CIGRE SCIENCE

& ENGINEERING
Volume No.14, June 2019

Innovation in the
Power Systems
industry
Enginers and specialists worldwide exchange
information and state-of-the-art world practices
to enhance knowledge related to power systems in
CIGRE’s latest publication.

SC C2: Power system restoration - World practices

& future trends
SC C4: The use of battery energystorage systems
for system integrity protection schemes in South
Australian power systems

Best young engineers papers from the CIGRE IEC

2019 Symposium in Hakodate, Japan Conference

CIGRE 21, rue d’Artois, 75008 Paris – ISSN: 1286-1146

 Volume N°14, June 2019

Journal edited by CIGRE

President Rob STEPHEN
Vice President Technical Marcio SZECHTMAN
Secretary General Philippe ADAM / philippe.adam@cigre.org

Editorial Committee
Chief Editor Konstantin O. PAPAILIOU
Editors Pierre ARGAUT Erli FIGUEIREDO
Christian FRANCK Susana Almeida de GRAAFF
Nikos HATZIARGYRIOU Xidong LIANG
Pouyan POURBEIK Rob STEPHEN
Copyright registration N° 014, June 2019
1286-1146 CIGRÉ
© Copyright CIGRE
Conseil International des Grands Réseaux Electriques
International Council on Large Electric Systems
21 rue d’Artois - 75008 Paris - France
Tel : 33 (0)1 53 89 12 90
http://www.cigre.org
"The author(s) of papers published in the CIGRE Science and Engineering Journal, are
granted permission to post a PDF copy of their paper, as published in the journal, for self-
archival purposes on their own personal websites, or similar outlets (e.g. researchgate.com).
This permission is granted with the following restrictions, requirements, and understanding:
1. CIGRE is the copyright owner of all articles. Please see the CIGRE copyright statement for
details. The articles cannot be disseminate for commercial use (i.e. for payment, etc.).
2. A clear reference must appear where it is posted stating that "The paper may also be
downloaded for free from the source CIGRE. Goto: http://www.cigre.org/Menu-links/
Publications/CIGRE-Science-Engineering. Then click to download the <Month/Year> Issue
of the CIGRE Science and Engineering Journal. The paper appears on pages <page numbers>.
3. The paper can only be posted on the author(s) website after it has been published and
released through the CIGRE website."

Submission Procedure https://www.cigre.org/GB/publications/cigre-cse

Dear readers,

With the longest day of the year (21st June) ante portas (in the northern hemisphere where I live) and
a prolonged period of nice sunny weather, I have to admit it is difficult to concentrate on reading the
12 excellent articles in the present issue of CIGRE Science & Engineering. The chances for our readers in
the southern hemisphere are much better since there, the longest night of the year is soon marching it
And believe you me it is more than worth! The present issue namely is a very nice mosaic of papers
displaying impressively the variety of knowledge sharing which makes CIGRE so unique.
To start with this issue contains two top Study Committee papers presenting the state of the art
on “Power system restoration – World practices & Future trends” from SC C2 and “The use of
battery energy storage systems for system integrity protection schemes in South Australian Power
System” from SC C4.
The second group contains five innovative papers on rather scarce issues such as “Wire
displacement of AAAC overhead line conductors at wedge tension clamp” and “Seismic response of
high-voltage cables in ducts – A preliminary study” as well as papers on “An aggregate dynamic model
for distributed energy resources for power system stability studies”, “A novel Wavelet selection
scheme for partial discharge signal detection under low SNR condition” and “Real-time co-
simulation model using electromagnetic transient and dynamic phasor simulations”.
It is followed by two papers, “IDtools: An automated tool for modal identification from time-domain
simulation results for establishing system operating limits” and “A direct calculation of locational
marginal value of distributed energy resources”, which have been selected as excellent papers from the
“The Grid of the Future” Symposium, sponsored by the CIGRE US National Committee (USNC) and
the Electric Power Research Institute (EPRI), and with the theme “Technology for the 21st Century
electric utility”, had been very successfully held on October 28 - 31, 2018 in Reston, Virginia, USA.
And last but not least, this issue includes the Grand Prize award - for authors under 35 years old
- for “Effect of Earth Surface on lightning electromagnetic pulse propagation” and the two
outstanding performance awards, i.e. “Material and emerging test techniques general overview of
AC and DC current injection on high voltage potential for HVDC long-term tests” and
“Application strategies for externally gapped line arresters against lightning outages on the 400 kV
overhead transmission lines in Hong Kong”, all three papers presented at the CIGRE-IEC 2019
Conference on EHV and UHV (AC & DC), which was held in Hakodate, Japan from April 23rd to
26th and which I had the honor to chair. The Hakodate Conference has been in my view one of the
highlights of this year with eight Study Committees and more than 330 participants from 27 countries,
which were more than impressed by the excellent organization and the unparalleled Japanese
hospitality. More details on this venue can be found in Electra No 304, June 2019.

