediction)
Activities to
optimize operations Network Other processes
(Strategy, Asset
optimization Management
(Load flow) Investments)
Figure 5.1 Illustration of integration of functionalities in grid management

RES prediction, State estimation and Load flow combined make It possible to anticipate future events and
proactively optimize accordingly to reduce risks or optimize operations like work flow management and
possibly even grid losses.
5.2 STANDARDIZED PROTOCALS AND INTERFACES

In this section we are reporting the external interfaces of grid management services (as reported above)
with NOBEL GRID tools that will further enable risks mitigation and optimized grid operation. The outcomes
from grid management services are further transformed to grid management strategies, triggered to the
associated NOBEL GRID components.
5.2.1 Standardized protocols between DSO and Aggregator - Interfaces with DRFM framework
Within NOBEL GRID project we are addressing the role of DSM Aggregator as the business stakeholders that
acts on behalf of DSOs towards optimal management of prosumers portfolio in abnormal grid conditions.
The goal of this section is to define interfaces between DSO and Aggregator towards the implementation of
grid management Demand Response strategies. This is a Demand-Side-Management (DSM) service for
congestion avoidance in the distribution grid, as a network service.
The Demand Side Aggregator gathers (aggregates) the flexibilities of consumers in order to help
electricity network utilities - grid and transmission operators - to shave peak power demands, balance
intermittent power generation, increase security of supply etc.. This is the case examined in NOBEL GRID
project, where DSO triggers (under specific grid conditions) the appropriate grid strategies to Aggregator
for further exploiting the amount of demand side flexibility.
An initial presentation of DSO - Aggregator interfaces was reported in D9.1 (G3M Framework functionalities
specification and design) and D12.1 (Specification and design of a NOBEL DRFM cockpit) but the analysis
was focusing mainly on the user perspective of DSO and Aggregator respectively (functional definition of
DSO Aggregator interfaces). In this section, the technical implementation of these interfaces is reported,
by adopting the USEF protocol which explicitly specifies the DSO - Aggregator interfaces. A summary of
USEF protocol is reported prior to the definition of technical interfaces.
5.2.1.1 Introduction to USEF framework

One of the main objectives of DRFM framework is to provide services to DSOs. Towards this direction we
are examining the adaptation of USEF framework in NOBEL GRID project. USEF provides a modular design
for smart energy systems that can be customized to the needs of smart grid project implementations. It
guarantees the interoperability of products and services and it ensures that solutions become repeatable
and future proof, thus safeguarding investments in the smart energy future.
USEF provides a minimal set of specifications to secure the essential interoperability between all the
components in a smart energy system. The main focus is on the business perspective and not on the
technical implementation of a DSM framework, where OPENADR standard is clearly covering this aspect. To
optimize the value of flexibility across all roles in the system, USEF introduces a new market-based
coordination mechanism (MCM) along with new processes. The USEF Market-based Coordination
Mechanism (MCM) facilitates the delivery of value propositions (i.e. marketable services) to various market
parties without imposing limitations on the diversity and customization of the propositions. The USEF MCM
is designed for all energy commodities and enables the market to optimize in time, capacity and power.
MCM provides access, under equal conditions, for all stakeholders to a single integrated market. This
unique approach aims for a future-proof design of the energy market. The USEF MCM operations scheme

distinguishes four phases and the corresponding processes can be summarized as:
Plan: In the Plan phase the demand and supply of energy are planned for the upcoming period, usually
a calendar day. Both the BRP and Aggregator carry out an initial portfolio optimization. During this
phase, the BRP can procure flexibility from its Aggregators. The Plan phase results in an agreed-upon
Aggregator plan (A-plan) between the Aggregator and the BRP.
Validate: In the Validate phase, the DSO validates whether the demand and supply of energy can be
distributed safely without any limitations, with the use of D-prognoses. If congestion occurs, the DSO
can procure flexibility from Aggregators to resolve the grid capacity issues. It is important to note that
iterations exist between the Plan and Validate phases. This means that after validation, it is possible to
go back to the Plan phase. These iterations take place until the foreseen energy flows can be
distributed safely in an economically optimized way.
Operate: In the Operate phase, the actual assets and appliances are dispatched and the Aggregator
adheres to its D-prognoses and A-plan. When needed, DSOs and BRPs can invoke additional flexibility
from the Aggregators to resolve unexpected congestions (for the DSO) or to provide balancing services
or re-optimize their portfolio (for the BRP).
Settle: in the Settle phase the flexibility that the Aggregator has sold to the BRPs or DSOs is settled. For
this purpose the actual consumed and produced volumes are first allocated to the responsible parties;
unresolved or disputed volumes are reconciled shortly thereafter.
The next schema presents the 4 business layers of USEF framework, highlighting also the different services
to be supported by each layer:
Figure 5.2 USEF Protocol Business Framework

NOBEL GRID project is not examining the impact of Demand Response Strategies for TSOs and BRPs though
special interest is delivered on the provision of services to DSOs (along with services to prosumers that are
not considered part of USEF framework). The USEF framework identifies different flexibility services for the
DSO which provide value by helping the DSO increase its performance and efficiency in managing the
distribution grid.
Congestion management refers to avoiding the thermal overload of system components by reducing
peak loads. In contrast with grid capacity management, this is a situation where failure due to
overloading may occur. It is a short-term problem (with respect to the duration of a grid reinforcement
project) for the DSO that requires a relatively swift response. The conventional solution is grid
reinforcement (e.g., cables, transformers). The alternative (load flexibility) may defer or even avoid the
necessity of grid investments.
Voltage problems typically occur when solar PV systems generate significant amounts of electricity.
This will push up the voltage level in the grid. Using load flexibility by increasing the load or
decreasing generation is an option to avoid exceeding the voltage limits. This mechanism can reduce
the need for grid investments (such as automatic tap changers) or prevent generation curtailment.
Grid capacity management aims to use load flexibility primarily to optimize operational performance
and asset dispatch by reducing peak loads, extending component lifetimes, distributing loads evenly,
and so forth. An added benefit may be the reduction of grid losses.
Controlled islanding aims to prevent supply interruption in a given grid section when a fault occurs in a
section of the grid feeding into it.
Redundancy (n-1) support refers to actions that help reduce the frequency and duration of outages. An
example is supplying emergency power (or shedding loads) in the event of a severe power shortage, or
supplying backup power during grid maintenance activities.
The USEF specification is technology and implementation agnostic. It standardizes the logical interfaces and
defines the minimum functionality of the components in the form of use cases and message transport
specifications and message descriptions. While the USEF specification is technology and implementation
agnostic, there are some basic principles to which any USEF implementation architecture must adhere. The
respective principles are also addressed in the definition of interfaces in NOBEL GRID project, where
specific USEF functionalities will be supported taking into account NOBEL GRID business scenarios and use
cases.
5.2.1.2 USEF Adaptation in NOBEL GRID Framework

It is clear that the focus in NG project is to enable Flexibility trading between Aggregator and DSO. In order
to solve any expected congestion, the following steps are defined by USEF framework:
1. The DSO requests Aggregator at the congestion point to provide flexibility. A Congestion Point is a set
of connections, which are (directly) related to that part in the grid where the grid capacity might exceed
for instance the secondary side of an LV transformer. In this request, the DSO indicates the
magnitude (amount of excess power) and timing (PTU) of the expected congestion, and how much
capacity is available in the remaining PTUs. More specifically USEF specifies that when sending a
flexibility request, the DSO provides the following information about expected congestion:
a. Information about expected congestion is provided to Aggregator

b. Information about expected congestion includes the reason for overload
(production/consumption)
c. Information about expected congestion includes the amount of reduction needed
d. Information about expected congestion includes available grid capacity for other PTUs
2. The Aggregator receives the flexibility request for adjusting distribution requirements.
3. Then, the Aggregator creates offers for the flexibility. Within this message structure the Aggregator can
offer two types of flexibility:
a. A reduction or increase in load for one or multiple PTUs (eg load shedding or increase of
production)
b. A reduction or increase in load for one or multiple PTUs combined with the inverse effect on
other PTUs (e.g. load shifting)
The FlexOffer message contains the flexibility (essentially a positive or negative volume per PTU) and
the price for the flexibility.
4. The DSO receives the flexibility offers
5. The DSO procures flexibility to resolve the congestion issues, either by placing an order for the
flexibility the Aggregator have offered for that specific time (short-term flexibility options), or by
activating flexibility options in prearranged bilateral contracts (long-term flexibility options) with an
Aggregator. The FlexOrder message is used by USEF for buying the flexibility from to the Aggregator. It
has a close relation with the original FlexOffer message. The FlexOrder is used to signal the acceptance
of the offer. If the Aggragator wants to withdraw a FlexOffer before it has been accepted the
FlexOfferRevocation message is used.
6. The DSO determines whether the expected congestion is solved using the ordered flexibility. If this is
not the case, the system moves to the Orange regime. In this regime the DSO temporary overrules the
market by limiting connections in the overloaded part of the grid
7. The Aggregator receives the flexibility orders, resulting in the actual procurement of flexibility by the
DSO from Aggregators.
For the procurement of flexibility by the DSOs, USEF specifies the following rules:
A flexibility offer is valid until a new flexibility offer is sent by the Aggregator, until it expires or is
revoked. It is up to the Aggregator to determine the acceptance deadline.
A flexibility order is definite and binding once it has been placed.
The DSO chooses which flexibility offers it accepts to solve the congestion. In this regard, it is not
obligatory for the DSO to start with the offer that has the lowest price. The DSO has the freedom to
assess the balance between that price and quality of the flexibility offered in both long term
contracts and short term offers.
Bidding takes place at congestion point level, making every congestion point a local flexibility
market.
The DSO only sends a flexibility request when the result of the grid safety analysis leads to
congestion. This ensures that biddings are only made if the DSO has need for them.
A flexibility order is linked to a flexibility offer for settlement purposes.
For the last part of the process, where settlement between DSO and Aggregator takes place, the DSO is
responsible for settling the flexibility that is has acquired from the Aggregator. Within this settlement, the
DSO needs to check whether the acquired flexibility has been delivered according to the agreements. If not,
this can lead in certain circumstances to penalties, which is considered an integral part of the settlement

process.
Figure 5.3 - Sequence Diagram Cockpit/G3M
The purpose of this part of the paragraph is to give a formal approach to the process described in the
previously. The starting point is defining the sequence diagram of the Figure 5.3: the various steps of the
before list of the corresponds to a particular message of this diagram.
In Figure 5.3 there is the list of all messages with their descriptions. All messages contain Metadata
information (for example sender and receiver, roles, timestamp and others). Also each message is linked to
its previous message and to the following one using Sequence indexes, so the entire process is
synchronized at semantic level. The response messages contain only metadata information and their aim is
only acknowledging the request message, so they will not be described at this level.
A. FlexRequest
A FlexRequest element is composed substantially by a list of PTUs (Program Time Unit). It is
important to understand the meaning about this PTUs: this is the minimum granularity at which
the energy will be managed. So for example if the minimum granularity is of 1 hour, a Period (that
is 1 day) will be of 24 PTUs. Again, if the minimum granularity is of 15 minutes, a Period will be of
96 PTUs. Each PTU includes a Power value, the sign of which is used to distinguish between
production and consumption, from the perspective of the Prosumer. A positive value indicates that
power flows towards the Prosumer (consumption), a negative value indicates flow towards the grid
(production). So:
+ means Power flows towards prosumer

- means power flows towards the grid

Figure 5.4 - Sign interpretation
The two different type of Power in a FlexRequest can be obtained in two different ways (or
combination of these two ways) by the Aggregator, as descried in the Figure 5.4.
For example a request with a Power of -20 in a PTU means that DSO wants 20 watts (in the unit
measure of the energy) come in its grid: this quantity will be obtained by the Aggregator using
DRs increasing production or reducing consumption (or a combination of them). Lastly, attention
should be given to the managing of the PTUs sequentiality: as described previously each
FlexRequest has a related day Period and each PTU is the minimum granularity of an energy
event: depending on this setting each Period contains a variable total number of PTUs. The PTU
field can refer to one or more PTUs depending on the filed duration, and these PTUs starts after
an offset of PTUs as described in the field Start, which means that a certain numbers of PTUs
have been already passed and/or processed. An example will better clarify this: supposing PTUs
have duration of 60 minutes (so a Period is of 24 PTUs), supposing DSO receives a Forecasting (D-
Prognosis) message specifying an issue (high consumption) between the 12:00 and the 16:00: this
means that the FlexRequest related to this forecast refers to 2 PTUs (two hours so 120
minutes) starting from the 12-th PTU (supposing hours will be indexed 1 to 24).
B. FlexOffer
FlexOffer messages are used by Aggregators to make to the DSOs an offer in order to provide
flexibility. A FlexOffer message contains a list of PTUs, with for each PTU the change in

consumption or production offered, plus the price for this amount of flexibility. FlexOffer
messages should only be sent once a FlexRequest message has been received and must never
be sent unsolicited. Note that multiple FlexOffer messages may be sent based on a single
FlexRequest: for example, one offer that exactly matches the power reduction requested, plus
one with a different amount of reduction, with more favourable pricing. When responding to a
FlexRequest, an Aggregator may send an empty FlexOffer message (i.e. a message not
containing any PTU elements) in order to indicate that no flexibility is available. The PTU entity
should contain also the price attribute for the flex Aggregator is offering to the DSO.
C. FlexOrder
FlexOrder messages are used by DSOs to purchase flexibility from an Aggregator based on a
previous FlexOffer. A FlexOrder message contains a list of PTUs, with, for each PTU, the
change in consumption or production to be realized by the Aggregator, plus the accepted price to
be paid by the DSO for this amount of flexibility. This PTU list should be copied from the
FlexOffer message without modification: Aggregator implementations will (and must) reject
FlexOrder messages where the PTU list is not exactly the same as offered.
D. FlexRevocation
The FlexOfferRevocation message is used by the Aggregator to revoke a FlexOffer
previously sent to a DSO. It voids the FlexOffer, even if its validity time has not yet expired,
even if a FlexOrder has already been issued based on this offer. The FlexOffer should exist and
have been previously acknowledged, though, and may NOT apply to a period of which one PTU is
already in the operate phase.
E. SettlementMessage
The SettlementMessage is sent by DSOs on a regular basis (typically monthly) to AGRs in order
to initiate settlement. It includes a list of all FlexOrders placed by the originating party during
the settlement period.
It contains a list of FlexOrderSettlement elements and a FlexOrderSettlement contains
settlement details for a single FlexOrder.
Also the FlexOrderSettlement entity is composed by a series 0 to N of PTU-Settlement
entities where a PTU-Settlement element contains settlement details for a single PTU (or set of
identical PTUs) belonging to the FlexOrder placed by the party originating the message during
the settlement period.
Figure 5.5 - Forecasting load

In order to better understand these parameters the following figures are useful. In the Figure 5.5
we can se the forecast about power for each PTU (PrognosisPower). The yellow section of PTU
number 3 is the amount of power the Aggregator want to use as flexibility. For example, assume a
residential Prosumer with an amount of uncontrollable load and an electric vehicle (EV) with
smart charging equipment. The initial forecasted total load profile of both the controllable and
uncontrollable load is the blue trend in the graph.
Figure 5.6 - Flexibility obtained applying DR by Aggregator
The Aggregator (AGR) is able to indirectly control the EV charging process and change its load
pattern by activating the available flexibility. In the forecast, EV charging would originally take
place at PTU=3. To create additional value, AGR shifts charging to PTUs 4 and 5. This is depicted in
Figure 5.6.
Figure 5.7 - Measurement by DSO after PTU is over
The measured load profile at the main meter (SMX) is shown in graph of Figure 5.7: the figure
describe the effective redistribution of power load. For example in the PTU number 4 the blue
section is the measured ActualPower.
In conclusion the SettlementMessage can be seen as a periodically (monthly, weakly, )

balance in which the DSO reports settlements of all the FlexOrders ordered during this period
with others Aggregators.
At the end, these concepts have to be clear to the components and actors have to use these standard:
o Keep consistence between the sequence attributes across the different messages in the flow;
o Understand the usage of PTU element;

o Keep in mind that the process starts always after a forecast both if this cause an issue
(CongestionPoint element) both no (optimization, Critical Peak Price use case, etc.) and this
forecast is referenced both in the FlexRequest both in the last Settlement message.
5.2.2 NOBEL GRID Demand Flexibility Framework

Through DSO-Aggregator interfaces definition in previous section, we highlight the term of demand
flexibility as a main factor for DSM strategies implementation. Therefore, we describe here the framework
for demand flexibility quantification as examined in NOBEL GRID project. This work is mainly reported in
WP10 and WP12 (as demand flexibility calculation is an internal process of NOBEL GRID DRMF framework)
through a brief summary is provided in this section.
Within NOBEL GRID framework, we examine the role of Aggregator as the entity responsible for Demand
Side management services. The Demand Side Aggregator gathers (aggregates) the flexibilities of
consumers to build Active Demand (AD) services: creates agreements with industrial, commercial,
institutional and residential electricity consumers to aggregate their capability to reduce energy and/or
shift loads on short notice in order to help utilities, grid and transmission operators to shave peak power
demands, balance intermittent power generation, increase security of supply, and avoid unnecessary
operations of CO2 intense power plants. The aggregators business model is built around receiving a
percentage of the value generated through avoided or shifted consumption. Aggregators manage its
clients consumption through, for example:
Switching aggregated electrical loads, for example pumps, fans, compressors, HVAC, cooling,
lightning, chilling
Shifting assembly line operations and/or utilizing times of equipment maintenance
Controlling electric vehicle charging process.
The next schema specifies the interaction of Aggregator with DSO stakeholders towards offering the
amount of demand flexibility available from customers of portfolio. Interfaces between Aggregator and
DSO interfaces were reported in previous section, and thus the focus is on the definition of NOBEL GRID
demand flexibility (FLEXIBILITY) offered by prosumers to Aggregators.
Figure 5.8 Demand Flexibility as a service for DSOs
The detailed analysis of NOBEL GRID demand flexibility framework is reported in D10.2 (D10.2: Holistic
Demand Flexibility Models), where we are presenting the overall framework towards the extraction of

demand flexibility profiles. This information is then provided to Aggregator DRFM for further exploitation
during Demand Response Services. Different types of demand flexibility profiles are reported in the project:
- Context-Aware Load Flexibility Profiling Framework, taking into account information related to
events from user behaviour to specific devices (smart appliances) , towards defining robust and
dynamic demand profiles, reflecting real-time demand flexibility as a function of multiple
parameters, such as time, device operational characteristics, environmental context/ conditions
and individual/group occupant comfort preferences. Therefore, the goal of this engine is to provide
estimation of demand flexibility from specific devices, taking into account low level information
from sensors and device actuators.
- In lack of low-level context information, high-level Demand Elasticity Profiles are developed,
reflecting temporal real-time demand elasticity as a function of multiple contextual
(environmental) and market (price and incentive schemes) variables. This engine, will enable the
implementation of price based strategies, exploiting that way the impact of energy prices to total
energy consumption
- As a third option, Load disaggregation methodology is developed in order to estimate and predict
composition of load categories. Total load will be divided into controllable and uncontrollable load,
taking into account low level/device information coming from building environment. Therefore,
even in cases where we are lacking detailed information, we can extract the amount of potential
demand flexibility offered by each customer of the portfolio. This information may be further
exploited by Aggregator, towards triggering the appropriate control strategies.
For the DSO case, where fast response is required towards grid stability and balancing support, Context-
Aware Demand Flexibility and Load Disaggregation are considered (Demand Elasticity Framework requires
triggering a price signal to customers and thus this service is not appropriate for fast response scenarios
requested by the DSO). Context-Aware Load Flexibility Framework, takes into account (real time and
historical) DER characteristics along with users preferences towards real-time extraction of the amount of
available demand flexibility offered by the devices that consist of the Smart Home Environment. The same
approach is considered for Load Disaggregation/ Customer Profiling case, where in lack of low level
information, we can extract the amount of potential demand flexibility taking into account total energy
consumption data. In both cases, we are extracting (in real time) the amount of potential demand
flexibility, to be further exploited by Aggregator (through automated Demand Response Strategies) on DSO
scenarios related to grid stability.
On the other hand, where no fast response is required for DSO grid management (e.g. Grid capacity
management), we are considering also Demand Elasticity Profiles through tariff based demand response
strategies implementation. In this case, the Aggregator is accepting the request from DSO and then
appropriate tariff based demand response strategies are triggered by DSM Aggregator, taking into the
selection process the responsiveness of prosumers on different tariff signals (expressed in terms of
Demand Elasticity Profiles).
We have presented the end to end framework for Demand-Side-Management (DSM) services
implementation towards congestion avoidance in the distribution grid, as a network service. The DSO
(through grid management services) directly interfaces with DRFM Aggregator, by triggering demand
management requests. Then, the Aggregator, by taking into account the amount of portfolio demand
flexibility, further triggers the appropriate automated or manual DR strategies at customers level.