Cigre Science & Engineering • N°14 June 2019

3
 Before closing it is my great pleasure to announce the appointment of Dr.
Susanna Almeida de Graaff as a new Member in our Editorial Board. Susana, a dear
colleague, who has dedicated her professional life to power system operations.
She received her Licenciatura (5 years degree), M.Sc and PhD degrees in Electrical
Engineering from Porto University, Portugal, respectively in 2000, 2006 and
2010. Her PhD work «Portuguese transmission grid incidents - Risk assessment”
was approved with distinction.
She was with REN, the Portuguese TSO, since the beginning of 2001, with System Operations in
the development of activities such as network security analysis, online risk assessment and
disturbance/blackout analysis.
Since November 2010, she is with TenneT TSO B.V. in the Netherlands, in International
Development within System Operations, focusing in innovation, research and development,
cooperation and coordination among TSOs, developing methodologies for capacity calculation,
network security on a European level, including R&D projects.
Dr. Susana Almeida de Graaff is the Chair of CIGRE Study Committee C2 on System Operations
and Control since August 2016, and contributing actively to CIGRE SC C2 activities since 2006, as
Working Group Member, Special Reporter and National Member and has received the following
prestigious awards:
• Hidde Nijland Prize in the Netherlands, in recognition for contributions to the development
of the electrical power system, December 2018.
• CIGRE Technical Committee Award, in recognition of contribution to the Study Committee C2,
2013
• REN prize with her M.Sc. thesis in electrical power systems, 2005

If not enough, it is a remarkable coincidence, that Susana is the coordinator and corresponding author
of the SC C2 paper featuring on the top of this issue.