5.2.3 Direct Interface between G3M and NOBEL GRID DER
In general, the G3M control center is exchanging measurement and control data with distributed
generators, storages and electrical appliances (DER) at the prosumers premises as specified by the NOBEL
GRID architecture. Information elements from DER to G3M include active and reactive power
measurements and status information. In addition, the SMM component at the prosumer may send grid
connection point voltage and frequency to the G3M. From G3M to generators and batteries, active and
reactive power setpoints can be transferred. In addition, there is information related to validation and
authentication services. Control commands to electrical appliances are not directly sent by G3M, but G3M
might request services from the DRFM for grid related usage of appliance flexibility.
According to the NOBEL GRID Architecture, this information interchange is routed through several
hardware and software components which basically perform protocol translation, but may also gather,
check, or pre-process the data. A possible manifestation of the communication chain is sketched in Figure
5.9. The components denominated D in the SMX trusted and non-trusted zone are Drivers which may or
may not be associated with an Application (App). If we take a look at the figure from the prosumer
perspective (at the right side), we find generators, batteries or appliances as information source or receiver
which is represented by either an PV or battery inverter or a SHIC (for electric appliances). The next
element in the chain are the trusted-zone drivers (D) in the SMX, followed by the SMX real-time data base
which serves as connector to non-trusted zone applications and drivers. Those may process data found at
the RTDB and send it through a customer-level router and local area network to the central control level,
where we find the DACF, the ESB and finally the G3M control center.
DER
meas.
G3M CIM ESB DACF SMX
61850 DER
DR & ancillary
D
services DER meas.

61850
gen./batt ctrl
Generator /
Inverter
D
MODBUS
battery
RTDB
Appl. flex.
DR & ancillary App(s)
D
oADR
requests
CIM Appl. ctrl
DRFM ESB DACF SH App SHIC Appliance
D
D
oADR 6LowPan
oADR Appl. Meas.
Non-trusted Trusted
Figure 5.9: Information chain between DER and G3M
For the functionality integration at the SMX non-trusted zone, OGEMA Apps may be used which essentially
provide the function of DER status data acquisition and provision towards the G3M. This is one of the
functions identified in D7.1 as low-level functions. OGEMA Apps may also be used for handling and
supervision of generator and battery operation.
For handling of the data flow from prosumer to G3M (measurements and status data), we conceive that
the following Apps will be needed at the OGEMA framework:
An OGEMA driver App which synchronizes the RTDB and OGEMA resources storing data from
different DER device drivers. This App is reading from the RTDB via the RBAC interface. The RTDB
data points are written by trusted-zone drivers connected to DER respectively.
At least one OGEMA App which reads DER data from OGEMA resources, eventually processing
them by recording, bundling, plausibilization, conversion of physical units, or any other preparation
needed before sending to the G3M. The results are written back to the RTDB, from which they are
read by the external IEC 61850 driver. This driver is providing an IEC 61580 compliant server which
the G3M can access. At the time of writing this deliverable, it is still unknown which data processing
will be needed in detail.
Eventual OGEMA Apps used for configuration of SHIC and inverter resources and setting up the
OGEMA resource tree. This function may be provided by an external software accessing the SMX
through the OGEMA REST interface, allowing to set up the resource tree from remote.
Appendix C contains a list of concrete data points provided by DER types with direct interface with G3M
and their representation in OGEMA. SHIC are included here since G3M may receive submetering data from
them. Data points consequently match the ones for the SMM; however, OGEMA resource types are
arranged at a different location in the resource tree. If, for example, a SHIC would be measuring the power
consumption of an air conditioning device, there would be a resource for the air conditioner. The active
power sensor actually implemented by the SHIC would then be a subresource of that air conditioner. In
contrast to that, the active power meter of the SMM would be a subresource of the buildings electricity
connection. Hence, both sensors provide active power data, but are accessed through a different path in
the OGEMA resource tree.
The OGEMA resource structure representing the information from the SMX grid connection point and from
SMX sub meters is designed as given in Figure 5.10. The sub metering data structure is essentially the same
for all devices providing metering. The device SampleBatteryInverter is given as a sample inverter. The
paths to the relevant information data points are given in Appendix C.
Figure 5.10 OGEMA Resource structure representing the grid connection point metering (sample
SHIC)
DER control is usually carried out by the DRFM, but it is currently planned to allow the G3M to directly
control PV generators and batteries. In addition, local control of generators and batteries is relevant for
some use cases. In order to avoid that control accesses from different actors have unwanted effects on
generators and batteries, again OGEMA Apps may be used. At this point, the following Apps may be needed
at the OGEMA framework:
An OGEMA driver App listening to incoming control messages from G3M, which are received by the
SMX over IEC 61850 and written to the RTDB by the IEC 61850 driver. The OGEMA driver App
would update internal OGEMA DER resources.
An OGEMA App processing changes of modelled DER setpoints, maybe applying checks and unit
translation, and writing the setpoints into corresponding DER OGEMA Resources, eventually taking
into account setpoint requests from other actors (e.g. by priority).
Eventual OGEMA Apps used for configuration of SHIC and inverter resources with respect to their
control capabilities. These Apps are corresponding to configuration apps mentioned above and and
thus may be combined with them. Also, they may again be replaced by an external software
accessing the SMX through the OGEMA REST interface.
It shall be noted that the complexity of the system of Apps and Drivers thus defined is not to be
underestimated, especially given that the system works on a platform which is performing a diversity of
other functions with limited processing resources. Also, implementation of configuration OGEMA Apps
might be arbitrarily complex depending on their capabilities and type of user interface. Complexity of the
complete IT system chain is very high also because the configuration of system elements will be different at
each prosumer site. For example, DER availability, type, and capabilities will not be the same at each
prosumer. Designing a software driven system for this is a most demanding task even if it is the very
domain of the OGEMA platform to support handling such complexity in variable environments.
5.3 GRID STABILITY BALANCING SUPPORT AND FAULT ANALYSIS
5.3.1 PV-inverter Management Functionalities

The interface between photovoltaic (PV) plants and the grid is implemented by PV inverters. Originally,
these devices were intended to provide active power to the grid. National regulations typically required
them to automatically disconnect (and reconnect) from the latter in cases of relevant under/over voltages
or frequencies. However, as PV penetration increases, namely at residential or small commercial level, it
has been realized that inverters are able to contribute to grid stability by means of intelligent control
strategies assisted by powerful communications abilities.
Besides its common task of automatically disconnection/reconnection under voltage and/or frequency
disturbances, the PV inverter will be able to support grid stability using several functionalities, intended to
compensate previously mentioned disturbances. These functionalities performance is related with storage
availability and are further described:
1. Assisting voltage control at local level: by absorbing or injecting reactive power, the PV inverter
can contribute to voltage control, even if the / ratio is relatively low in low voltage grids. The
flow of reactive power (from or to the inverter) is controlled by the displacement angle between
(grid) voltage and (PV) current, or, equivalently the power factor of the device. The amount of
active power injected in the grid (which increases voltage at that node) is controlled simultaneously
with reactive power, and its curtailed if required.
2. Implementing voltage ride through operation: under periods of voltage dips (due to faults in the
grid), the PV inverters are required to remain connected without generating overcurrents as
disconnection could cause cascading of other plants. Under these occurrences the PV inverter
increases its reactive power (if available or stored) in order to recover voltage level locally.
3. Contributing to frequency control: grid frequency is determined by active power balance
(generation vs. consumption). By increasing or curtailing its production, the PV inverters will assist
frequency control avoiding their disconnection in events of under/over frequencies (frequency
sensitive operation mode).
4. Compensating grid unbalances: besides its role in assisting grid stability, the PV inverter is also able
to address other issues when considering dispersed generation, namely the unbalance that arises
due to increased penetration of single phase sources. These were typically not considered when
designing low voltage grids and may not be addressed by conventional means as regulating tap
change transformers in substations. Negative consequences of unbalances are e.g. neutral
overcurrents, voltage asymmetries or increased losses. Grid unbalances will be compensated by PV
inverter through an adequate management of transferred power, which is independently injected
in each phase. This approach is represented in Figure 5.11, where a four-wire device is required for
this purpose.
Figure 5.11 Connection of the three-phase four-leg PV inverter, intended to compensate grid
unbalances.
5.3.2 Voltage Control in Distribution Networks with DGs

Todays advanced technology and the conversion of the distribution networks to active networks provide
the ability of distribution networks voltage control by the distributed generators (DGs). At present, most
DGs do not participate in voltage control. The voltage regulation of distribution systems is mainly provided
by on-load tap changer transformers installed in the substation. However, this control cannot react fast
enough to provide voltage control in emergency conditions.
In the context of future smart grids, modern converter-connected distributed generation should be able to
provide voltage control in medium-voltage distribution networks. The participation of DGs to voltage
support is achieved by controlling their active and reactive power output and was proposed in [84-86]
Moreover, the increased penetration of DGs provides the ability of decentralized control. The latter
eliminates the need for global communications and allows the parallel voltage regulation of the zones. In
addition, the voltage regulation problem is decomposed into several smaller sub-problems. Figure 5.12
presents schematically the partition of the network into control zones.
Figure 5.12 Decentralized control of power system
To achieve the decentralized control, the distribution network is divided into zones and the DGs provide
voltage regulation for the nodes inside their zone. The partition of the system into zones must ensure that
the DGs are able to provide voltage support to the nodes inside their zone and the change of voltage inside
a zone has a minimum effect to node voltage changes outside the zone. The division of power networks
into zones has been proposed in [87-90].
5.3.2.1 The decomposition method

The decomposition method is used to divide the system into weakly coupled zones. The method uses the
sensitivity matrix, which can be obtained from inversion of the Jacobian matrix. The sensitivity matrix given
in (5.1) describes a linear relationship between power variations of nodes and voltage changes. The power
variation is achieved by changing the active and reactive power generation of DGs.

[
] [ ]
[ ] [ ]
( )= ( ) (5.1)
[ ] [ ]
[ ] [ ]
( )
Each of the sub-matrices of sensitivity matrix has dimensions , where is the total number of
nodes of the network excluding the infinite node.

For the decomposition, the columns of the matrices
and , which do not correspond to nodes with
DGs, are deleted. Considering a certain node with initial voltage ,0 and target voltage , , the
adjustment of DGs power can be computed from the following equation.

, = ,0 + ( + ) (5.2)

=1
is the total number of DGs and , are the active and reactive power adjustment of DG ,
respectively.
The decomposition method breaks the system into sub-networks (zones). An efficacious partition has
been achieved when inside the formed sub-networks the coupling between the nodes are strong while the
coupling between the sub-networks is weak. In that way, the DGs provide voltage regulation only to the
nodes inside their zone.

Consider the matrix , which contains all the elements of the matrix larger than . The matrix

maintains only the strong couplings of the matrix and describes the influence range of each DG and

the formation of sub-networks. The same method can be applied for the matrix . The matrix is used

for the partition of the network, since the proposed voltage control algorithm is mainly based on reactive
power adjustment.
Based on the proposed decomposition method, it is possible to have nodes which are not included in any
zone or to have intersections among the ranges of influences of DGs. The Deep First Search is used to divide

the system into sub-networks based on the matrix .

The choice of parameter plays a key role in the partition of the network as it determines which couplings
are weak and has to be removed and which are strong enough to be hold. Therefore, the formation of the
zones depends on the value of parameter . The greater the value of , the greater the number of zones

with fewer nodes each zone. The value of obviously depends on the values of matrix . By selecting the
smallest value, the matrix remains unchanged and the formed zone contains all the nodes of the network.
On the other hand, by selecting the largest value only one zone is formed which contains two nodes.
Therefore, the choice of should be in the range of matrix elements and depends on the desired number of
zones and number of nodes that will be contained in each zone.
As the sensitivity of nodes voltage to the active and reactive power change of distributed generators
depends on network conditions, the formation of the zones must be adapted to continuously changes of
network conditions. Therefore, the zones are updated in specified time intervals.
5.3.2.2 Voltage control

In case of voltage violations in a zone (deviations higher than 5%) the optimum adjustment, to eliminate
voltage violations, of DGs inside the zone is computed. Using the sensitivity matrix, the linear programming
problem shown in (5.3)-(5.8) is solved in order to compute the power adjustment. All DGs are considered to
operate in Power Factor Control mode and be able to adjust their reactive power by changing their power
factor.
min (1 | | + 2 | |) (5.3)
=1
,
_ _
_, + ( + ) (5.4)

=1
, + , , (5.5)
, + , , (5.6)
, = tan(cos 1 ) , (5.7)
, = tan(cos 1 ) , (5.8)
and are the lower and upper voltage limits, respectively. , is the set of DGs and _ is an
index for the nodes inside a zone. , are the active and reactive power generation of DG , respectively.
, , , , , and , are the lower and upper bounds for reactive and active power generation
of DG . and are the lower and upper bounds of the controllable inverter power factor and
finally 1 and 2 are weights. Choosing 1 2 the problem is solved by firstly adjusting the reactive
power of DGs. If this procedure is infeasible, the active power is adjusted too.
Considering that the sensitivities which are used in (4) correspond to specific operation point, the
application of voltage regulation (Eq. (3)-(8)) may not eliminate the voltage limit violations. Therefore, the
voltage regulation is applied recursively, until there is no voltage limit violations or the linear problem is
infeasible due to constraints (5) and (6).
5.3.2.3 Case study

The distribution network of Sperchiada is used to examine the results of decentralized voltage regulation.
Sperchiada is a rural area in the centre of mainland Greece. It comprises a radial distribution network
including a HV/MV substation in Greece and the corresponding feeders P-21, P-22, P-23, P-24, P-25 and P-
26. The distribution lines are feeding rural loads and are characterized by high penetration of Photovoltaic
Systems (PVs).
The primary substation of Sperchiada has been designed and installed taking into account three main
aspects in the grid; feeding of a large MV customer (industrial load), connection of hydro generators in the
area and provision for the intersection of long distribution lines between two already existing primary
substations (Lamia, Karpenisi). However, this situation has drastically changed due to the disconnection of
the industrial load and the rapid increase of distributed generation, mainly PVs, leading to overvoltage at
the end users.
Due to high penetration of PVs, feeder P-25 is used to test the proposed decomposition method and
decentralized voltage control. The PVs are installed in 24 nodes and the total installed capacity is 13858
kW. The minimum and maximum load demand are 294 and 4122 kW, respectively. Figure 5.13 shows the
single-line diagram of feeder P-25.
Figure 5.13 Single-line diagram of feeder P-25
Table 5.1 shows the results of the network partition for different values of parameter .
Table 5.1 Partition results of P-25 for different values of parameter

Number of Percentage of Max number of
zones nodes in zone nodes in a zone
(%)
0.00008 1 100 100
0.40 2 89 85
0.45 3 84 41
0.50 5 72 37
0.55 4 64 36
0.60 3 47 26

The values 0.00008 in Table 5.1 is the lower value of the sensitivity matrix . The value 0.5 is chosen for
the partition as it leads to the most satisfactory results. The network is divided into five zones, which
contain the 72% of the total number of nodes. For this value, the maximum number of nodes in a zone is
37. Figure 5.14 presents the results of the partition for the chosen value of the parameter .
Figure 5.14 Partition results of P-25 for =0.5
For the application of decentralized voltage control the following assumptions have been made
The DGs are able to vary their power factor in the range of 0.707 inductive/capacitive.
The DGs have unit power factor before voltage control
Considering that all the DGs are PVs, the upper bound of active power generation is equal to their
production (, = ). Therefore, the adjustment of active power could be effective only for
over-voltage problems. In addition, PVs are able to curtail their active power in order to eliminate
voltage violations
The upper and lower voltage limits are equal to 0.95 p.u. and 1.05 p.u, respectively.
The scenario of maximum generation minimum demand is considered.
Figure 5.15 shows node voltage profile before and after centralized voltage control. Table 5.2 presents the
results of the linear programming problem for the centralized voltage control. The second and third column
of Table II present the adjustment of reactive power and the power factor for the DGs that participate in
voltage control, respectively.
1.08
1.06
1.04
1.02
Voltage (p.u.)
1
After control
Before control
0.98 Upper limit
Lower limit
0.96
0.94
0 10 20 30 40 50 60 70 80 90 100
Nodes
Figure 5.15 Voltage profile before and after centralized voltage control
Table 5.2 Results of centralized voltage control

DG (Node) Qadj (MVAr) PF
43 -2.3068 0.763 (inductive)
90 -1.0278 0.959 (inductive)
Total -3.3346 -
Figure 5.16 and Table 5.3 present node voltage profile and the results of linear programming problem for
the decentralized voltage control, respectively.
1.08
1.06
1.04
1.02
Voltage (p.u.)
After control
0.98
Before control
Upper limit
0.96 Lower limit
0.94
0 10 20 30 40 50 60 70 80 90 100
Nodes
Figure 5.16 Voltage profile before and after decentralized voltage control
Table 5.3 Results of decentralized voltage control

DG (Node) Zone Qadj (MVAr) PF
43 1 -2.6829 0.712 (inductive)
90 5 -0.7945 0.975 (inductive)
Total - -3.4774 -
By comparing Table 5.2 and Table 5.3, it is observed that the total reactive adjustment of centralized and
decentralized control are very close, proving the effectiveness of decentralized control as well as the
network partition.
Furthermore, the scenario that the DGs are able to vary their power factor in the range of 0.95 inductive
/capacitive is examined. By decreasing the capability of reactive power adjustment, the DGs are forced to
curtail active power in order to eliminate voltage limit violations. The voltage profile and the results of
decentralized voltage control are shown in Figure 5.17 and Table 5.4, respectively.
1.08
1.06
1.04
1.02
Voltage (p.u.)
After control
0.98 Before control
Upper limit
Lower limit
0.96
0.94
0 10 20 30 40 50 60 70 80 90 100
Nodes
Figure 5.17 Voltage profile before and after decentralized voltage control with active power
adjustment
Table 5.4 Results of decentralized voltage control

DG (Node) Zone Qadj (MVAr) Padj (MW) PF
29 1 -0.2465 0 0.95 (inductive)
33 1 -0.0493 0 0.95 (inductive)
35 1 -0.2465 0 0.95 (inductive)
40 1 -0.0493 0 0.95 (inductive)
42 1 -0.6281 0 0.95 (inductive)
43 1 -0.6433 -0.55 0.95 (inductive)
92 1 -0.0493 0 0.95 (inductive)
93 1 -0.099 0 0.95 (inductive)
97 1 -0.0493 0 0.95 (inductive)
99 1 -0.0493 0 0.95 (inductive)
100 1 -0.0247 0 0.95 (inductive)
90 5 -0.9651 0 0.964 (inductive)
Total - -3.0106 -0.55 -
As can be seen form Table 5.3, despite the fact that the DGs in zone 1 consume the maximum reactive
power that are able to, the voltage violations are not eliminated. Therefore, the DG at node 43 decrease
the active power generation.
5.3.2.4 Conclusions
The proposed method investigate the participation of the DGs in voltage regulation. The partition of the
network is based on decomposition method. The aim of the partition is to form zones with weak
couplings between them while the couplings inside the zones are strong. The network division eliminates
the need for global communications and allows the parallel voltage regulation of the zones. Furthermore,
the linear programming problem for eliminating voltage violations is decomposed into several smaller sub-
problems. The results of the case study prove the efficacy of the proposed method.
5.3.3 Analysis of outage and the black-start process

5.3.3.1 Introduction
Last decades, extreme weather events have led to widespread blackouts. Considering that the frequency
and severity of such events is expected to increase in the future as a direct impact of climate change, it is
becoming more and more important to develop techniques for mitigating the impact of such events and
get things back to normal operation as quickly as possible after the disaster. Consequently, a significant
aspect is power system restoration.
Microgrids (MG) have been introduced [92] as an effective way to mitigate the problem of large scale
integration of distributed generation systems. The MGs have great flexibility so that they can operate ether
connected to the distribution grid or work isolated when grid fault occurs. The restoration process after a
partial or complete blackout relies on black start resources. In this context, micro-grid Black Start (BS) is a
very innovative aspect that can be used in order to fully profit from the potentialities of dispersed micro-
sources. Up to now, functionalities for operation restoration are available only for conventional power
systems. Following a top down approach the restoration procedure is based on division of the restoration
procedure into the following stages restart, reintegration, load pick-up and interconnection phases [93].
Restoration strategies of the MG can be divided into parallel (or bottom-up) strategies and series
strategies. While serial strategies - which are similar with the conventional top-down approaches - can
simplify the complexity of the design process and improve the stability of the system, parallel operations
lead to shorter restoration times.
According to national regulations regarding the DER inverters, reconnection of the DER units should be
done after the clearance of grid failure once the distribution system is energized and rated voltages and
frequency have been restored. However, active management strategies in microgrids could lead to
distributed control in island mode, which may not require the disconnection of microgrid DERs in a major
disturbance [94]. In blackouts, distribution system downtime can be reduced by providing a fast blackstart
of local generation resources in island operation. A microgrid that has remained fully functional and lies in
the restoration path can be started in island mode to supply its loads.
The emerging microgrids embedded in distribution systems allow distribution systems to recover faster in
the event of an outage. The feasibility of the new concepts islanded operation of small distribution
systems and black start functionalities are tested through simulations, being the results described and
discussed in this study. The focus of our analysis is targeted exclusively to standalone microgrids that have
no interconnection with a stiff grid. That could be the case of remote islands which should rely only on
autonomous power solutions. Grid forming units, such us diesel generators and grid forming inverters, that
carry the main duty of stabilizing the MG is the center of attention of this investigation.
5.3.3.2 Modeling and Control
Several Generator technologies were considered to coexist in the MG and the corresponding dynamic
models were derived. The generators are an essential part of the MG, their dynamic models used in the
simulation platforms are briefly described, as well as the analysis of their control structure. As MG are
inverter dominated networks, the use of adequate control strategies in the solid state converters are
crucial for MG operation. A detailed overview is given about the two most common types of generators:
Synchronous generator: A model of an electrically excited synchronous generator is presented.
This model can be employed to represent the characteristics of a diesel generator, which prevail in
real microgrids, such as islanded grids.
Inverter-based generators: The topology selected for the grid inverter is the standard 2-level
voltage source inverter (VSI) with a capacitor fed DC-ling as input.
Synchronous generator
The synchronous generator model is shown in Figure 5.18. There are two control loops present: The
automatic voltage regulator (AVR) and the speed governor. Both control loops implement what is known as
droop characteristics for active and reactive power respectively. More specifically, for the speed governor,
the droop characteristic is implemented as a setpoint for the mechanical torque (which is identical to the
setpoint for active power) while for the AVR the voltage deference from the reference input is used to
control the exciter field.
Pset Qset
Governor m* AVR vSG

vf vSG*
iSG vSG Net
m
Tm Prime
V
SG A
Mover
Figure 5.18 Overview of the synchronous generator structure
The synchronous generator consists of two main components: the armature and field windings. The model
takes into account the dynamics of the stator, field, and damper windings. The equivalent circuit of the
model is represented in the rotor reference frame (dq-frame). The machine model developed in this section
is oriented to the structure and operation of a salient-pole machine [95]. Figure 5.19 represents the
equivalent circuit of the machine.
qe de
Rs Ll Rs Ll
+ - + -
id Lfd
L1d iq L1q L2q
Rfd
+ -
Dd /dt Lad
Dq /dt
Ed
+ -
Eq Laq
R1d R1q R2q
+ -
Efd
Figure 5.19 Equivalent circuit of synchronous generator in dq-frame.
Variable Description
Ed,q Induced terminal voltage

d,q Flux linkage
e Rotor angular velocity
L l & Rs Self inductance & resistance of stator winding
Ladq Mutual inductance (DQ) between stator & rotor windings
L1dq , L2dq Self inductances of DQ damper circuits
Lfd & Rfd Self inductance & resistance of rotor circuit
Ldq DQ-axis reactance
The Expressions for the Equation of Motion is giver, in terms of the inertia constant, as:

{ }, =
+ = 2

To properly simulate a machine we need to have not only a representative model but also accurate
parameters for the model. These parameters cannot be directly measured from the machine terminals,
they are obtained through experimental test and they are available in the manufacturers datasheets.
Speed Governor and AVR: The design is based on the assumption that the excitation circuit and the prime
mover can be represented by first-order models. The automatic voltage regulator is modeled according to a
modified AC5A excitation type model from the IEEE 421.5 standard [96].
The plant transfer function and the PID controller can be emulated with the following way:
1 2 + +
() = , () =
(1 + )(1 + )
Where, tg, te is the generator and exciter open-circuit time-constants. kd, kp, ki are the gains of the
controller.
The model for the machine is also following the first order representation:
1
() =
(1 + , )(1 + )
Where, tg,m and tt are the time constants for the machine and the prime mover respectively
The design methodology uses the known model of the plant and determines the controller settings that will
force the closed-loop poles to designer-selected, desirable locations The closed loop system thus has the
desired transient response [97].
Grid forming inverter
There are mainly two different operation modes that an inverter can operate. The inverter model is derived
according to the control strategy followed:
P-Q control mode, where the converter is working as a current source injecting/absorbing power
to/from the grid. In this way, the main objective is to control the power that have to be
interchanged with the grid.
Grid forming mode or V-F mode, where the inverter controls the output voltage and frequency,
trying to form an autonomous grid. In this case, the energy flow coming out from the DC-part of
the inverter is always determined through the load demand.
The model of the islanded system connected through an LC(L) filter is depicted in Figure 5.20 where we see
the DC bus as a DC source, the DC/AC inverter, the LC(L) filter and the Load. The grid forming inverter
requires a DC input source that provides the energy required from the grid.
L1 R1 L2 R2
Vc
DC_BUS
3 Leg + C
LLoad RLoad
f
Bridge
3 Phase LCL Rf
Filter
3 Phase
Load
Figure 5.20 Grid forming inverter model.

The objective of the inverters control algorithm is to regulate the amplitude voltage and frequency at its
output and provide the necessary active and reactive power to the rest of the microgrid. Figure 5.21
illustrates the control structure for the grid forming VSC system, the control is implemented in a dq-frame.
A cascaded PI controller scheme is adopted in order to accomplish the current and voltage regulations of
the VSC. Coupling between Vsq and Vsq as well as id and iq is eliminated by decoupling feed-forward
compensation. Figure 5.21 also shows that iLd and iLq are compensated by means of another feed-forward
compensation strategy. Therefore, the compensated system is capable to perform similarly under all load
conditions. The controller design of the grid forming VSC was in detail explained in chapter 3 of the
deliverable D7.4 [98].
Figure 5.21 Control block diagram of the grid-forming VSC system [99].
Greed forming inverter in parallel mode: The control principle of a VSI emulates the behavior of a
synchronous machine. Thus it is possible to control voltage and frequency on the AC system by means of
inverter control. Frequency variation in the MG provides an adequate way to define power sharing among
several VSI, since it is related with system frequency. Figure 5.22 presents the form of the P-f
characteristics that was designed and incorporated into the general control strategy of the inverter.
However, in the case of parallel operating inverters, it is advantageous to couple also reactive power with
voltage amplitude to optimize reactive power sharing, Figure 5.23. By introducing frequency and voltage
amplitude deviation and coupling them with active and reactive power respectively, the following equation
can be formed:
0 = ( 0 )
0 = ( 0 )
m ax Vm ax
0 V0
m in
Vm in
P m in P ou t P m ax
Q m in Q ou t Q m ax
Figure 5.22 Frequency vs power droop Figure 5.23 Voltage vs reactive power droop
characteristic. characteristic
In parallel operation when the grid forming inverter is required to connect to an already up and running
micro-grid, the micro-grid voltage is required. Therefore, a phase-locked-loop (PLL) is used to facilitate the
synchronisation. As shown in Figure 5.24 the micro-grid voltage VSG is used to provide the synchronization
signals to the inverter controls. The objective is to control the output voltage and angle of the inverter to
match the micro-grid voltage and angle, so that a smooth synchronisation can take place. The Low pass
Filter (LPF) at the point after the calculation of the active and reactive power is applied to reduce the
dynamic behaviour of the power flux controller to levels much smaller than the one of the voltage control
loop of the inverter. A typical range of cut-off frequencies for the filter are = 5 10.
vSG *
LPF
f n om
f
vC P f*
P, Q
IL2 Computation
V vd *
Q
Vn om Vq *= 0
Figure 5.24 Droop control with synchronization scheme
5.3.3.3 Microgrid Black Start
In conventional stiff grid structures, black start is defined as the ability to boot up the system without
importing external power. In this way, also in micro-grids, black start means that the micro-grid has the
ability to start operating on its own, without external support. Communication between generators is
essential during the black start process: not only if more than one generators are involved to generate the
connecting sequence, as well as to control the loads to be connected.
If a microgrid is located near to a central generator that does not have a black start capability, it has the
potential to provide power necessary for starting the plant and thus accelerate the restoration process. The
unit that starts up the booting process is called master generator. The criteria for the appointment of a
generator unit as the master unit in a black start process are listed below:
Short start-up time for fast restoration.

Self-starting with self-powered auxiliary systems.
Able to handle the heavy inrush load current at the starting of re-
energizing process due to prior de-energized conditions of the loads.
In case of distribution network, it is common for the diesel generator to overtake the black start process as
a master generator and the inverters to follow. Battery based distributed generators can also be employed
as master generators. The appointment of the master generator depends also on its rated power, so that
significant number of loads can be connecter to the network as soon as possible. The synchronization
technique for the grid forming inverters is similar to the one that modern synchronous generators employ
to connect to the grid. More specifically, the following measures should be taken prior to synchronization:
The produced electrical frequency of the generator is equal to the one of the grid. That means
that the generators rotating speed should be adjusted to match the grid frequency.
Synchronous generator and grid have the same phase sequence. That could arise a problem
right after the generators installation and during the first grid synchronization attempt.
Output voltage magnitude of the generator matches the grid voltage. If the magnitudes do not
match, the generator will be connected and work in over-excited or under-excited mode,
depending on which voltage is higher.
Not only the magnitude but also the voltage phase angle should match, this can be supervised
by observing the zero crossing points of both voltages.
If not all four synchronizing conditions are met, equal load sharing cannot be achieved and a transient
process will occur that will lead to a massive reactive power circulation between generator and grid.
The black start procedure at least one grid forming unit is needed during all operational modes. In this way,
the SG and the inverter are operated in grid forming mode. The following sequence of actions is the
fundamental step that should be carried out in order to restore after a blackout event.
Disconnection of all loads, in order to avoid large transients when energizing the network.
Building the network. Grid forming generators are employed in VF mode so as to re-energize the
micro-grid.
Connecting controllable loads. The amount of loads to be connected should take in to account the
available power and storage capacity of the generators in order to avoid large voltage and
frequency deviations.
The chain of command between generators and loads is also crucial for the black-start process but exceeds
the focus of this work.
5.3.3.4 Black Start test scenarios
A low voltage hybrid micro-grid (Figure 5.25) is considered, in order to evaluate the restoration time and the
ability of the grid forming inverter to be appointed as master generator. Two cases are first considered. The
D8.1 Design and specifications of stable and secure distribution grids 118
generator and the gird forming inverter overtake the start-up process. The grid forming inverter is sized
equally to the generator (S = 20 kV A) to ease drawing conclusions on their behavior. Tipical values for the
parameters of the generator have been used while the following values are used for the filter of the
inverter: L1 = 2.3mH, R1 = R 2 = R f = 0.01, L2 = 0.93mH, Cf = 10 F and VDC = 730V.
iSG vSG
m
V
Prime SG A
Mover
Load
vDC iDG vDG

L1+R1 L2+R2 Load
V V
A
Cf+Rf
Figure 5.25 Microgrid configuration for black start scenarios.
The scope of these first scenarios is to investigate the fundamental difference in the restoration time of the
grid forming units. The test scenario proceeds as following: The master generator is initially booted until
steady state is reached. A step load change that equals half of the nominal load of the generator follows.
Once the system reaches its new steady state point, a second step load change occurs, so that in the end of
the test scenario the generators current reaches its nominal value.
Figure 5.26 Black start test scenario with (right) grid forming inverter and (left) with SG.
Figure 5.26 shows the results of the two test scenarios, on the right side are depicted the response of the
grid forming inverter while on the left side is shown the response of SG. It is apparent that the time scale
difference in both processes is significant. Since the inverter is a static electric device without moving parts,
it can follow any transient change in a matter of milliseconds. Inertia of rotor mass resists in any
instantaneous speed variation making the synchronous generator considerable slower to boot up. It is
worth to mention that the steady state frequency drop shown in Figure 5.26, is due to the implementation
of the droop characteristic and depend on the preselected droop gains. The droop gains kP, kQ are defined
through the maximum frequency & power or voltage & reactive power deviation. Here, for the needs of the
simulations the following values have been chosen: kP=0.01, kQ=0.02.
Figure 5.27 Active power sharing for grid forming inverter synchronisation (P/Sn).
In the following test scenario the synchronization of grid forming units is presented. The synchronous
generator is the first generator to be connected to the grid, providing the power required from the initial
load. Before the connection of the inverter the SG it delivers active power close to its nominal power. Once
the synchronization of the grid forming inverter is completed, the load is effectively shared between both
generators and the microgrid has reached a new steady state point. Figure 5.27, shows the active power
sharing between generators, the inverter power smoothly equalizing the outputs of the SG. The active and
reactive power outputs of each distributed generator in the following figures are scaled with their rated
power. The duration of matching the active power at the output of the generators is dependent on the
droop gains selected as well as the selection of LPF filter at the droop control loop.
Figure 5.28 Matlab/Simulink microgrid model for the black start scenarios.
5.4 DEVELOPMENT OF VERIFICATION AND CERTIFICATION SERVICES
5.4.1 Necessity for Verification and Certification of Smart Grid Services

Demand response as well as ancillary services fromprosumer ownedinverters are very interesting to
DNOs as tools to help manage their grid. These tools can be used to reduce losses and to keep power
quality, like voltage and harmonics, within the required restrictions. However, in order to be a real
alternative to traditional measures like enforcing the grid, these tools should be cost effective and
dependable.
The main responsibility of a network operator is to guard the integrity of its grid. So for a DNO to actually
depend on demand response from consumers that are in principle free to do act as they desire, in order to
avoid congestions requires a lot of confidence and trust.
Establishing standards and the associated verification and certification thus will likely play a major role in
creating the necessary confidence that demand response and demand side ancillary services are cost
efficient and trustworthy alternatives to traditional measures.
Besides using flexibility for multiple causes and multiple stakeholders (DNO, TSO and BRP), this
trustworthiness is on one of the main reasons for the introduction of a new market role: the role of
Aggregator. An Aggregator is a commercial party responsible for the actual delivery of flexibility from
demand response to other stakeholders. Individual prosumers themselves will not be likely to be able or
willing to bear this responsibility. Being commercial, forces aggregators to entice prosumers to participate
in demand response or be out of business.
The smart grid verification services focus thus predominately on the standardization and quality of service
between an aggregator and DNO. Although the other stakeholders that are able to purchase the services of
the aggregator, their existence is an important requirement for the success of the aggregator business
model.
5.4.2 Levels and Structure of Smart Grid Verification Services