Prof. Dr. Konstantin O. Papailiou

Welcome dear Susana on board and good luck,
Chief Editor
konstantin@papailiou.ch

Cigre Science & Engineering • N°14 June 2019

4
Contents of this issue

page

23

38
49
64

73

86

96
105
111

118

125

Cigre Science & Engineering • N°14 June 2019

5
Study Committee C2 – System operation
and control
Power system restoration –
World practices & future trends
F. CRISCI (AU), P. GOMES (BR), B. LI (CA), W. SATTINGER (CH), Y. FANG (CN),
C. ROGGATZ (DE), S. LEWIS (IE), R. PORWAL (IN), G. GIANNUZZI (IT), R. ZAOTTINI (IT),
S. ALMEIDA DE GRAAFF* (NL), V. SEWDIEN (NL), N. CUKALEVSKI (RS), T. RAMEY (US)
Restoration of the power system following an interruption
is an important aspect of a System Operator’s (SO) role in
managing the bulk power system. Electric power grids in
developed economies generally exhibit a very high degree
of reliability thanks to the well-established standards
and criteria for the design, planning, construction and
operations of the integrated network and the close
interconnection in certain continents or regions. Despite
prudent planning and operations, major interruptions to the
electric power grid (complete or partial blackout) do occur
from time to time due to events (disturbances) that either
exceed the basic design criteria, or due to various causes
such as natural disasters, multiple equipment failure,
protection relay miscoordination or malfunctioning, and
human errors.
When such disturbances occur, the power system may
experience wide-area, regional or local area blackouts
with or without damage to, or prolonged outages of,
major facilities. Restoring the integrity of the electrical
grid and supply to end-use customers is of paramount
importance to reduce and minimise undue hazard to
social welfare, public safety, infrastructure security, and
business activities.
To help restore the power system after major disturbances,
control centre’s operating personnel are trained and
provided with a set of guidelines and procedures, defining
the strategy, to placing top priority on restoring stable
operation of the power system with sufficient energized
corridors so that power supply to other generation
facilities and end-use customers can be restored as
expeditiously as possible to minimise interruptions to
social life and businesses. The key to successful and
expeditious restoration thus depends to a great extent on
the SO’s preparedness, which includes operator training,
availability of guidelines and procedure documents,
effective communication protocols, provision and
assurance of blackstart capability, verification of cranking
path’s sustainability, etc.
* susana.de.graaff@tennet.eu
Cigre Science & Engineering • N°14 June 2019
6
Overview of past major incidents
Depending on its operating point and defined constraints,
a power system finds itself in one of the predefined
states. Figure 1 [1] depicts possible operating states of the
power system, including different types of controls and
contingency effects that facilitate the transition between
these states (figure originally from [1]).
These states depend on the operational characteristics of
the power system and each jurisdiction can have its own
definition of these states. For example, the European
Network of Transmission System Operators for Electricity
(ENTSO-E) defined these states as follows [2]:
• Normal State: means a situation in which the system
is within operational security limits in the N-situation
and after the occurrence of any contingency from the
contingency list, taking into account the effect of the
available remedial actions.
• Alert State: means the system state in which the
system is within operational security limits, but
a contingency from the contingency list has been
detected and in case of its occurrence the available
remedial actions are not sufficient to keep the normal
state.
• Emergency State: means the system state in which
one or more operational security limits are violated.
• Blackout state (also in extremis state): means the
system state in which the operation of part or all of
the transmission system is terminated.
• Restoration state: means the system state in which
the objective of all activities in the transmission
system is to re-establish the system operation and
maintain operational security after the blackout state
or the emergency state.
The objectives of power system restoration (PSR) are to
enable the power system to return to normal conditions
(i.e. normal or alert operating state) securely and rapidly,
minimize restoration time and associated losses, and
diminish adverse impacts on society [3]. Despite having
 Normal State

By restorative By preventive Gradual Due to

control control security occurrence of
deterioration less severe
contingency
By restorative Due to
control occurrence of
Restoration State Alert State severe
contingency

By emergency
control Due to
occurrence of
more severe
In Extremis State contingency

Due to ineffective
or unavailable
emergency control
By emergency control
Emergency State

Figure 1 Power System Operating States [1]

the same objectives, a review conducted by the CIGRE the effectiveness and efficiency in system restoration [5].
C2.23 working group concluded that system restoration An overview of the past major incidents is given in Table 1.
preparedness may vary from one SO to another, or may
This paper presents the more commonly adopted
differ between interconnected systems due to differences
approaches for power system restoration and speculates
in system characteristics and/or market design/rules
on several innovative approaches to tackle specific
[4]. In addition, the lessons learnt from actual events
challenges encountered during system restoration.
suggest that there are challenges which would warrant the
exploration of modified or innovative ways to improve
Table 1 Overview of Past Major Incidents [5]

Date Location Lessons Learned (restoration related)