Services from demand response and for power quality control require verification of multiple levels. These
levels range from the quality of the organization to the proper use of the right protocols and standards for
communication. The different layers that can be distinguished are described below and summarized in
Figure 5.29, using the SGAM framework as a reference.
Figure 5.29 Verification of smart grid communication layers
Because there is no generally accepted smart grid standard and because as one of the founders DNV GL has
extensive knowledge, the USEF framework has been chosen as a reference1.
Capability assessment is the assessment of the capabilities of a provider of (ancillary or flexibility) services
1 USEF stands for Universal Smart Energy Framework and is published on http://www.USEF.eu and
https://github.com/USEF-Foundation/ri.usef.energy.
with the aim to assure the service/flexibility is guaranteed to be delivered. This may include assessment of
technical capabilities, organizational capabilities (a.o. internal processes, ISO, etc.), financial health, etc. It
may be part of a qualification and/or selection process for suppliers of (ancillary or flex) services.
Performance assessment is assessing the (potential) performance of a portfolio of DR and/or storage. This
can be performed both as an audit (e.g. test if the portfolio or individual device is meeting the
requirements), or as a check of the actual performance with in practice (e.g. after a large incident that
affected the grid frequency, the response of FCR suppliers is monitored and checked if they performed
according to the requirements). Audit services for ancillary services to TSOs already are common.
Performance assessment can be expanded towards aggregated offers of flexibility. This requires more
advanced methods and ways to extrapolate performance samples of devices to the whole flexibility
offering.
Process compliancy is about the verification that relevant stakeholders comply with prescribed market
processes. This is especially important for pilots and in the initial stages of market development. As the
aggregator market matures, the added value of this kind of verification (between stakeholders) will decline.
Protocol assessment is the verification of the ICT messages and sequences of messages between
stakeholders exchanging (flexibility) services. Protocol certification on device level (like OpenADR) is
currently out of scope of USEF. Verifying if equipment complies with this kind of communication will likely
be too expensive for small (household) devices. Certification on device level thus might be similar organized
as CE certification (i.e. self-certification).
Digital twin verification is the verification that a digital copy of a device, services or (Demand response)
product (including intelligence and logic) has the same behaviour as the original (for a specified range of
applications).
The value of these digital copies is that they can be used in system and scenario studies, the assessment of
future grid stability and business cases for BRPs and aggregators. With digital twins unforeseen interference
between devices or agents can be found in advance and the flexibility of a certain portfolio of smart devices
under a certain demand response regime/service can be assessed.
5.4.3 Smart Grid Verification Services Within NOBEL GRID

Within NOBEL GRID verification services have been developed for process verification and protocol
verification. For each of these verification services two sets of tests have been developed, which are shown
in Figure 5.30.
Figure 5.30 Layers for verifying Process and IT communication
In the next paragraphs this is elaborated upon for both the process communication as the IT (protocol)
verification.
5.4.4 Process Verification

The process verification service follows an audit approach where the USEF is used as a reference to audit
against. This assessment can be performed during all stages of implementation (design, develop, test,
implement) and also includes advice for a pilot to become USEF compliant in case of non-conformance
issues.
The audit is based on the 2015 USEF Specifications [91]. Because of Privacy & Security (Chapter 10 of
USEF) of this version concerns principles, rather than (normative) specifications. Privacy & Security is
omitted from the audit. The Grid Operations specifications (Chapter 7 of the USEF specification) were still
under development, which is why they are omitted from the audit as well.
The audit consists of 110 questions divided over five sections:
Market model concerns specifications related to the USEF Market-based control mechanism
(Chapter 5)
Process concerns specifications related to the processes defined in the Market-based control
mechanism (Chapter 5)
Interaction concerns specifications related to the interactions between roles in the Market-based
control mechanism (Chapter 5)
Contractual and technological specifications concern requirements for participating in USEF
(Appendix I)
The division of questions between the topics is indicated in Table 5.5. Table 5.6 gives a small sample of the
results of an audit for a Dutch smart grid pilot.
Table 5.5 division of questions among the sections in the process verification audit.
Sections # Questions
Market model 32 (29%)
Process 22 (20%)
Interaction 44 (40%)
Contractual 7 (6%)
Technological 5 (5%)
Total 110
Table 5.6 a sample of results of a preliminary process audit for the process between an aggregator
and a DSO in an existing Dutch Smart grid pilot.
5.4.5 Protocol Assessment
For protocol verification several tests have been developed, both on the exchange of data level as on the
communication behaviour level. These tests complement the already available smart grid protocol
verification as is shown in Figure 5.31.
Figure 5.31 Overview of smart meter and smart grid protocols for which verification services already
exists
Within NOBEL GRID the verification services have been developed for the protocols used between the
different stakeholders in the USEF on two levels: for individual messages (message test) as well as for
message sequences (i.e. process test).
The System under test, i.e. the system that has to be verified is connected to a Test system (in the form
of a web server) that emulates the counterpart. The system checks the reaction of the system under test
under various circumstances, using both positive as negative (wrong) messages. The structure of the test
is shown in Figure 5.32.
Figure 5.32 Structure of the message and process (conversation) test. The lowest row indicates
which system initiates the test
The test system has the form of a web server that can be approached by the system under test. The user
can select a scenario to be tested and perform the test, see Figure 5.33 and Figure 5.34.
After the test is performed the results are shown. Figure 5.35 shows an example of how these results are
presented.
Figure 5.33 Scenario selection.
Figure 5.34 overview of the test scenarios for the aggregators system.
Figure 5.35 example of test results (in this case the system of a BRP).
5.5 IDENTIFICATION OF SUB-OPTIMAL SITUATIONS
5.5.1 Introduction
A key aspect in the management of the distribution grid is to know the status of the grid at all times. An
added-value to the knowledge of this state is to have information about their evolution. In NOBEL GRID we
are going to evaluate the use of clustering techniques to process and analyse the grid status, towards the
identification of sub-optimal situation with the aim of notifying the operator and eventually perform
automatic control.
Cluster analysis divides data into groups (clusters) that are meaningful, useful, or both. If meaningful groups
are the goal, then the clusters should capture the natural structure of the data. In some cases, however,
cluster analysis is only a useful starting point for other purposes, such as data summarization. Whether for
understanding or utility, cluster analysis has long played an important role in a wide variety of fields and it
is a goal of NOBEL GRID to evaluate their use for enhancing distribution grid monitoring.
The system status is defined by a vector x that characterizes its behaviour. It is therefore possible to
identify different types of situations in the distribution grid from the knowledge of the state variables.
For each of these situations, the distribution grid manager must perform a series of actions that enable the
distribution grid to perform properly, especially in those cases occurring or expected to be produced
undesirable situations.
The status of the distribution grid is characterized by different electrical parameters associated with the
various elements that make up the grid so its interpretation, even for an expert, requires a careful analysis.
This implies that the operation of the grid by those responsible for the system can become very tedious and
complex. One way to solve this problem is to have automated methods of analysis and identification of
situations. These methods determine the status of the distribution grid and communicate those responsible
for their management so that they take the necessary actions that allow optimal operation. It is possible
that the control system itself performs the necessary actions to achieve optimum grid performance.
The purpose of the Strategic Manager of the distribution grid is the identification of the different situations
that occur on the grid and the activation of strategies that respond to these situations.
Unwanted
Future Situations
Situations
by Probability
by Risk
Future Evolution
Unwanted
Future Situations
situations
S1 0.1 0.2 0.1 0.3 0.2 0.1
S2 0.2 0.1 0.3 0.1 0.1 0.2
S3 0.1 0.1 0.2 0.1 0.2 0.3
S4 0.3 0.2 0.1 0.1 0.2 0.1
S5 0.1 0.1 0.4 0.1 0.1 0.2
0.1 0.1 0.1 0.1 0.5 0.1

S6
S1 S2 S3 S4 S5 S6
Probability & Risk
S1
Strategy
Situation 1
101
X S3
S6 Strategy
* Situation 2
025
Vector Status
X S5
Situation 3
Strategy
200
Strategy
Situation 4
151
S2 Strategy
S4 Situation 5
305
Strategy
Situation 6
222
Clasification Strategy Selection Control System
Figure 5.36 Situations Management
5.5.2 Strategic Manager
The Strategic Manager is a component of G3M that is continuously observing the status of the grid,
identifying the situation where the grid is and activating control strategies to maintain optimal grid
performance.
The following figure shows a simplified diagram of the architecture of the Strategic Manager. It shows the
following items:
The data defining the status of the distribution grid.
The Strategic Manager is responsible for analysing the status information, identify situations, store
them in the database and send action orders to the control system.
Database where the information of the situations and strategies associated with these situations is
stored.
The user interface enables system administrators to manage information of situations and
strategies.
The control system that receives the information of the strategies to be executed.
User Interface
Control Strategic Situations/

Data Base
System Management Strategies
x
System
status
Figure 5.37 G3M Strategic Manager
5.5.3 Initial Identification of Situations

In normal operation, the Strategic Manager identifies situations comparing the current state of the grid
with a library of situations, stored in the database. This library can be created previously and take as input
the historic of situations.
One way to create the library is that experts on the distribution grid define situations which may be of
interest. To do this, it has to be specified the vector state defining each of the situations.
This way of defining situations requires a very deep knowledge and laborious work. A simpler option is to
use historical data available in the Control System for an initial classification of situations and create the
situations library by using big data techniques over the historical set of records.
Expert
Non Supervised Historical
Situations Data Base
Classifier Data
Definition
Expert User
Figure 5.38 Initial Classification
This initial classification can be performed using different classification techniques. One of the ways to
classify the historical data is the use of unsupervised methods since it is not necessary to have previous
knowledge of the grid. The Strategic Manager analyses the historical data and through different operations
can classify the data into groups. Each group will define a situation. An unsupervised classification method
that provides good results is based on K-means algorithm.
* *
* * * *
* *
* * * * * S1* S1
* * * *
* *
* *
* * * *
* * * * * *
S3 S3
* S6 * S6
* * * * * *
* * * *
* * * * * *
* * * * S5 S5
* * * * * *
* * * *
* *
* * * *
* * * * * * * *
* *
* * * * * *
* * * *
* S2 * S2
* * S4 * * S4
* *
* * * *
* * * * * *
* *
Historic Data Clustering Situations
Figure 5.39 Cluster Creation
K-means clustering aims to partition n observations into k clusters in which each observation belongs to the
cluster with the nearest mean, serving as a prototype of the cluster.
In the normal operation of the Strategic Manager is possible to define new situations from data received as
detailed in section 5.5.5.
5.5.4 Strategies
In order to optimize the operation of the distribution grid, each situation must have an associated strategy
that will be applied by the Manager of the distribution grid when the situation occurs. A strategy consists of
a series of actions on the control system. These actions include direct orders on the grid components,
modification of control parameters, generation of incidents or simply no action.
Situation 1 Strategy 101
Figure 5.40 Strategy Selection
The definition of the most appropriate strategies to each situation requires the participation of experts on
the distribution grid and support of the various tools offered by G3M.
5.5.5 New Situations and Learning

The Strategic Manager, every time classifies a situation, indicates the quality of the classification. A "poor"
rating indicates that possibly the situation is new and is not in the library of situations. The system warns
those responsible for the system that the situation may be new to define it as such if they consider it.
X S7
S1
* S1
S3 S3
S6 S6
S5 S5
S2 S2
S4 S4
Figure 5.41 New Situation
Creating new situations is simple using the tools offered by the Strategic Manager. It is sufficient to perform
the selection of those time instants in which it is considered to have occurred the new situation. With this
selection the Strategic Manager recalculates new situations.
The Strategic Manager is able to learn as it receives new data. With each new data and after classifying the
situation, recalculates the state vector that characterizes it. This allows that the definition of the situations
is dynamic and continuously adapted.
5.5.6 User Interface

The Strategic Manager provides graphical components that allow operators to manage situations and
strategies. These components are integrated into the User Interface of the G3M.
The User Interface allows the display of graphical and alphanumerical information associated with
situations and strategies. At all times the User Interface displays the status of the grid and the Active
situation.
The user interface allows the graphical representation of the situations on a map of the facility and an
electrical schematic thereof. The different variables are shown by symbols and colours so that the
interpretation is simple. A very useful tool is the comparison of situations on the map or the electrical
schematic. This allows you to see through symbols and colours differences between the situations it is not
always easy with multidimensional situations.
S3 S6 S3-S6
- =
Figure 5.42 Situations Comparison
The user interface displays a historical record which shows each of the detected situations and activation of
their strategies.
5.5.7 Future Situations

The Strategic Manager shows the active situation and the past situations produced. It also allows viewing
the forecasted situations and the risk that occur unwanted situations. This allows the grid managers for
taking early action to achieve optimal grid performance.
The expected situations are those most likely to be reached from the active situation. The user interface
displays a list sorted according to the likelihood of these situations.
Evolution Risk
Situation S3 0.4 Situation S2 0.1 Cost 30K

Situation S6 0.2 Situation S6 0.2 Cost 10K
S1 0.1 0.2 0.1 0.3 0.2 0.1

Situation 1 Cost
S2 0.2 0.1 0.3 0.1 0.1 0.2

Situation 2 Cost
S3 0.1 0.1 0.2 0.1 0.2 0.3

Situation 3 Cost
Situation S4 0.3 0.2 0.1 0.1 0.2 0.1 Situation 4 Cost
S5 0.1 0.1 0.4 0.1 0.1 0.2 Situation 5 Cost
S6 0.1 0.1 0.1 0.1 0.5 0.1 Situation 6 Cost
S1 S2 S3 S4 S5 S6
Probabilities Cost
Figure 5.43 Evolution and Risk
Unwanted situations are those that are conflictive and that has an associated cost (damage) if they occur.
The Strategic Manager assesses the likelihood that any of these situations and calculates the associated
risk. The user interface displays a sorted list of situations with increased risk, indicating the associated cost
and probability.
5.5.8 Complementary Visualization

The user interface allows visualization of the data defining each of the situations and the data defining the
active situation of the grid. These data are the corresponding to the state vector x that characterizes the
grid. These electric magnitudes are easily interpretable by grid managers. However when the grid is
complex and must manage many variables, it is difficult an interpretation of the actual status.
To facilitate this interpretation the user interface displays information about the Principal Components. For
this, the Strategic Manager calculates the Principal Components of the data and shows these components
for each situation and the actual state vector. This information supplements the information provided by
the state vector and facilitates user interpretation.
Principal component analysis is a statistical procedure that uses an orthogonal transformation to convert a
set of observations of possibly correlated variables into a set of values of linearly uncorrelated variables
called principal components. This transformation is defined in such a way that the first principal component
has the largest possible variance.
6 REFERENCES AND ACRONYMS
6.1 REFERENCES
[1] NISTIR 7628 Guidelines for Smart Grid Cyber Security: Vol. 1, Smart Grid Cyber Security Strategy,
Architecture, and High-Level Requirements, August 2010. [Online]:
http://www.nist.gov/smartgrid/upload/nistir-7628_total.pdf: NISTIR 7628, 2010.
[2] "The Unbundled Smart Meter concept in a synchro-SCADA framework," in I2MTC - 2016
International In-strumentation and Measurement technology Conference, Taipei, Taiwan, 2016.
[3] EN50160, "Voltage characteristics of electricity supplied by public distribution systems," 1999.
[4] I. 61000-4-30, "Electromagnetic compatibility (EMC) - Part 4-30: Testing and measurement tech-
niques - Power quality measurement methods," 2008.
[5] I. 61000-4-7, "Testing and measurement techniques General guide on harmonics and
interharmonics measurements and instrumentation, for power supply systems and equipent
connected hereto," 2002.
[6] I. 61000-4-15, "Testing and measurement techniques - Flickermeter - Functional and design
specifications."
[7] M.Sanduleac, M.Albu, J.Martins, L.Alacreu, and C.Stanescu, "Power Quality Assessment in LV
networks using new Smart Meters design," in CPE, 2015.
[8] M. Bollen, S. Ronnberg, F. Zavoda, J. Zhong, S. Djokic, R. Das, and M. Halpin, "CIGRE/CIRED Working
Group C4.24-Power Quality and EMC Issues Associated with Future Electricity Networks-Status
Report," in CIRED, 2015.
[9] F. Zavoda, S. Ronnberg, M. Bollen, J. Meyer, and J. Desmer, "CIGRE/CIRED/IEEE Working Group
C4.24-New Measurement Techniques in the Future Grid-Status Report," 2015.
[10] MTE, "Portable Reference Standard PRS 400.3 Technical Brochure, Available: http://www.mte.ch."
[11] L. Gyr, "energy meter type ZMD310 Technical Brochure, Available: http://www.landisgyr.eu."
[12] "Funzionalit Che Devono Essere Assicurate Dai Sistemi Di Smart Metering Di Seconda
Generazione,chapter 2 in http://www.autorita.energia.it/allegati/docs/16/087-16alla.pdf," 2016.
[13] A. Abur and A. G. Exposito, Power System State Estimation: Theory and Implementation. New York:
Marcel Dekker, 2004.
[14] F. C. Schweppe and J. Wildes, "Power system static-state estimation, part I: exact model," IEEE
Trans. Power Appar. and Syst., vol. PAS-89, pp. 120-125, 1970.
[15] H. Yih-Fang, S. Werner, H. Jing, N. Kashyap, and V. Gupta, "State estimation in electric power grids:
meeting new challenges presented by the requirements of the future grid," IEEE Signal Proc. Mag.,
vol. 29, pp. 33-43, 2012.
[16] B. Hayes and M. Prodanovic, "State estimation techniques for electric power distribution systems,"
in Proc. Euro. Mode. Symp., 2014, pp. 303-308.
[17] A. T. Saric and R. M. Ciric, "Integrated fuzzy state estimation and load flow analysis in distribution
networks," IEEE Trans Power Del., vol. 18, pp. 571-578, 2003.
[18] D. Youman, H. Ying, and Z. Boming, "A branch-estimation-based state estimation method for radial
distribution systems," IEEE Trans Power Del., vol. 17, pp. 1057-1062, 2002.
[19] F. Baalbergen, M. Gibescu, and L. van der Sluis, "Modern state estimation methods in power
systems," in Proc. Power Syst. Conf. Expo., 2009, pp. 1-6.
[20] M. Pau, P. A. Pegoraro, and S. Sulis, "Efficient branch-current-based distribution system state
estimation including synchronized measurements," IEEE Trans Instrum. Measure., vol. 62, pp. 2419-
2429, 2013.
[21] R. F. Arritt, R. C. Dugan, R. W. Uluski, and T. F. Weaver, "Investigation load estimation methods with
the use of AMI metering for distribution system analysis," in Proc. IEEE Rural Electric Power Conf.,
2012, pp. B3-1-B3-9.
[22] A. Alimardani, F. Therrien, D. Atanackovic, J. Jatskevich, and E. Vaahedi, "Distribution system state
estimation based on nonsynchronized smart meters," IEEE Trans Smart Grid, vol. 6, pp. 2919-2928,
2015.
[23] G. Lopez, J. Moreno, H. Amaras, and F. Salazar, "Paving the road toward smart grids through large-
scale advanced metering infrastructures," Elec. Power Syst. Resear., 2014.
[24] S. M. S. Alam, B. Natarajan, and A. Pahwa, "Distribution grid state estimation from compressed
measurements," IEEE Trans Smart Grid, vol. 5, pp. 1631-1642, 2014.
[25] L. Junqi, T. Junjie, F. Ponci, A. Monti, C. Muscas, and P. A. Pegoraro, "Trade-offs in PMU deployment
for state estimation in active distribution grids," IEEE Trans Smart Grid, vol. 3, pp. 915-924, 2012.
[26] E. Manitsas, R. Singh, B. C. Pal, and G. Strbac, "Distribution system state estimation using an
artificial neural network approach for pseudo measurement modeling," IEEE Trans Power Syst., vol.
27, pp. 1888-1896, 2012.
[27] W. Jianzhong, H. Yan, and N. Jenkins, "A robust state estimator for medium voltage distribution
networks," IEEE Trans Power Syst., vol. 28, pp. 1008-1016, 2013.
[28] I. Dzafic, M. Gilles, R. A. Jabr, B. C. Pal, and S. Henselmeyer, "Real time estimation of loads in radial
and unsymmetrical three-phase distribution networks," IEEE Trans Power Syst., vol. 28, pp. 4839-
4848, 2013.
[29] Y. Chakhchoukh, V. Vittal, and G. T. Heydt, "PMU based state estimation by integrating correlation,"
IEEE Trans Power Syst., vol. 29, pp. 617-626, 2014.
[30] Y. Yue and S. Roy, "PMU placement for optimal three-phase state estimation performance," in Proc.
IEEE In. Conf. Smart Grid Comm., 2013, pp. 342-347.
[31] P. N. P. Barbeiro, H. Teixeira, J. Krstulovic, J. Pereira, and F. J. Soares, "Exploiting autoencoders for
three-phase state estimation in unbalanced distributions grids," Electric Power Systems Research,
vol. 123, pp. 108-118, 2015.
[32] S. Nanchian, A. Majumdar, and B. C. Pal, "Three-phase state estimation using hybrid particle swarm
optimization," IEEE Trans Smart Grid, vol. PP, pp. 1-1, 2015.
[33] M. M. Nordman and M. Lehtonen, "Distributed agent-based State estimation for electrical
distribution networks," IEEE Trans Power Syst., vol. 20, pp. 652-658, 2005.
[34] V. Kekatos and G. B. Giannakis, "Distributed robust power system state estimation," IEEE Trans
Power Syst., vol. 28, pp. 1617-1626, 2013.
[35] A. Gomez-Exposito, A. Abur, A. de la Villa Jaen, and C. Gomez-Quiles, "A Multilevel State Estimation
Paradigm for Smart Grids," Proc. IEEE, vol. 99, pp. 952-976, 2011.
[36] S. Kuang-Rong and H. Shyh-Jier, "Application of a robust algorithm for dynamic state estimation of
a power system," IEEE Trans Power Syst., vol. 17, pp. 141-147, 2002.
[37] J. Arrillaga, N. R. Watson, and S. Chen, Power System Quality Assessment. Chichester: Wiley, 2000.
[38] C.-M. Kuan, Generalized least squares theory. Taipei, Taiwan: Inst. Econ., Academia Sinica, 2004.
[39] R. Singh, B. C. Pal, and R. B. Vinter, "Measurement placement in distribution system state
estimation," IEEE Trans. Power Syst., vol. 24, pp. 668-675, 2009.
[40] C. W. Hansen, "Power system state estimation using three phase models," IEEE Trans. Power Syst.,
vol. 10, pp. 818-824, 1995.
[41] J. Arrillaga and N. R. Watson, Computer modelling of power systems. New York: Wiley, 2001.
[42] C. L. Fortescue, "Method of symetrical co-ordinates to the solution of polyphase networks," in 34th
Annu. Convention of the Amer. Inst. Elect. Eng., Atlantic City, NJ, , 1918.
[43] F. Corcoles, L. Sainz, J. Pedra, J. Sanchez-Navarro, and M. Salichs, "Three-phase transformer
modelling for unbalanced conditions," IET Elect. Power Appl., vol. 2, pp. 99-112, 2007.
[44] N. C. Woolley and J. V. Milanovi, "Statistical estimation of the source and level of voltage
unbalance in distribution networks," IEEE Trans. Power Del., vol. 27, pp. 1450-1460, 2012.
[45] I. Roytelman and S. M. Shahidehpour, "State estimation for electric power distribution systems in
quasi real-time conditions," IEEE Trans Power Deli., vol. 8, pp. 2009-2015, 1993.
[46] K. A. Clements, "The impact of pseudo-measurements on state estimator accuracy," in Proc. Power
and Energy Soc. General Meeting, 2011, pp. 1-4.
[47] A. Angioni, C. Muscas, S. Sulis, F. Ponci, and A. Monti, "Impact of heterogeneous measurements in
the state estimation of unbalanced distribution networks," in Proc. Int. Instrum. Meas.
Technol.Conf., 2013, pp. 935-939.
[48] C. Muscas, S. Sulis, A. Angioni, F. Ponci, and A. Monti, "Impact of different uncertainty sources on a
three-phase state estimator for distribution networks," IEEE Trans. Instru. and Measure., vol. 63,
pp. 2200-2209, 2014.
[49] L. Ke, "State estimation for power distribution system and measurement impacts," IEEE Trans
Power Syst., vol. 11, pp. 911-916, 1996.
[50] D. Macii, G. Barchi, and D. Petri, "Uncertainty sensitivity analysis of WLS-based grid state
estimators," in Proc. Int. Workshop on Applied Measure. for Power Syst., 2014, pp. 1-6.
[51] G. D'Antona and M. Davoudi, "Effects of parameter and measurement uncertainties on the power
system WLS state estimation," in Proc. Instrum. Measure. Tech.y Conf., 2012, pp. 1015-1020.
[52] C. Muscas, F. Pilo, G. Pisano, and S. Sulis, "Considering the uncertainty on the network parameters
in the optimal planning of measurement systems for Distribution State Estimation," in Proc.
Instrum. Measure. Tech. Conf., 2007, pp. 1-6.
[53] J. Pereira, J. T. Saraiva, and V. Miranda, "An integrated load allocation/state estimation approach
for distribution networks," in Proc. Int. Conf. Prob. Methods Applied to Power Syst., 2004, pp. 180-
185.
[54] R. Singh, E. Manitsas, B. C. Pal, and G. Strbac, "A Recursive Bayesian Approach for Identification of
Network Configuration Changes in Distribution System State Estimation," IEEE Trans Power Syst.,
vol. 25, pp. 1329-1336, 2010.
[55] C. Muscas, M. Pau, P. A. Pegoraro, and S. Sulis, "Impact of input data correlation on Distribution
System State Estimation," in Proc. Int. Work. on Applied Measure. Power Syst., 2013, pp. 114-119.
[56] Y. Zhang and J. V. Milanovi, "Voltage sag cost reduction with optimally placed FACTS devices," in
Proc. Int. Conf. on Elec. Power Quality Utili., 2007, pp. 1-6.
[57] H. L. Liao, S. Abdelrahman, Y. Guo, and J. V. Milanovi, "Identification of weak areas of power
network based on exposure to voltage sagsPart II: assessment of network performance using sag
severity index," IEEE Trans. Power Del. , vol. 30, pp. 2401 - 2409, 2015.
[58] W. H. Kersting, Distribution System Modeling and Analysis. New York: CRC Press, 2002.
[59] J. R. Berg, E. S. Hawkins, and W. W. Pleines, "echanized calculation of unbalanced load flow on
radial distribution circuits," IEEE Trans. Power App. Syst., vol. 86, pp. 415-421, 1967.
[60] M. E. Baran and F. F. Wu, "Optimal sizing of capacitors placed on a radial distribution system," IEEE
Trans. Power Del., vol. 4, pp. 735742, 1989.
[61] C. S. Cheng and D. Shirmohammadi, "A three-phase power flow method for real-time distribution
system analysis," IEEE Trans. Power Syst., vol. 10, pp. 671-679, 1995.
[62] G. X. Luo and A. Semlyen, "Efficient load flow for large weakly meshed networks," IEEE Trans.
Power Syst., vol. 5, pp. 1309-1316, 1990.
[63] D. Shirmohammadi, H. W. Hong, A. Semlyen, and G. X. Luo, "A compensation-based power flow
method for weakly meshed distribution and transmission networks," IEEE Trans. Power Syst., vol. 3,
pp. 753762, 1988.
[64] H. D. Chiang, "A decoupled load flow method for distribution power networks: Algorithms, analysis
and convergence study," Int. J. Elect. Power Energy Syst., vol. 13, pp. 130138, 1991.
[65] W. H. Kersting and D. L. Mendive, "An application of Ladder network theory to the solution of three
phase radial load flow problem," in IEEE PES Winter Meeting, 1976.
[66] R. D. Zimmerman and H. D. Chiang, "Fast decoupled power flow for unbalanced radial distribution
systems," IEEE Trans. Power Syst., vol. 10, pp. 20452052, 1995.
[67] R. Zimmerman, "Comprehensive distribution power flow: modeling, formulation, solution
algorithms and analysis," Cornel Univercity, 1995.
[68] "IEEE PES Distribution System Analysis Subcommittee Distribution Test Feeders, Available:
http://ewh.ieee.org/soc/pes/dsacom/testfeeders/," 2013.
[69] T. Senjyu, S. Higa, and K. Uezato, "Future load curve shaping based on similarity using fuzzy logic
approach," , vol. , pp. , ." in IEE Proceedings Generation, Transmission and Distribution, 1998, pp.
375-380.
[70] E. H. Barakat, M. A. Qayyum, M. N. Hamed, and S. A. Al-Rashed, "Short-term peak demand
forecasting in fast developing utility with inherit dynamic load characteristics," IEEE Transactions on
Power Systems, vol. 5, pp. 813-824, 1990.
[71] W. History, "Weather.org [Online]. Available:
http://weather.org/weatherorg_records_and_averages.htm."
[72] S. Xu and L. Chen, "A novel approach for determining the optimal number of hidden layer neurons
for FNN's and its application in data mining," in 5th ICITA, 2008, pp. 683-686.
[73] Y. Liu, J. A. Starzyk, and Z. Zhu, "Optimizing number of hidden neurons in neural network," in 25th
IASTED International Multi-Conference: Artificial Intelligence and Applications, 2007, pp. 121-126.
[74] I. Ilean, C. Rotar, and A. Incze, "The optimization of feed forward neural netowrks structure using
genetic algorithms," in ICTAMI, Thessalonili, Greece, 2004, pp. 223-230.
[75] H. Demuth, M. Beale, and M. Hagan, "MATLAB Neural network tookbox user's guide ed, ," 2009.
[76] K. Gurney, An introduction to neural network: University College London Press, 1997.
[77] T. G. Dietterich, Ensemble Learning. Cambridge, 2002.
[78] W. Yao, X. Chen, Y. Zhao, and M. v. Tooren, "Concurrent Subspace Width Optimization Method for
RBF Neural Network Modeling," IEEE Trans. Neural Networks Learning Syst., vol. 23, pp. 247-259,
2012.
[79] G. Sideratos and N. Hatziargyriou, "An Advanced Statistical Method for Wind Power Forecasting,"
IEEE Transaction on Power System, vol. 22, pp. 258-265, 2007.
[80] G. N. Sideratos and Hatziargyriou, "An Advanced Radial Base Structure for Wind Power
Forecasting," Int. journal on Power and Energy Systems, vol. 12, 2008.
[81] L. Breiman, "Random forests," Machine Learning, vol. 45, pp. 5-32, 2001.
[82] A. Liaw and M. Wiener, "Classification and regression by randomForest," R News, vol. 2, pp. 18
22, 2002.
[83] A. Criminisi, J. Shotton, and E. Konukoglu, "Decision Forests for Classification, Regression, Density
Estimation, Manifold Learning and Semi-Supervised Learning," Microsoft Research technical report
TR-2011-114, p. 24, 2011.
[84] M. E. Baran and I. M. El-Markabi, "A multiagent-based dispatching scheme for distributed
generators for voltage support on distribution feeders," IEEE Trans. Power Syst., vol. 22, pp. 5259,
2007.
[85] M. H. J. Bollen and A. Sannino, "Voltage control with inverter-based distributed generation," IEEE
Trans. Power Del., vol. 20, pp. 519520, 2005.
[86] P. M. S. Carvalho, P. F. Correia, and L. A. F. Ferreira, "Distributed reactive power generation control
for voltage rise mitigation in distribution networks," IEEE Trans. Power Syst., vol. 23, pp. 766772,
2008.
[87] S. R. Islam, K. M. Muttaqi, and D. Sutanto, "A Decentralized Multiagent-Based Voltage Control for
Catastrophic Disturbances in a Power System," IEEE Transactions on Industry Applications, vol. 51,
pp. 1201-1214, 2015.
[88] P. Pachanapan, O. Anaya-Lara, A. Dysko, and K. L. Lo, "Adaptive Zone Identification for Voltage
Level Control in Distribution Networks With DG," IEEE Transactions on Smart Grid, vol. 3, pp. 1594-
1602, 2012.
[89] L. Yu, D. Czarkowski, and F. d. Leon, "Optimal distributed voltage regulation for secondary networks
with DGs," IEEE Transactions on smart grid, vol. 3, 2012.
[90] H. Mehrjerdi, S. Lefebvre, M. Saad, and D. Asber, "A Decentralized Control of Partitioned Power
Networks for Voltage Regulation and Prevention Against Disturbance Propagation," IEEE
Transactions on Power Systems, vol. 28, pp. 1461-1469, 2013.
[91] USEF: THE FRAMEWORK SPECIFICATIONS 2015, (ISBN:
978-90-824625-1-7. online: http://www.usef.info/Framework/Download-the-framework.aspx, 2015.
6.2 ACRONYMS
Acronyms List
AI Artificial Intelligence
AMI Advanced Metering Infrastructure
AD Active Demand
AGR Aggregator
ANN Artificial Neural Network
BLF Basic Load Forecaster
BRBP Bayesian Regulation Backpropagation
BS Black Start
BF breadth First
CLF Change Load Forecaster
CAIDI Customer Average Interruption Duration Index
CSV Comma Separated File
CIM Common Information Model
CB Capacitor Banks
DMS Distribution Management System
DLMS Device Language Message Specification

DSM Demand Side Management
DG Distributed Generators
D.T.S. Data Transformation Service
DR Demand Response
DP Distribution of Probability
DRFM Demand Response Flexible Market
DACF Data Acquisition and Control Frontend
DSSE Distribution System State Estimation
DRMS Demand Response Management System
DSP Digital Signal Processor
DER Distributed Energy Resource

DSO/DNO Distribution System Operator/ Distribution Netowrk Operator
EMA Energy Monitoring and Analytics
ESB Enterprise Service Bus
EV Electric Vehicle
EMS Energy Management Systems
G3M Grid Management and Maintainance Master Framework
GLM Generalized Linear Models
GLS Generalized Least Squares
GDN Generic Distribution Network
GUI Graphical User Interface
GP Gaussian Processes
GDBP Gradient Descent Backpropagation
GPS Global Positioning System

HTTP Hypertext Transfer Protocol
IEC International Electrotechnical Commission

IR Internal Report
ISO International Organization for Standardization
IED Intelligent Electronic Device
ICT Information and Communication Technology
KCL Kirchhoff's Current Law
KPI Key Performance Indicators
LMBP Levenberg-Marquardt BackPropagation
LE Load Estimation
LDC Line Drop Compensator
LPQAM Light PQ Assessemnt Method
LIC Logical Interface Category
LPF Low pass Filter

LMS Language Message Specification
MLP Multilayer Perceptrons
MDM Meter Data Management

MDMS Meter Data Management System
MCM Market-based Coordination Mechanism
MG Micro-Grids
MAPE Mean Absolute Percentage Error
MAIFI Momentary Average Interruption Frequency Index
MM Mixed Model
M2M Machine to Machine
MQTT Message Queuing Telemetry Transport
MV/LV/HV Medium Voltage / Low Voltage/ High Voltage
NWP Numerical Weather Predictions
NTP Network Time Protocol
NIST National Institute of Standards and Technology
NISTIR NIST Internal Report
OLTC OnLoad Tap-Changing

OGEMA Open Gateway Energy Management
PV Photovoltaic
PLL Phase-Locked-Loop
PM Pseudo-Measurements
PD Probability Distribution
PTU Program Time Unit
PMU Phasor Measurement Units
PQ Power Quality
RBF Reverse Breadth First
RMS Root Mean Square
RBFNN Radial Basis Function Neural Network
RF Random Forests
RES Renewable Energy Sources
RTU Remote Terminal Unit
RTO Regional Transmission Organizations
SHIC Smart Home Intelligent Controller
SCADA Supervisory Control and Data Acquisition
SP Scheduled Power
SE State Estimation
SMX Smart Meter eXtension
SM Smart Meters
SLAM Smart Low-cost Advanced Meter
SMM Smart Metrology Meter
SGAM Smart Grid Architecture Model

SAIFI System Average Interruption Frequency Index
SAIDI System Average Interruption Duration Index
SG Smart Grid
TSO Transmission System Operator
USM Unbundled Smart Meter
USEF Universal Smart Energy Framework
VR Voltage Regulators
VSI Voltage source Inverter
VPP Virtual Power Plants
WLS Weighted Least Squares
WP Work Package
7 APPENDIX
7.1 APPENDIX A
The developed three-phase power flow software tool is validated using the IEEE13-bus and 123-bus
distribution test systems [68] and the obtained results for these systems were compared with the results as
given in [68]. Furthermore, it is applied to Meltemi distribution network. Meltemi distribution network is a
real world 0.4 kV distribution network with 57 buses.
7.1.1 IEEE 13-bus distribution test system

The IEEE 13-bus distribution test system is presented in Figure 7.1. The data of the IEEE 13-bus test system
are given in Table 7.1 [68]. The technical characteristics of the distribution lines can be found in Appendix A.
The results of the developed three phase power flow analysis tool are presented in Table 7.2 and Table 7.3.
Table 7.2 presents the bus voltage magnitudes and angles of the system. Table 7.3 presents the line current
magnitudes and angles of the system. The obtained results of the Table 7.2 and Table 7.3 are almost
identical to the results of [68]. The system is consisted of 8 concentrated loads, one distributed load
between buses 632 and 671, 10 distribution lines, one transformer, one voltage regulator and one closed
switch.
650
646 645 632 633 634
611 684 692 675