14 August 2003 Northeast USA/Canada Lack of testing and verification of:
• Blackstart capability
• Cranking path procedure1
4 November 2006 Europe Lack of tools to assist operators to effectively restore the system
10 November 2009 Mid-West and South of Brazil Lack of testing and verification of:
• Blackstart capability
• Cranking path procedure
4 February 2011 Northeast Brazil Lack of testing and verification of:
• Blackstart capability
• Cranking path procedure
8 September 2011 San Diego, USA Lack of tools (such as wide area observability) to assist operators to
effectively restore the system
30 July 2012 Northern India Lack of:
• dedicated communication infrastructure for the power system
• availability and preparedness of personnel at generation stations
and substations, leading to high start-up time of the restoration
• tools (such as wide area observability) to assist operators to
effectively restore the system
28 September 2016 South Australia • Changes in system dynamics with increased amounts of generation
from non-synchronous and inverter-connected plants
• Unknown critical settings of several generating systems

____________________
1- Cranking paths are transmission corridors that extend from blackstart generation units to the targeted facilities needing offsite power.

Cigre Science & Engineering • N°14 June 2019

7
1. International Experiences there was 330 MW of thermal generation, 833 MW of wind
generation and 613 MW import from the interconnector
In general, there are two basic strategies for power system
with neighbouring state Victoria. AEMO aims to restore
restoration, namely the bottom-up and the top-down
the power system to a secure operating state within 30
strategy.
minutes of either a credible or non-credible contingency
The bottom-up restoration strategy is based on use event. The system restart plan commenced at 16.30 hrs,
of blackstart generators (those able to re-energise the around 42 minutes after SA went black. However the first
system without relying on the external electric power power was not restored until 19.00 hrs, as explained in the
transmission network), and applies in case of total system restoration process below (Figure 2).
blackout and non-existent interconnection assistance.
Normally, blackstart units and selected UHV transmission
facilities are connected to create electrical islands that can
be further synchronized with similar islands to restore the
grid and gradually restore load.
On the contrary, the top-down restoration strategy is
based on neighbouring interconnections. These are used
to energize the bulk power transmission system first. Then
operators energize loads, start-up generation, balance
load and resynchronize areas to the backbone and finally
reconnect remaining loads. A more detailed overview of
these approaches is given in Appendix C of [4].
Due to many requirements like size, number and location
of the blackstart units, other base load units such as nuclear
power plants, points of interconnection, substation and
telecommunication equipment, communication protocols
and procedures, it is a regular practice to prepare
restoration plans in advance and to update these plans
regularly. These plans must comply with requirements
Figure 2 Restoration Process of AEMO
posed by regulators or similar authorities.
The first task in the restoration process is to identify the
Both approaches have their advantages and disadvantages,
state of the power system and to determine the extent
thus many system operators choose a hybrid approach to
of the disturbance. AEMO, in conjunction with the
restoration (i.e. combination of bottom-up and top-down
Transmission Network Service Providers (TNSPs) and
strategies), as the best fit to their conditions. Several real
generators, will determine:
life approaches are described in sequel.
• Which areas of the power system are blacked out.
Australia • Which areas of the power system have islanded
The security, stability and reliability of the Australian around generation.
National Electricity Market (NEM) is managed by the • Any requirements for stabilising the island.
Australian Energy Market Operator (AEMO). At one end • Where are the separation points in the system and
of this interconnected grid is the state of South Australia equipment that is not available.
(SA). In September 2016, the state suffered a blackout
After a major system disturbance, electrical islands can be
when at least seven tornadoes in SA’s mid north caused
formed. When the island(s) conditions have been assessed,
transmission line towers to collapse, interrupting supply
actions can be taken to stabilise their operation. This may
to 850,000 customers.
require additional load shedding or load reconnection to
At 15.48hrs (market time), just before the blackstart event, ensure adequate frequency and voltage control within