671
652 680
Figure 7.1 IEEE 13-bus distribution test system [68]
Table 7.1 Data of the IEEE 13-bus test system
1. DISTRIBUTION LINES
Type of phase Type of neutral Spacing Phase Length
From To Type
conductor conductor ID Sequence (km)
556,500 26/7
650 632 Overhead 4/0 6/1 ACSR 500 BACN 0,6096
ACSR
632 633 Overhead 4/0 6/1 ACSR 4/0 6/1 ACSR 500 CABN 0,1524
632 645 Overhead 1/0 ASCR 1/0 ASCR 505 CBN 0,1524
645 646 Overhead 1/0 ASCR 1/0 ASCR 505 CBN 0,0914
556,500 26/7
632 671 Overhead 4/0 6/1 ACSR 500 BACN 0, 6096
ACSR
692 675 Underground 250,00 AA, CN - 515 ABC 0,1524

671 684 Overhead 1/0 ASCR 1/0 ASCR 505 ACN 0,0914
684 611 Overhead 1/0 ASCR 1/0 ASCR 510 CN 0,0914
684 652 Underground 1/0 AA, TS 1/0 Cu 520 AN 0,2438
556,500 26/7
671 680 Overhead 4/0 6/1 ACSR 500 BACN 0,3048
ACSR
2. TRANSFORMERS
Rated Nominal voltage of Nominal voltage
R X
From To Connection Capacity the primary of the secondary
(%) (%)
(kVA) winding (kV) winding (kV)
Grounded Wye
633 634 500 4.16 0.48 1.1 2.0
Grounded Wye
3. VOLTAGE REGULATORS
BWreg R (V) X (V) Vrelay (V)

From To Connection Npt CTp
(V) A B C A B C A B C
Grounded
650 632 2 20 700 3 3 3 9 9 9 122 122 122
Wye
4. LOADS
Nominal apparent power ( kW + jkvar)
Bus Connection Type
Phase A Phase B Phase C
634 Grounded Wye Constant Power 160 + j110 120 + j90 120 + j90
645 Grounded Wye Constant Power 0 170 + j125 0
646 Open Delta Constant Impedance 0 230 + j132 0
652 Grounded Wye Constant Impedance 128 + j86 0 0
671 Open Delta Constant Power 385 + j220 385 + j220 385 + j220
692 Open Delta Constant Current 0 0 170 + j151
611 Grounded Wye Constant Current 0 0 170 + j80
632 17 + j10 66 + j38 117 + j68

Grounded Wye Constant Power
671
5. SHUNT CAPACITORS
Nominal reactive power (kvar)
Bus Connection
675 Grounded Wye 200 200 200
611 Grounded Wye 0 0 100
Table 7.2 Bus voltages of the IEEE 13-bus distribution system

Bus |V|ph_A (V) ph_A (deg.) |V|ph_B (V) ph_B (deg.) |V|ph_C (V) ph_C (deg.)
650 2401.755 0 2401.787 -120 2401.747 120
632 2462.705 -2.47 2509.078 -121.72 2456.803 117.84
645 0 -2.47 2487.063 -121.9 2452.027 117.87
646 0 -2.47 2482.876 -121.97 2447.058 117.92
633 2455.467 -2.53 2504.531 -121.76 2450.581 117.84
634 276.709 -3.2 283.911 -122.21 277.57 117.37
671 2388.506 -5.25 2535.299 -122.34 2362.375 116.05
684 2383.814 -5.28 0 -122.34 2357.557 115.95
611 0 -5.28 0 -122.34 2352.776 115.81
652 2370.371 -5.2 0 -122.34 0 115.95
680 2388.506 -5.25 2535.299 -122.34 2362.375 116.05
692 2388.503 -5.25 2535.299 -122.34 2362.373 116.05
675 2372.994 -5.5 2541.019 -122.51 2357.857 116.07
Table 7.3 Line currents of the IEEE 13-bus distribution system

Bus
Iph_A (A) ph_A (deg.) Iph_B (A) ph_B (deg.) Iph_C (A) ph_C (deg.)
From To
650 632 593.388 -28.5 435.512 -140.8 627.283 93.72
632 671 476.563 -26.94 214.19 -134.49 473.318 100.09
632 633 80.964 -37.71 60.961 -159.08 62.353 80.5
632 645 0 0 143.032 -142.58 65.46 57.88
645 646 0 0 65.462 -122.12 65.461 57.88
633 634 80.965 -37.71 60.962 -159.08 62.354 80.5
671 680 0.004 90.68 0.003 -40.93 0.003 -150.45
671 684 63.348 -39.07 0 0 71.175 121.84
671 692 227.994 -17.99 69.84 -55.04 177.522 109.73
684 611 0 0 0 0 71.175 121.84
684 652 63.348 -39.07 0 0 0 0
692 675 204.395 -4.88 69.84 -55.04 123.262 112.28
7.1.2 IEEE 123-bus distribution test system

The IEEE 123-bus distribution test system is presented in Figure 7.2. The data of the IEEE 123-bus test
system are given in Table 7.4. The technical characteristics of the distribution lines can be found in
Appendix A. The results of the developed three phase power flow analysis tool are presented in Table 7.5
and Table 7.6. Table 7.5 presents the bus voltage magnitudes and angles of the system. Table 7.6 presents
the line current magnitudes and angles of the system. The obtained results of the Table 7.5 and Table 7.6
are almost identical to the results of [68]. The system is consisted of 85 concentrated loads, 117
distribution lines, one transformer, 4 voltage regulators and 11 switches.
32 29 25 0 35 0
30
33 25 1 51 11 1 11 0 11 2 11 3 11 4
28 50 15 1 30 0
31 49
10 9 10 7
25 47
48 46
26 10 8 45 1
45 10 6 10 4
27 64
44 43 10 3
23 65 45 0
10 5 10 2
63 10 0
42 41
24 66 10 1
21 99
40 71
22 98
39 62 19 7 70
38
35 36 97 69
19 13 5
20 18 68
75
16 0 67
37 74
60
73
57
14 58
11 59 72 85
61 79
61 0
9 78
53 54 77
10 52
2 56
15 2 55
76
8 13
7 80
94 84
96
76
14 9 1 34
90 88 81
15 0 12 92
17
15 86 83
95 87 82
91 89
3 93
5 6
16
19 5
4
Figure 7.2 IEEE 123-bus distribution test system [68]
Table 7.4 Data of the IEEE 123-bus test system
From To Type
conductor conductor ID Sequence (km)
336,400 26/7
149 1 overhead 4/0 6/1 ACSR 500 ABC 0,1219
ACSR
1 2 overhead 1/0 ASCR 1/0 ASCR 510 B 0,0533
1 3 overhead 1/0 ASCR 1/0 ASCR 510 C 0,0762
336,400 26/7
ACSR
336,400 26/7
ACSR
8 9 overhead 1/0 ASCR 1/0 ASCR 510 A 0,0686
336,400 26/7
ACSR
336,400 26/7
13 18 overhead 4/0 6/1 ACSR 500 CAB 0,2515
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
25 26 overhead 4/0 6/1 ACSR 505 AC 0,1067
ACSR
336,400 26/7
26 27 overhead 4/0 6/1 ACSR 505 AC 0,0838
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
135 35 overhead 4/0 6/1 ACSR 500 CBA 0,1143
ACSR
336,400 26/7
35 36 overhead 4/0 6/1 ACSR 505 AB 0,1981
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
49 50 overhead 336,400 26/7 4/0 6/1 ACSR 500 CBA 0,0762
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
54 57 overhead 4/0 6/1 ACSR 500 BCA 0,1067
ACSR
336,400 26/7
ACSR
336,400 26/7
60 61 overhead 4/0 6/1 ACSR 500 BAC 0,1676
ACSR
60 62 underground 1/0 AA, CN - 515 ABC 0,0762
336,400 26/7
160 67 overhead 4/0 6/1 ACSR 500 ACB 0,1067
ACSR
336,400 26/7
ACSR
72 76 overhead 336,400 26/7 4/0 6/1 ACSR 500 BCA 0,0610
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
98 99 overhead 336,400 26/7 4/0 6/1 ACSR 500 BCA 0,1676
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
336,400 26/7
ACSR
2. TRANSFORMERS
Rated Nominal voltage Nominal voltage
R
From To Connection Capacity of the primary of the secondary X (%)
(%)
61 610 Delta-Delta 150 4.16 0.48 1.27 2.72
3. VOLTAGE REGULATORS
BWreg R (V) X (V) Vrelay (V)

From To Connection Npt CTp
(V) A B C A B C A B C
Grounded
150 149 2 20 700 3 - - 9 - - 120 - -
Wye
Phase A
9 14 2 20 50 0.4 - - 0.4 - - 20 - -
Ground
Phase A, C
25 26 1 20 50 0.4 - 0.4 0.4 - 0.4 20 - 20
Ground
Grounded
160 167 2 20 300 0.6 1.4 0.2 1.3 2.6 1.4 124 124 124
Wye
4. LOADS
Bus Connection Type
1 Grounded Wye Constant Power 40 + j20 0 0
4 Grounded Wye Constant Power 0 0 40 + j20
5 Grounded Wye Constant Current 0 0 20 + j10
6 Grounded Wye Constant Impedance 0 0 40 + j20
10 Grounded Wye Constant Current 20 + j10 0 0
22 Grounded Wye Constant Impedance 0 40 + j20 0
35 Open Delta Constant Power 40 + j20 0 0
38 Grounded Wye Constant Current 0 20 + j10 0
43 Grounded Wye Constant Impedance 0 40 + j20 0
47 Grounded Wye Constant Current 35 + j25 35 + j25 35 + j25
48 Grounded Wye Constant Impedance 70 + j50 70 + j50 70 + j50
65 Open Delta Constant Impedance 35 + j25 35 + j25 70 + j50
76 Open Delta Constant Current 105 + j80 70 + j50 70 + j50
5. SHUNT CAPACITORS
Nominal reactive power (kvar)
Bus Connection
6. SWITCHES
From To Status
13 152 Closed
18 135 Closed
60 160 Closed
61 610 Closed
97 197 Closed
150 149 Closed
250 251 Open
450 451 Open
54 94 Open
151 300 Open
300 350 Open
Table 7.5 Bus voltages of the IEEE 123-bus distribution system

150 2401.714 0 2401.799 -120 2401.797 120
149 2511.596 0 2511.685 -120 2511.684 120
1 2481.757 -0.64 2505.804 -120.31 2490.419 119.62
2 0 -0.64 2505.254 -120.31 0 119.62
3 0 -0.64 0 -120.31 2486.452 119.59
5 0 -0.64 0 -120.31 2483.303 119.57
6 0 -0.64 0 -120.31 2481.671 119.55
4 0 -0.64 0 -120.31 2485.231 119.58
7 2459.521 -1.1 2501.637 -120.56 2476.753 119.37
8 2445.095 -1.41 2498.71 -120.72 2467.697 119.2
12 0 -1.41 2498.011 -120.73 0 119.2
9 2441.614 -1.44 0 -120.72 0 119.2
14 2422.4 -1.47 0 -120.72 0 119.2
10 2421.61 -1.48 0 -120.72 0 119.2
11 2420.808 -1.49 0 -120.72 0 119.2
13 2425.998 -1.84 2493.471 -120.95 2454.147 118.92
34 0 -1.84 0 -120.95 2451.772 118.9
15 0 -1.84 0 -120.95 2450.842 118.89
17 0 -1.84 0 -120.95 2449.758 118.89
16 0 -1.84 0 -120.95 2448.518 118.88
18 2404.363 -2.26 2483.513 -121.2 2436.38 118.85
19 2401.203 -2.29 0 -121.2 0 118.85
20 2399.149 -2.31 0 -121.2 0 118.85
135 2404.364 -2.26 2483.513 -121.2 2436.379 118.85
35 2397.7 -2.35 2477.4 -121.29 2433.836 118.79
36 2395.457 -2.37 2476.284 -121.33 0 118.79
37 2393.568 -2.38 0 -121.33 0 118.79
38 0 -2.37 2474.728 -121.35 0 118.79
39 0 -2.37 2473.731 -121.35 0 118.79
40 2394.053 -2.38 2474.648 -121.33 2431.329 118.75
41 0 -2.38 0 -121.33 2430.315 118.74
42 2390.234 -2.42 2471.942 -121.38 2429.211 118.71
43 0 -2.42 2468.695 -121.41 0 118.71
44 2387.608 -2.45 2470.241 -121.41 2427.231 118.68
45 2386.34 -2.46 0 -121.41 0 118.68
46 2385.386 -2.47 0 -121.41 0 118.68
47 2385.115 -2.47 2467.666 -121.44 2424.825 118.63
48 2384.451 -2.48 2466.971 -121.45 2424.293 118.62
49 2384.457 -2.48 2466.346 -121.45 2424.067 118.61
50 2384.408 -2.49 2466.409 -121.44 2423.254 118.6
51 2384.013 -2.5 2466.6 -121.44 2423.22 118.6
151 2384.013 -2.5 2466.6 -121.44 2423.22 118.6
21 2403.035 -2.32 2483.64 -121.2 2433.58 118.83
22 0 -2.32 2480.215 -121.23 0 118.83
23 2402.029 -2.37 2484.556 -121.19 2430.889 118.82
24 0 -2.37 0 -121.19 2427.451 118.79
25 2400.434 -2.42 2485.643 -121.18 2428.784 118.82
28 2399.543 -2.45 2486.19 -121.18 2427.972 118.82
29 2399.143 -2.47 2486.556 -121.17 2426.895 118.81
30 2399.77 -2.48 2486.454 -121.16 2425.806 118.79
250 2399.77 -2.48 2486.454 -121.16 2425.806 118.79
26 2399.82 -2.45 0 -121.18 2412.513 118.81
31 0 -2.45 0 -121.18 2411.097 118.79
32 0 -2.45 0 -121.18 2410.152 118.79
27 2398.955 -2.47 0 -121.18 2412.44 118.82
33 2395.796 -2.49 0 -121.18 0 118.82
152 2425.998 -1.84 2493.471 -120.95 2454.147 118.92
52 2411.734 -2.22 2490.482 -121.19 2446.394 118.67
53 2405.265 -2.39 2488.647 -121.31 2442.59 118.54
54 2401.625 -2.5 2487.282 -121.38 2440.253 118.46
55 2401.149 -2.5 2487.104 -121.39 2440.41 118.46
56 2401.076 -2.49 2486.689 -121.4 2440.612 118.46
57 2394.19 -2.8 2480.545 -121.58 2434.068 118.24
58 0 -2.8 2478.983 -121.59 0 118.24
59 0 -2.8 2478.218 -121.6 0 118.24
60 2378.732 -3.47 2468.52 -121.97 2419.596 117.79
61 2397.207 -2.67 2430.194 -121.97 2439.173 117.01
610 480.689 27.91 489.049 -92.42 482.502 146.88
62 2376.882 -3.47 2465.79 -121.94 2414.656 117.78
63 2375.423 -3.46 2463.705 -121.94 2412.251 117.77
64 2374.623 -3.43 2459.169 -121.9 2407.066 117.74
65 2372.949 -3.44 2458.354 -121.86 2399.868 117.73
66 2373.518 -3.47 2458.929 -121.83 2396.262 117.73
160 2378.732 -3.47 2468.52 -121.97 2419.596 117.79
67 2499.447 -3.72 2481.818 -122.15 2492.529 117.65
72 2500.426 -3.81 2479.682 -122.25 2492.053 117.54
73 0 -3.81 0 -122.25 2486.894 117.5
74 0 -3.81 0 -122.25 2482.467 117.46
75 0 -3.81 0 -122.25 2480.019 117.44
76 2500.238 -3.87 2478.433 -122.34 2493.566 117.49
77 2503.045 -3.94 2481.196 -122.42 2495.729 117.41
78 2503.784 -3.96 2482.19 -122.44 2496.12 117.39
79 2503.063 -3.97 2482.508 -122.44 2496.002 117.4
80 2508.85 -4.02 2486.233 -122.5 2498.218 117.28
81 2514.074 -4.09 2491.717 -122.53 2499.623 117.18
84 0 -4.09 0 -122.53 2493.454 117.13
85 0 -4.09 0 -122.53 2490.56 117.11
82 2516.171 -4.13 2494.729 -122.56 2501.496 117.15
83 2519.019 -4.15 2497.414 -122.59 2503.483 117.11
86 2498.158 -3.9 2474.178 -122.5 2497.01 117.45
87 2496.95 -3.92 2472.263 -122.59 2498.338 117.43
88 2496.853 -3.95 0 -122.59 0 117.43
89 2495.918 -3.92 2471.625 -122.64 2499.312 117.42
90 0 -3.92 2471.433 -122.68 0 117.42
91 2495.438 -3.92 2470.838 -122.65 2499.834 117.41
92 0 -3.92 0 -122.65 2499.671 117.36
93 2494.685 -3.92 2470.45 -122.67 2500.065 117.41
94 2493.011 -3.94 0 -122.67 0 117.41
95 2494.59 -3.91 2469.535 -122.69 2500.51 117.41
96 0 -3.91 2468.92 -122.69 0 117.41
68 2495.787 -3.74 0 -122.15 0 117.65
69 2491.595 -3.78 0 -122.15 0 117.65
70 2488.621 -3.8 0 -122.15 0 117.65
71 2486.943 -3.81 0 -122.15 0 117.65
97 2497.046 -3.77 2480.61 -122.17 2490.788 117.64
98 2496.565 -3.78 2479.991 -122.18 2490.292 117.63
99 2497.261 -3.77 2478.026 -122.18 2489.47 117.59
100 2497.739 -3.77 2477.868 -122.17 2488.499 117.57
450 2497.739 -3.77 2477.868 -122.17 2488.499 117.57
197 2497.046 -3.77 2480.61 -122.17 2490.788 117.64
101 2495.083 -3.81 2479.966 -122.18 2489.498 117.63
102 0 -3.81 0 -122.18 2486.057 117.6
103 0 -3.81 0 -122.18 2482.08 117.57
104 0 -3.81 0 -122.18 2477.793 117.53
105 2491.904 -3.85 2479.596 -122.23 2490.16 117.66
106 0 -3.85 2476.837 -122.25 0 117.66
107 0 -3.85 2473.309 -122.28 0 117.66
108 2488.361 -3.92 2481.143 -122.24 2489.809 117.7
300 2488.362 -3.92 2481.143 -122.24 2489.81 117.7
109 2478.494 -3.99 0 -122.24 0 117.7
110 2473.754 -4.03 0 -122.24 0 117.7
111 2471.988 -4.05 0 -122.24 0 117.7
112 2472.163 -4.04 0 -122.24 0 117.7
113 2467.14 -4.08 0 -122.24 0 117.7
114 2466.141 -4.09 0 -122.24 0 117.7
Table 7.6 Line currents of the IEEE 123-bus distribution system

Bus
From To
150 149 656.762 -21.58 426.333 -139.56 524.43 101.6
149 1 628.029 -21.58 407.681 -139.56 501.487 101.6
1 7 610.098 -21.41 398.83 -139.4 455.519 102.47
1 2 0 0 8.925 -146.88 0 0
1 3 0 0 0 0 46.543 93
3 4 0 0 0 0 17.995 93.02
3 5 0 0 0 0 28.549 93
5 6 0 0 0 0 19.239 92.99
7 8 601.062 -21.32 398.831 -139.4 455.519 102.47
8 13 555.187 -20.76 389.966 -139.22 455.519 102.47
8 9 46.219 -28.03 0 0 0 0
8 12 0 0 8.951 -147.29 0 0
9 14 27.902 -28.05 0 0 0 0
14 11 18.767 -28.05 0 0 0 0
14 10 9.31 -28.04 0 0 0 0
13 152 331.795 -14.12 244.011 -129.19 264.899 112.22
13 34 0 0 0 0 46.399 92.32
13 18 228.852 -30.41 155.607 -155.07 153.219 88.63
34 15 0 0 0 0 27.392 92.32
15 16 0 0 0 0 18.264 92.31
15 17 0 0 0 0 9.127 92.32
18 21 55.836 -29.03 19.226 -147.78 55.29 92.23
18 135 135.822 -31.4 136.56 -156.1 98.102 86.6
18 19 37.244 -28.86 0 0 0 0
19 20 18.62 -28.87 0 0 0 0
135 35 135.822 -31.4 136.56 -156.1 98.102 86.6
35 40 108.617 -34.93 108.516 -155.4 98.102 86.59
35 36 18.555 -28.94 18.348 -147.91 0 0
36 38 0 0 18.349 -147.91 0 0
36 37 18.556 -28.95 0 0 0 0
38 39 0 0 9.039 -147.91 0 0
40 42 108.618 -34.93 108.516 -155.4 88.95 86.02
40 41 0 0 0 0 9.2 92.18
42 44 99.318 -35.49 89.573 -156.98 88.951 86.02
42 43 0 0 19.138 -147.97 0 0
44 47 80.781 -36.98 89.573 -156.98 88.951 86.02
44 45 18.684 -29.03 0 0 0 0
45 46 9.374 -29.03 0 0 0 0
47 49 27.341 -34.95 34.877 -156.99 35.071 90.53
47 48 35.558 -38.02 36.788 -156.99 36.152 83.08
49 50 9.379 -29.05 0.002 -27.02 18.455 92.04
50 51 9.379 -29.06 0.001 -27.02 0.001 -159.79
51 151 0 0 0 0 0 0
21 23 55.837 -29.03 0.005 -34.78 55.291 92.23
21 22 0 0 19.228 -147.79 0 0
23 25 55.837 -29.03 0.004 -34.78 36.868 92.24
23 24 0 0 0 0 18.423 92.23
25 26 18.619 -29.05 0 0 18.435 92.23
25 28 37.218 -29.02 0.003 -34.77 18.434 92.24
28 29 18.599 -29.03 0.003 -34.77 18.434 92.23
29 30 0.002 83.54 0.002 -34.77 18.435 92.23
30 250 0.001 83.54 0.001 -34.77 0.001 -143.62
26 27 18.619 -29.05 0 0 0.001 -142.27
26 31 0 0 0 0 18.551 92.23
31 32 0 0 0 0 9.277 92.23
27 33 18.62 -29.06 0 0 0 0
152 52 331.795 -14.12 244.011 -129.19 264.899 112.22
52 53 313.891 -13.27 244.011 -129.19 264.899 112.22
53 54 296.034 -12.29 244.011 -129.19 264.899 112.22
54 57 287.135 -11.76 235.516 -128.48 264.899 112.21
54 55 9.307 -29.06 8.991 -147.95 0.002 -155.48
55 56 0.001 96.1 8.992 -147.96 0.001 -155.48
57 60 287.135 -11.76 218.341 -126.86 264.899 112.21
57 58 0 0 18.332 -148.16 0 0
58 59 0 0 9.023 -148.16 0 0
60 160 240.081 -5.7 171.739 -119.85 191 120.51
60 61 0.002 91.85 0.002 -39.99 0.002 -149.38
60 62 45.384 -41.33 52.28 -150.52 80.753 92.25
61 610 0 0 0 0 0 0
62 63 45.389 -41.34 52.283 -150.52 62.041 92.56
63 64 27.185 -49.16 52.286 -150.53 62.043 92.55
64 65 27.192 -49.18 18.032 -157.47 62.048 92.54
65 66 0.01 86.54 0.01 -31.85 34.535 92.73
160 67 240.081 -5.7 171.739 -119.85 191 120.51
67 97 82.499 -30.54 54.176 -148.79 64.347 91.01
67 68 53.875 -30.35 0 0 0 0
67 72 118.879 23.03 126.01 -107.84 132.861 134.31
72 76 118.879 23.03 126.01 -107.84 100.19 156.58
72 73 0 0 0 0 55.26 90.9
73 74 0 0 0 0 37.278 90.89
74 75 0 0 0 0 18.032 90.88
76 86 32.558 5.99 57.576 -129.75 21.048 157.87
76 77 78.212 60.67 77.579 -57.15 77.705 -177.27
77 78 78.211 60.66 80.206 -44.17 77.704 -177.27
78 80 80.975 74.53 80.205 -44.17 77.703 -177.27
78 79 19.405 -30.53 0.001 -28.88 0.001 -148.29
80 81 80.974 74.53 86.59 -32.59 77.701 -177.27
81 82 80.972 74.53 86.589 -32.59 83.19 -158.4
81 84 0 0 0 0 26.923 90.56
84 85 0 0 0 0 17.956 90.55
82 83 87.337 85.85 86.588 -32.59 83.189 -158.4
86 87 32.557 5.99 49.139 -126.26 21.047 157.87
87 89 17.937 -30.49 33.23 -114.03 21.046 157.86
87 88 21.035 36.44 0 0 0 0
89 91 17.938 -30.49 18.11 -149.25 21.045 157.86
89 90 0 0 21.186 -84.5 0 0
91 93 17.938 -30.49 18.111 -149.25 0.002 -148.29
91 92 0 0 0 0 21.043 157.85
93 95 0.001 78.21 18.111 -149.25 0.001 -148.29
93 94 17.938 -30.5 0 0 0 0
95 96 0 0 9.057 -149.26 0 0
68 69 44.916 -30.36 0 0 0 0
69 70 26.967 -30.37 0 0 0 0
70 71 17.982 -30.38 0 0 0 0
97 197 64.589 -30.61 36.133 -148.82 45.057 91.01
97 98 17.911 -30.33 18.044 -148.73 19.29 91.02
98 99 0.005 82.78 18.044 -148.74 19.29 91.02
99 100 0.003 82.78 0.004 -24.21 19.291 91.01
100 450 0.002 82.78 0.003 -24.21 0.003 -156.83
197 101 64.589 -30.61 36.133 -148.82 45.057 91.01
101 105 64.589 -30.61 36.134 -148.82 0.006 -156.79
101 102 0 0 0 0 45.06 91
102 103 0 0 0 0 36.066 90.99
103 104 0 0 0 0 18.048 90.97
105 108 64.59 -30.61 0.004 -24.23 0.005 -156.79
105 106 0 0 36.137 -148.82 0 0
106 107 0 0 18.081 -148.84 0 0
108 109 64.591 -30.61 0 0 0 0
108 300 0.003 82.66 0.003 -24.23 0.004 -156.79
109 110 46.548 -30.63 0 0 0 0
110 112 37.503 -30.64 0 0 0 0
110 111 9.045 -30.6 0 0 0 0
112 113 28.193 -30.65 0 0 0 0
113 114 9.067 -30.65 0 0 0 0
7.1.3 Meltemi distribution network

The Meltemi distribution network is a real world low voltage distribution network with 57 buses located at
Rafina, Attiki, Greece and it is presented in Figure 7.3. The data of the Meltemi distribution network are
given in Table 7.7. The technical characteristics of the distribution lines can be found in Appendix A. The
system is consisted of 27 concentrated loads, 56 distribution lines, one transformer and one switch. The
24h load profile of all loads is presented in Figure 7.4.
Table 7.8 presents the bus voltage magnitudes and angles of the network during the peak load demand.
Table 7.9 presents the line current magnitudes and angles of the network during the peak load demand.
The total power losses at the peak load demand are 1.32 kW in Phase A, 0.71 kW in Phase B and 1.38 kW in
Phase C. The voltage magnitude deviation of buses 25, 34 and 55 for the load profile of Figure 7.4 is shown
in Figure 7.5, Figure 7.6, Figure 7.7, respectively. The 24h voltage profile per phase of buses 25, 34, and 55
considering the load profile of Figure 7.4 is shown in Table 7.10, Table 7.11, Table 7.12, respectively. The
current magnitude deviation of the MV/LV transformer is shown in Figure 7.8. The 24h profile of the
outgoing current from the MV/LV transformer is presented in Table 7.13.
10 8 9
1 2 4 6 11 13 15
3 5 7 12 14 16
20 23 26
17 18 21 24
19 22 25
27
28
29
20 kV
301
30 31 33 34 36 39
32 35 37
40 42 38
41 53
43 44 46 50 55
54
0.4 kV
15 49 48 51 56
47
52
Figure 7.3 Meltemi distribution network
1.00
1,00
0.80
0,80
0.60
0,60
(p.u.)
0.40
0,40
0.20
0,20
0.00
0,00
0 4 8 12 16 20 24
Hour
Figure 7.4 Load profile (p.u.)
Table 7.7 Data of the Meltemi distribution network
From To Type
conductor conductor ID Sequence (m)
266,800 CLASS
1 2 Overhead 2 7 STRD Cu 500 ABCN 32,49
A AA
2 3 Overhead 3/0 6/1 ACSR 1 CLASS A AA 500 ABCN 37,30
8 9 Overhead 6 AWG SLD Cu 6 AWG SLD Cu 510 AN 28,00
8 10 Overhead 6 AWG SLD Cu 6 AWG SLD Cu 510 BN 15,65
13 14 Overhead 6 AWG SLD Cu 6 AWG SLD Cu 510 CN 22,68
266,800 CLASS
A AA
266,800 CLASS
A AA
1 30 Overhead 266,800 CLASS 2 7 STRD Cu 500 ABCN 63,45
A AA

266,800 CLASS
A AA
266,800 CLASS
A AA
2. TRANSFORMERS
Rated Nominal voltage Nominal voltage
R X
From To Connection Capacity of the primary of the secondary
(%) (%)
DeltaGrounded
301 1 50 20 0.40 1.03 3.86
Wye
3. LOADS
Bus Connection Type
3 Grounded Wye Constant Power 3.08 + j1.01 0.544 + j0.18 1.2 + j0.39
5 Grounded Wye Constant Impedance 0.45 + j0.15 0.96 + j0.32 0.96 + j0.32
7 Grounded Wye Constant Current 0.91 + j0.3 1.12 + j0.37 0.624 + j0.21
9 Grounded Wye Constant Power 0.8 + j0.26 0 0
10 Grounded Wye Constant Power 0 0.8 + j0.26 0
14 Grounded Wye Constant Impedance 0 0 0.4 + j0.13
16 Grounded Wye Constant Current 0 0 0.8 + j0.26
29 Grounded Wye Constant Current 0.8 + j0.26 0 0
35 Grounded Wye Constant Impedance 0.8 + j0.26 0 0
38 Grounded Wye Constant Power 0 0 0.4 + j0.13
4. SWITCHES
From To Status
33 54 Open
Table 7.8 Bus voltages of the Meltemi distribution network during peak load demand
301 20000 0.00 20000 -120.00 20000 120.00
1 223.63 -32.34 224.41 -152.05 224.20 87.87

2 223.32 -32.40 224.30 -152.10 224.03 87.84
3 222.98 -32.47 224.41 -152.10 223.87 87.85
4 222.06 -32.42 222.03 -152.42 223.01 87.55
5 222.04 -32.42 221.92 -152.43 222.93 87.53
6 221.79 -32.43 221.63 -152.49 222.86 87.51
7 221.70 -32.44 221.55 -152.50 222.84 87.50
8 221.60 -32.44 221.54 -152.52 222.91 87.52
9 221.44 -32.45 0.00 -152.52 0.00 87.52
10 0.00 -32.44 221.45 -152.53 0.00 87.52
11 221.84 -32.42 221.50 -152.49 222.75 87.48
12 0.00 -32.42 221.40 -152.49 0.00 87.48
13 221.99 -32.43 221.45 -152.45 222.39 87.41
14 0.00 -32.43 0.00 -152.45 222.33 87.40
15 222.03 -32.44 221.44 -152.44 222.30 87.39
16 0.00 -32.44 0.00 -152.44 222.15 87.38
17 223.48 -32.37 224.31 -152.08 224.04 87.84
18 221.68 -32.65 223.01 -152.25 222.15 87.56
19 221.36 -32.69 222.88 -152.28 221.94 87.54
20 221.52 -32.67 222.88 -152.27 222.01 87.53
21 221.36 -32.70 222.75 -152.28 221.73 87.49
22 221.03 -32.79 222.75 -152.27 221.23 87.44
23 220.99 -32.72 222.42 -152.35 221.63 87.47
24 221.24 -32.72 222.57 -152.29 221.49 87.45
25 221.24 -32.72 222.45 -152.30 221.35 87.41
26 221.14 -32.73 222.51 -152.31 221.42 87.44
27 223.56 -32.35 224.46 -152.05 224.19 87.88
28 223.12 -32.44 224.68 -152.07 224.15 87.93
29 223.01 -32.44 0.00 -152.07 0.00 87.93
30 223.46 -32.39 224.27 -152.07 223.92 87.81
31 222.61 -32.54 223.57 -152.14 222.65 87.61
32 222.61 -32.55 223.36 -152.14 222.39 87.55
33 222.17 -32.62 223.49 -152.16 222.34 87.59
34 222.00 -32.65 223.45 -152.17 222.22 87.58
35 221.87 -32.66 0.00 -152.17 0.00 87.58
36 221.82 -32.68 223.28 -152.19 221.97 87.54
37 221.84 -32.68 223.27 -152.19 221.92 87.53
38 0.00 -32.68 0.00 -152.19 221.84 87.53
39 221.39 -32.72 222.96 -152.25 221.71 87.50
40 223.37 -32.38 224.22 -152.10 223.99 87.82
41 223.28 -32.40 224.14 -152.11 223.91 87.81
42 223.30 -32.39 224.16 -152.11 223.92 87.81
43 223.11 -32.42 224.04 -152.16 223.90 87.79
44 221.51 -32.59 222.73 -152.39 222.82 87.61
45 221.44 -32.60 222.67 -152.39 222.76 87.60
46 219.29 -32.82 220.92 -152.71 221.40 87.38
47 218.54 -32.93 220.45 -152.79 220.83 87.30
48 218.42 -32.94 220.35 -152.81 220.73 87.28
49 218.27 -32.97 220.34 -152.82 220.66 87.28
50 218.97 -32.84 220.59 -152.77 221.25 87.34
51 218.84 -32.86 220.52 -152.78 221.16 87.33
52 218.75 -32.88 220.48 -152.79 221.10 87.32
53 218.85 -32.86 220.49 -152.78 221.15 87.33
54 218.96 -32.83 220.47 -152.79 221.31 87.34
55 218.94 -32.82 220.38 -152.81 221.35 87.34
56 0.00 -32.82 220.20 -152.82 0.00 87.34
Table 7.9 Line currents of the Meltemi distribution network during peak load demand
Bus
From To
301 1 104.27 -51.34 91.52 -171.07 94.78 68.85
1 2 20.74 -50.65 16.53 -170.63 15.40 69.39
1 17 26.79 -50.91 23.71 -170.49 27.87 69.28
1 27 3.10 -50.64 0.00 -66.24 0.00 174.44
1 30 8.55 -50.86 7.45 -170.39 10.52 69.34
1 40 6.84 -50.59 6.84 -170.30 6.84 69.61
1 43 22.18 -51.07 22.99 -170.94 19.72 69.15
2 4 8.43 -50.63 14.37 -170.68 10.62 69.27
2 3 12.31 -50.66 2.16 -170.30 4.78 69.66
4 6 6.73 -50.64 10.78 -170.70 7.01 69.24
4 5 1.70 -50.62 3.59 -170.62 3.60 69.34
6 11 0.00 65.08 3.22 -170.69 4.60 69.20
6 7 3.51 -50.63 4.34 -170.70 2.42 69.31
6 8 3.22 -50.64 3.22 -170.72 0.00 174.14
8 9 3.22 -50.65 0.00 0.00 0.00 0.00
8 10 0.00 0.00 3.22 -170.72 0.00 0.00
11 13 0.00 65.08 0.00 -66.61 4.60 69.20
11 12 0.00 0.00 3.22 -170.69 0.00 0.00
13 15 0.00 65.08 0.00 -66.61 3.10 69.19
13 14 0.00 0.00 0.00 0.00 1.50 69.21
15 16 0.00 0.00 0.00 0.00 3.10 69.19
17 18 26.79 -50.91 23.71 -170.49 27.87 69.28
18 21 15.68 -50.94 13.96 -170.50 18.07 69.24
18 19 5.15 -50.89 3.84 -170.48 3.85 69.34
18 20 5.95 -50.86 5.92 -170.47 5.94 69.34
21 24 6.20 -50.92 7.44 -170.50 9.53 69.23
21 22 5.97 -50.98 2.18 -170.46 6.12 69.25
21 23 3.51 -50.92 4.34 -170.54 2.42 69.28
24 25 1.69 -50.91 3.60 -170.49 5.67 69.22
24 26 4.51 -50.93 3.84 -170.50 3.86 69.24
27 28 3.10 -50.64 0.00 -66.22 0.00 174.46
28 29 3.10 -50.64 0.00 0.00 0.00 0.00
30 31 8.55 -50.86 7.45 -170.39 10.52 69.34
31 33 6.85 -50.89 3.84 -170.44 4.82 69.32
31 32 1.70 -50.74 3.61 -170.33 5.69 69.36
33 34 6.85 -50.89 3.84 -170.44 4.82 69.32
34 36 3.86 -50.92 3.84 -170.44 4.82 69.32
34 35 2.99 -50.85 0.00 0.00 0.00 0.00
36 37 0.00 65.03 0.00 -66.41 1.61 69.33
36 39 3.86 -50.92 3.84 -170.44 3.22 69.31
37 38 0.00 0.00 0.00 0.00 1.61 69.33
40 41 3.83 -50.59 3.82 -170.31 3.82 69.61
40 42 3.01 -50.59 3.02 -170.30 3.02 69.61
43 44 22.18 -51.07 22.99 -170.94 19.72 69.15
44 46 19.08 -51.11 19.89 -171.00 16.62 69.11
44 45 3.10 -50.79 3.10 -170.59 3.10 69.41
46 50 7.82 -51.06 10.03 -170.99 6.77 69.13
46 47 11.26 -51.15 9.87 -171.01 9.85 69.09
47 48 6.04 -51.14 5.99 -171.00 5.97 69.09
47 49 5.23 -51.16 3.88 -171.01 3.88 69.09
50 54 0.00 64.76 3.24 -171.01 0.00 173.89
50 51 4.89 -51.07 3.88 -170.99 3.87 69.13
50 53 2.93 -51.05 2.91 -170.98 2.90 69.13
51 52 4.89 -51.07 3.88 -170.99 3.87 69.13
54 55 0.00 64.76 3.24 -171.01 0.00 173.89
55 56 0.00 0.00 3.24 -171.01 0.00 0.00
1,00
1.00
Phase A
Voltage magnitude of Bus 25 (p.u.)
Phase B
0.99
0,99
Phase C
0,98
0.98
0,97
0.97
0,96
0.96
0,95
0.95
00 44 88 12
12 16
16 20
20 24
24
Hour
Figure 7.5 Bus 25 voltage magnitude profile (p.u.) of the Meltemi distribution network
Table 7.10 Bus 25 voltage profile of the Meltemi distribution network