Cigre Science & Engineering • N°14 June 2019

8
the surviving island. Once any islanded generation has
been stabilised, AEMO, in conjunction with the TNSPs,
determines the restoration sequence. This may be in
accordance with the Restart Procedures of each region, or
may be varied depending on the state of the power system
[6].
Firstly, supply is provided to power station auxiliaries,
using either System Restart Ancillary Services or any
online generation. To do this, cranking paths are required
between the blackstart generation units and the auxiliaries
of receiving power stations.
The TNSPs and Distribution companies disconnect all
circuits and then re-energise the transmission network
with adequate load to ensure effective voltage and
frequency control.
Once the transmission corridors have been stabilised with
auxiliary supply, restoration of the remaining network
and loads commences, but will depend on how much
generation is available, and how much load the TNSPs
and Distribution companies can provide.
SA’s average demand is around 1500 MW, and the
instantaneous output of wind generation in SA has
exceeded 120% of demand at times, for example on 25
April 2017 [7]. This high penetration of non-synchronous
generation challenges power system security, particularly
system strength. Consequently, AEMO requires a
minimum amount of synchronous generation to be
connected in SA at all times, commensurate with the
quantity of renewable generation in service.
AEMO meets periodically with TNSPs and Distribution
companies to review Blackstart procedures, and
annually conducts training of AEMO and Transmission
and Distribution system operators, to reinforce those
procedures.
Brazil
The Brazilian Interconnected Power System’s (BIPS) supply the service and to guarantee that such service is
restoration philosophy is based on a two-stage process:
the fluent and the coordinated restoration process. The
main objective is to restore all transmission lines of 230
kV and above within 2 to 3 hours.
During the fluent restoration process, control and
switching actions are performed by Utilities Operators
or their control centres, according to pre-established
Cigre Science & Engineering • N°14 June 2019
9
procedures. This phase allows the restoration of important
load centres simultaneously and independently of other
areas, without the involvement of Operador Nacional do
Sistema Elétrico (ONS, the only operator of the BIPS),
minimizing coordination work and ensuring agility to the
process.
The fluent restoration can become coordinated whenever
difficulties appear, such as the unavailability of some
equipment. The BIPS contains 44 fluent restoration areas.
The coordinated restoration begins only after the Utilities
Operators communicate to ONS’ regional control centres.
In this phase Utilities Operators’ actions are coordinated
by the ONS’ regional and national control centres.
In Brazil, mainly hydro plants are currently used to
provide blackstart services. For new hydro power plants,
ONS decides whether or not they should be equipped
with blackstart capability based on technical analysis
conducted as part of connection studies.
ONS evaluates the need for additional cranking paths
using available blackstart hydro power plants (new or
old ones), according to the following implementation
strategy:
• For blackstart units not belonging to the existing
fluent restoration paths, new fluent restoration areas
must be created, or they must be incorporated to the
existing restoration paths;
• For new plants to be integrated to the BIPS, a
procedure must be elaborated to evaluate whether the
new generator should be required to have blackstart
capability.
Hydro power plants equipped with blackstart capability
must undergo restoration tests on a yearly basis, according
to a schedule coordinated by ONS. Owners of blackstart
hydro power plants must test their facilities’ blackstart
capability under realistic conditions. These tests are
carried out to certify that the plant is currently able to
maintained over time. In case of identifying problems in
an equipment during a blackout, it must undergo a series
of tests, which may result in the need to be either repaired
or replaced, followed by re-testing.
Before putting the restoration plan in effect, the following
studies are normally performed. Power flow studies
aim to assess the overall status of the power system
throughout all stages of the restoration process, checking
voltage profiles, transmission lines’ thermal limits
and generators’ reactive power limits before and after
switching operations. These studies check the feasibility
of a candidate restoration procedure considering the
standards defined by the grid code. These studies are
performed using a steady state analysis software suite,
which includes a module specially tailored for restoration
strategy evaluation. Electromechanical stability studies
check whether switching actions simulated during power
flow studies are feasible from a dynamic point of view.
Electromagnetic transient studies define maximum
voltages for equipment energization and assess the
possibility to energize equipment without risk of transient
overvoltage or system resonance. These studies focus on
short term transient conditions such as line or transformer
energization and load rejection that may lead to severe
equipment strain or damage and consequently delay the
restoration process.
Through its operational procedures ONS defines a set of
simulated trainings sessions (using an Operator Training
Simulator, OTS) that train and prepare operators for
the restoration process. These training sessions (drills)
happen at least once a year at each ONS control centre
(South, Southeast, Northeast, North-Middle East and
National) and involve ONS and operators of generation,
transmission and distribution companies.
Drills include at least one restoration area and are
perceived as an indispensable tool to ensure the technical
adequacy of teams, procedures and processes. Each
training session results in a technical report that points
out positive aspects, inadequacies and opportunities for
improvement that are treated as an action plan.
India
The restoration plans in India comprise of bottom-up as
well as top-down procedures, to be chosen depending on
the actual situation and locations of the blackout, with
the overall objective of faster restoration. The majority
of hydro power plants spread across the system and the
gas fired generation units are equipped with blackstart
capability. Gas generation units and hydro stations are
used for bottom-up approaches. The subsystems with
availability of blackstart units and away from the healthy
part of the power system are started according to the
bottom-up approach, while subsystems or islands with
Cigre Science & Engineering • N°14 June 2019
10
easy availability of a nearby healthy systems are restored
using the top-down approach. The restoration strategy is
executed using the following priorities:
• Extension of power for synchronizing islanded
nuclear power stations or start-up power to nuclear
power plants.
• Extension of start-up power to thermal power plants.
• Extension of start-up power to non self-starting hydro
power plants.
• Restoration of supply to other essential loads such as
hospitals.
• Restoration of traction power supply.
• Building up subsystems and synchronizing them
with each other.
• Restoration of power supply to communication
system nodes and repeater stations.
• Restoration of supply to all other customers.
The country’s power system is demarcated into five
Regions with each region containing the state/provincial
control areas. There are State, Regional and National
Load Despatch Centres that coordinate the restoration
process in their jurisdiction. Thus the restoration process
starts from many locations, independently, with active
coordination through regional and national control centres
for interconnections and synchronisations.
A disturbance in the Northern Region of India took place
at 02.33 hrs of 30 July 2012 [8]Eastern & North-Eastern
Region On 31st July 2012 (Aug. 16, in which almost
the entire Northern Region catering a load of about 38
GW was lost. The power was restored using the hybrid
approach.
The major issues faced during the restoration were:
• Real power balance and control of frequency,
including low inertia situations
• Reactive power balance and control of voltages,
including tripping of high voltage lines
• Protection systems and local controllers
In order to minimise the restoration time in the future,
regular actual mock exercises with blackstart units are
undertaken on a yearly basis. These exercises have three
main components:
• Creation of an electrical island with the blackstart
unit and testing the running of the unit in an islanded
environment.
 Figure 3 Illustration of an actual Indian blackstart exercise

• Creation of a blackout in the island and revival of the • Energise an isolated transmission test path to the
island according to the bottom-up approach. synchronisation point;
• Resynchronisation of the island with the main grid. • Manually synchronise the test system to the main
The steps of such an exercise are illustrated in Figure 3. transmission system. This is achieved by the controlled
synchronisation facility on the circuit breaker of a
The restoration trainings and associated debriefing
220 kV double busbar coupler, where one busbar is
sessions involve personnel from load dispatch centres,
energised by the blackstarted interconnector, and the
generating stations and distribution utilities and are being
other busbar energised by the transmission system;
organised for increased preparedness. This approach has
• Ensure pre-test conditions are met to protect the
shown to be helpful in keeping all stakeholders prepared
Irish transmission system from short circuiting or
and in reducing the time during actual restorations.
significant power flows in either direction.
Ireland Pre-Test Conditions
The East West Interconnector (EWIC) is a Voltage Source
After EWIC has successfully blackstarted, it operates
Converter (VSC) HVDC submarine cable connecting
in Islanded Mode, where the interconnector controls
Ireland to Great Britain. The interconnector employs
the voltage and frequency of the energised island. This
HVDC Light technology, which provides blackstart
network operation mode should not transfer to Normal
capability on both grids. EWIC is contracted by EirGrid,
Mode (power transmission mode) until the short circuit
the Irish TSO, to provide blackstart capability to the
level of the system it is connected with reaches 1800
Irish transmission system, and is required to perform one
MVA. This short circuit level is the lowest specified short
successful TSO requested local blackstart test per calendar
circuit capacity, determined by a dynamic performance
year. As part of the agreement, the TSO is also entitled to
study during the design phase of the interconnector. This
request EWIC to participate in a power system restoration
level is higher than what can be typically expected on the
test up to every five years, where EWIC will blackstart to
Irish transmission system in May, so additional generators
either provide a conventional generator with start-up load,
were requested to be synchronised in advance of the test.
or energise an isolated subsystem and synchronise with
the main transmission system. To assist with regulating the main system frequency for the
synchronisation of the test system and main system, it was
EWIC is one of two top-down blackstart sources for
agreed that the system should be synchronised after the
EirGrid; the other being the AC interconnection to SONI
morning demand peak had been reached at approximately
(Northern Ireland TSO). Bottom-up blackstart capability
10.00 hrs on the day of the test.
is provided by hydro, pumped storage and open cycle gas
turbine generators. Sequence of Events
1. HV switching to isolate a transmission test path from
Irish Restoration Test using VSC HVDC
HVDC converter station --> 400 kV transmission
In May 2018, EirGrid agreed a test plan with EIDAC
station --> 220 kV transmission station
(owner of EWIC) and the Distribution System Operator
2. AC supply to the converter station in Ireland is
(ESB Networks) with the following objectives:
disconnected
• Blackstart the interconnector locally from the 3. The low frequency trip level is artificially simulated
converter station blackout configuration; in the DC converter station

Cigre Science & Engineering • N°14 June 2019

11
4. DC converter station trips to blackstart configuration.
UPS batteries provide auxiliary power supply to
converter station, whilst on-site back-up diesel
generator is run up to then take over the auxiliary
power supply from the batteries
5. EWIC is energised at rated DC voltage of ±200 kV in
“Islanded Mode”, and Great Britain is now providing
the converter station with auxiliary power supply
6. Power converted to AC, and HV switching to energise
busbar in 400 kV transmission station to target
voltage. Target voltage achieved within contracted
blackstart achievement time of one hour
7. HV switching to energise path to busbar in 220 kV
transmission station
8. NCC (EirGrid’s National Control Centre) confirms
short circuit level of the transmission system is
greater than 1800 MVA. The NCC has an online
short circuit study tool to calculate this
9. NCC confirms voltage and phase angle of both
busbars in 220 kV transmission station are matched
within synchronisation tolerance
10. NCC controls transmission system frequency to
within the range 50 – 50.05 Hz
11. 220 kV busbar coupler circuit breaker closed
successfully
12. The transmission system frequency is marginally
above 50 Hz, resulting in a power flow of
approximately 20 MW in the GB direction due to
the EWIC “island” attempting to govern the system
frequency to 50 Hz
13. EWIC is promptly returned to normal mode
following synchronisation and ordered to ramp to 0
MW flow. EWIC has been successfully blackstarted
and synchronised to the transmission system
Test Findings
The recent test resulted in two main conclusions.
Firstly, although the 1800 MVA short circuit level for
EWIC to be transferred from islanded mode to normal
mode was achieved for the day of the test, this could be
difficult to achieve in a blackstart scenario. In this scenario,
EWIC will provide load restoration in the East region of