Hour |V|ph_A (V) ph_A (deg.) |V|ph_B (V) ph_B (deg.) |V|ph_C (V) ph_C (deg.)
0 226.82 -30.99 227.22 -150.85 226.82 89.04
1 227.52 -30.78 227.84 -150.66 227.52 89.25
2 228.22 -30.56 228.45 -150.48 228.22 89.46
3 228.22 -30.56 228.45 -150.48 228.22 89.46
4 227.83 -30.68 228.10 -150.58 227.82 89.34
5 227.02 -30.93 227.40 -150.79 227.02 89.10
6 225.18 -31.49 225.80 -151.27 225.19 88.56
7 224.35 -31.74 225.08 -151.48 224.36 88.33
8 223.52 -31.99 224.36 -151.69 223.53 88.09
9 222.67 -32.23 223.63 -151.90 222.70 87.85
10 222.32 -32.39 223.37 -152.02 222.40 87.72
11 225.18 -31.49 225.80 -151.27 225.19 88.56
12 224.97 -31.55 225.62 -151.32 224.98 88.50
13 226.21 -31.18 226.69 -151.01 226.21 88.86
14 225.18 -31.49 225.80 -151.27 225.19 88.56
15 223.52 -31.99 224.36 -151.69 223.53 88.09
16 222.88 -32.17 223.81 -151.85 222.91 87.91
17 221.78 -32.56 222.91 -152.16 221.88 87.57
18 220.91 -32.82 222.17 -152.38 221.03 87.32
19 221.24 -32.72 222.45 -152.30 221.35 87.41
20 219.59 -33.22 221.06 -152.71 219.76 86.95
21 222.88 -32.17 223.81 -151.85 222.91 87.91
22 223.52 -31.99 224.36 -151.69 223.53 88.09
23 226.00 -31.24 226.51 -151.06 226.01 88.80
1,00
1.00
Phase A
Phase B
0,99
0.99
Phase C
0,98
0.98
0,97
0.97
0,96
0.96
0,95
0.95
0 4 8 12 16 20
20 24
Hour
0 227.07 -30.97 227.61 -150.80 227.13 89.10
1 227.72 -30.76 228.14 -150.63 227.77 89.30
2 227.02 -30.93 227.40 -150.79 227.02 89.10
3 228.37 -30.55 228.66 -150.45 228.40 89.50
4 228.00 -30.67 228.36 -150.55 228.04 89.38
5 227.26 -30.91 227.76 -150.75 227.32 89.16
6 225.57 -31.46 226.38 -151.20 225.67 88.66
7 224.81 -31.70 225.76 -151.40 224.92 88.44
8 224.04 -31.94 225.13 -151.60 224.18 88.21
9 223.26 -32.19 224.50 -151.80 223.42 87.99
10 222.98 -32.33 224.26 -151.91 223.16 87.87
11 225.57 -31.46 226.38 -151.20 225.67 88.66
12 225.38 -31.52 226.22 -151.25 225.48 88.60
13 226.51 -31.16 227.15 -150.95 226.59 88.94
14 225.57 -31.46 226.38 -151.20 225.67 88.66
15 224.04 -31.94 225.13 -151.60 224.18 88.21
16 223.45 -32.13 224.65 -151.75 223.61 88.05
17 222.49 -32.49 223.86 -152.04 222.69 87.73
18 221.70 -32.75 223.21 -152.25 221.93 87.50
19 222.00 -32.65 223.45 -152.17 222.22 87.58
20 220.50 -33.13 222.23 -152.56 220.78 87.15
21 223.45 -32.13 224.65 -151.75 223.61 88.05
22 224.04 -31.94 225.13 -151.60 224.18 88.21
23 226.33 -31.22 227.00 -151.00 226.40 88.88
1,00
1.00
Phase A
0,99
0.99 Phase B
Phase C
0,98
0.99
0.97
0,97
0,96
0.96
0,95
0.95
0,94
0.94
0 4 8 12 16 20
20 24
Hour

0 226.00 -31.03 226.49 -151.03 226.84 89.02
1 226.89 -30.80 227.27 -150.80 227.54 89.23
2 227.26 -30.91 227.76 -150.75 227.32 89.16
3 227.77 -30.58 228.04 -150.58 228.23 89.45
4 227.26 -30.71 227.60 -150.71 227.84 89.32
5 226.25 -30.96 226.71 -150.96 227.04 89.08
6 223.94 -31.54 224.69 -151.54 225.22 88.53
7 222.90 -31.79 223.79 -151.80 224.40 88.28
8 221.86 -32.04 222.87 -152.05 223.57 88.04
9 220.80 -32.30 221.95 -152.31 222.74 87.80
10 220.31 -32.48 221.56 -152.47 222.41 87.66
11 223.94 -31.54 224.69 -151.54 225.22 88.53
12 223.69 -31.60 224.47 -151.60 225.01 88.47
13 225.23 -31.22 225.82 -151.22 226.23 88.83
14 223.94 -31.54 224.69 -151.54 225.22 88.53
15 221.86 -32.04 222.87 -152.05 223.57 88.04
16 221.06 -32.24 222.18 -152.24 222.95 87.86
17 219.63 -32.65 220.97 -152.64 221.88 87.50
18 218.53 -32.92 220.02 -152.91 221.03 87.24
19 218.94 -32.82 220.38 -152.81 221.35 87.34
20 216.85 -33.34 218.58 -153.32 219.74 86.85
21 221.06 -32.24 222.18 -152.24 222.95 87.86
22 221.86 -32.04 222.87 -152.05 223.57 88.04
23 224.98 -31.28 225.60 -151.28 226.03 88.77
120 Phase A
Current throughthe MV/LV S/S (A)
100 Phase B
Phase C
80
60
40
20
0
0 4 8 12 16 20 24
Hour
Figure 7.8 Bus 1 outgoing current magnitude profile (p.u.) of the Meltemi distribution network
Table 7.13 Bus 1 outgoing current profile of the Meltemi distribution network
Hour |I|ph_A (A) ph_A (deg.) |I|ph_B (A) ph_B (deg.) |I|ph_C (A) ph_C (deg.)
0 32,68 -49,16 28,84 -169,08 29,97 70,88
1 25,48 -48,95 22,50 -168,89 23,39 71,09
2 30,62 -49,10 27,03 -169,03 28,09 70,94
3 18,31 -48,73 16,18 -168,69 16,82 71,29
4 22,41 -48,86 19,79 -168,80 20,57 71,18
5 30,62 -49,10 27,03 -169,03 28,09 70,94
6 49,25 -49,65 43,41 -169,53 45,08 70,42
7 57,59 -49,89 50,74 -169,76 52,66 70,19
8 65,97 -50,13 58,09 -169,98 60,26 69,96
9 74,40 -50,37 65,46 -170,20 67,89 69,73
10 77,58 -50,53 68,24 -170,33 70,76 69,60
11 49,25 -49,65 43,41 -169,53 45,08 70,42
12 51,33 -49,71 45,24 -169,59 46,97 70,36
13 38,87 -49,34 34,29 -169,25 35,63 70,71
14 49,25 -49,65 43,41 -169,53 45,08 70,42
15 65,97 -50,13 58,09 -169,98 60,26 69,96
16 72,29 -50,31 63,62 -170,15 65,98 69,79
17 82,88 -50,69 72,87 -170,48 75,54 69,46
18 91,40 -50,95 80,31 -170,72 83,22 69,22
19 88,20 -50,85 77,52 -170,63 80,34 69,31
20 104,27 -51,34 91,52 -171,07 94,78 68,85
21 72,29 -50,31 63,62 -170,15 65,98 69,79
22 65,97 -50,13 58,09 -169,98 60,26 69,96
23 40,94 -49,40 36,12 -169,31 37,51 70,65
7.2 APPENDIX B
0.762 m 1.372 m 2.134 m
1.524 m
1.219 m 1.219 m
0.154 m 0.154 m 0.154 m
7.315 m 7.315 m 7.315 m
ID - 500 ID - 505 ID - 510
Figure 7.9 Overhead distribution line spacing [58]
0.1524 m 0.1524 m 0.0254 m
ID - 515 ID - 520
Figure 7.10 Underground distribution line spacing [58]
Table 7.14 Conductor data [58]

Size Stranding Material Diam (m) GMR (m) R (/km) Capacity (A)
4/0 6/1 ACSR 0.0143 0.00248 0.953 480
250,00 CON LAY AA 0.0144 0.00043 0.660 350
336,400 26/7 ACSR 0.0183 0.00701 0.550 490
556,500 26/7 ACSR 0.0236 0.00954 0.116 730
1/0 ACSR 0.0101 0.00136 0.696 230
1/0 7 STRD Cu 0.368 0.01113 0.607 310
3/0 6/1 ACSR 0.0128 0.00183 0.449 300
6 AWG SLD Cu 0.0041 0.00160 1.485 120
2 7 STRD Cu 0.0074 0.00269 0.599 230
1 CLASS A AA 0.0083 0.00302 0.761 177
7.3 APPENDIX C
Table 7.15 Sub-metering data as OGEMA representation

Data point Type OGEMA Resource Path
of
data
Total active power Float AirConditioner/subMetering/connection/energySensor/reading
(consumption
positive)
Total positive active Float AirConditioner/subMetering/consumptionMeter/energyReading
energy
Positive active energy Float AirConditioner/subMetering/consumptionInTariffX/energyReading
in tariff x
Total negative active Float AirConditioner/subMetering/generationMeter/energyReading
energy
Total positive reactive Float AirConditioner/subMetering/capacitiveMeter/energyReading
energy
Total negative Float AirConditioner/subMetering/inductiveMeter/energyReading
reactive energy
Total positive active Sche- Example assuming storage of daily maximum values:
maximum demand dule
AirConditioner/subMetering/consumptionMeter/
of energyReading/maximumTotalConsumption/dayValue
Float
Instantaneous current Float AirConditioner/subMetering/connection/
phase L1 (L2, L3) subPhaseConnections/subPhaseConnections_0/currentSensor/reading
Instantaneous voltage Float AirConditioner/subMetering/connection/
phase L1 (L2, L3) subPhaseConnections/subPhaseConnections_0/voltageSensor/
reading
Frequency (optional) Float AirConditioner/subMetering/connection/frequencySensor/reading
Recorded load Sche- Example assuming 15-minute-values in the profile for total active
profiles dule power (including consumption and generation aggregated):
of AirConditioner/subMetering/connection/energySensor/reading/
Float loadProfile/fifteenMinuteValue
Active power phase Float AirConditioner/subMetering/connection/subPhaseConnections/
L1 (L2, L3) subPhaseConnections_0/powerSensor/reading
Reactive power phase Float AirConditioner/subMetering/connection/
L1 (L2, L3) subPhaseConnections/subPhaseConnections_0/reactivePowerSensor/
reading
Asymtric voltage: Float AirConditioner/SubMetering/connection/extendedData/
Zero Sequence rms asymetricVoltageRMS/reading
(optional)
Asymetric current: Float AirConditioner/SubMetering/connection/extendedData/
Zero sequence rms asymetricCurrentRMS/reading
(optional)
Voltage phase angle Float AirConditioner/SubMetering/connection/extendedData/
deltas L1-L2, (L1-L3) u1_u2_angle/reading
(optional)
Voltage-current phase Float AirConditioner/SubMetering/connection/subPhaseConnections/
angle L1, (L2, L3) subPhaseConnections_0/reactiveAngleSensor/reading
(optional)
Table 7.16 Sample inverter as OGEMA representation

Data point Type OGEMA Resource Path
of
data
Total active power Float SampleBatteryInverter/subMetering/connection/energySensor/reading
(consumption
positive)
Total positive active Float SampleBatteryInverter/subMetering/consumptionMeter/energyReading
energy
Positive active energy Float SampleBatteryInverter/subMetering/consumptionInTariffX/energyReading
in tariff x
Total negative active Float SampleBatteryInverter/subMetering/generationMeter/energyReading
energy
Total positive reactive Float SampleBatteryInverter/subMetering/capacitiveMeter/energyReading
energy
Total negative Float SampleBatteryInverter/subMetering/inductiveMeter/energyReading
reactive energy
Total positive active Sche- Example assuming storage of daily maximum values:
maximum demand dule SampleBatteryInverter/subMetering/consumptionMeter/
of
energyReading/maximumTotalConsumption/dayValue
Float
Instantaneous current Float SampleBatteryInverter/subMetering/connection/
phase L1 (L2, L3) subPhaseConnections/subPhaseConnections_0/currentSensor/reading
Instantaneous voltage Float SampleBatteryInverter/subMetering/connection/
phase L1 (L2, L3) subPhaseConnections/subPhaseConnections_0/voltageSensor/ reading
Frequency (optional) Float SampleBatteryInverter/subMetering/connection/frequencySensor/reading
Recorded load Sche- Example assuming 15-minute-values in the profile for total active power
profiles dule (including consumption and generation aggregated):
of SampleBatteryInverter/subMetering/connection/energySensor/reading/
Float loadProfile/fifteenMinuteValue
Active power phase Float SampleBatteryInverter/subMetering/connection/subPhaseConnections/
L1 (L2, L3) subPhaseConnections_0/powerSensor/reading
Reactive power phase Float SampleBatteryInverter/subMetering/connection/
L1 (L2, L3) subPhaseConnections/subPhaseConnections_0/reactivePowerSensor/
reading
Asymtric voltage: Float SampleBatteryInverter/SubMetering/connection/asymetricVoltageRMS/
Zero Sequence rms reading
(optional)
Asymetric current: Float SampleBatteryInverter/SubMetering/connection/asymetricCurrentRMS/
Zero sequence rms reading
(optional)
Voltage phase angle Float SampleBatteryInverter/SubMetering/connection/u1_u2_angle/reading
deltas L1-L2, (L1-L3)
(optional)
Voltage-current phase Float SampleBatteryInverter/SubMetering/connection/subPhaseConnections/
angle L1, (L2, L3) subPhaseConnections_0/reactiveAngleSensor/reading
(optional)

D8.1 PU Design and Specifications of Stable and Secure Distribution Grids - Compressed

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

D8.1 PU Design and Specifications of Stable and Secure Distribution Grids - Compressed

Caricato da

Copyright:

Formati disponibili

Ref.

Title: Document Version:

Project Number: Project Acronym: Project Title:

Contractual Delivery Date: Actual Delivery Date: Deliverable Type*-Security*:

M19 (July 2016) M19 (July 2016) R-PU

Responsible Organization Contributing WP

Jovica Milanovi UNIMAN WP8

ETRA I+D ETRA INVESTIGACIN Y DESARROLLO, S.A.

Revision Date Description Author (Organisation)

V0.1 15.04.2016 First draft of deliverable ToC Jovica Milanovi (UNIMAN)

V0.2 19/04/2016 Updated ToC Marcel Eijgelaar(DNV), Kostas

v.0.3 30/06/2016 Eng contribution to USEF-NG adaptation Pietro Fragnito (ENG),

V0.6 05/07/2016 State estimation, Load forecasting Huilian Liao (UNIMAN)

V0.8 28/07/2016 Final version Huilian Liao (UNIMAN)

V1.0 01/08/2016 Final version for submission

D8.1 Design and specifications of Stable and secure distribution grids 2

2 DATA ACQUISITION AND INFORMATION EXCHANGE ................................................. 14

3 DATA ANALYSIS ........................................................................................................ 26

4 FUNCTIONALITIES AND TOOL DEVELOPMENT ............................................................ 42

D8.1 Design and specifications of Stable and secure distribution grids 3

5 INTEGRATION OF FUNCTIONALITIES IN GRID MANAGEMENT .................................... 89

D8.1 Design and specifications of Stable and secure distribution grids 4

6 REFERENCES AND ACRONYMS................................................................................. 133

7 APPENDIX ............................................................................................................... 142

D8.1 Design and specifications of Stable and secure distribution grids 5

D8.1 Design and specifications of Stable and secure distribution grids 6

D8.1 Design and specifications of Stable and secure distribution grids 7

D8.1 Design and specifications of Stable and secure distribution grids 8

D8.1 Design and specifications of Stable and secure distribution grids 9

D8.1 Design and specifications of Stable and secure distribution grids 10

D8.1 Design and specifications of Stable and secure distribution grids 11

1.2 SCOPE OF THE DELIVERABLE

1.3 STRUCTURE OF THE DELIVERABLE

D8.1 Design and specifications of Stable and secure distribution grids 12

D8.1 Design and specifications of Stable and secure distribution grids 13

DRFM for Aggregator #1

D8.1 Design and specifications of Stable and secure distribution grids 14

D8.1 Design and specifications of Stable and secure distribution grids 15

Generation Transmission Distribution DER Customer Premise

D8.1 Design and specifications of Stable and secure distribution grids 16

Enterprise 28 Distribution 43 ESP

31 ISO/RTO 27 DMS 25 Distributed 33 MDMS

D8.1 Design and specifications of Stable and secure distribution grids 17

Framework connections Logical Interfaces Interface Categories

Framework Connections High-level Security Requirement

D8.1 Design and specifications of Stable and secure distribution grids 18

2.3 DEVELOPMENT OF COMMON INFRASTRUCTURE FOR INFORMATION EXCHANGE

D8.1 Design and specifications of Stable and secure distribution grids 19

Electrical System GUI

wrapper wrapper wrapper wrapper

Nobelgrid ESB middleware

Figure 2.4 ESB overall connectivity diagram

D8.1 Design and specifications of Stable and secure distribution grids 20

Figure 2.5 Basic ESB communication schema

Figure 2.6 ESB subscribe

D8.1 Design and specifications of Stable and secure distribution grids 21

Figure 2.7 ESB publish

(5) Data results Nobelgrid ESB

Figure 2.8 ESB client app with local database

D8.1 Design and specifications of Stable and secure distribution grids 22

(7) store Nobelgrid ESB

(3) DLMS message for

Figure 2.9 MDM - DLMS integration

Contractual Delivery Date: Actual Delivery Date: Deliverable Type-Security: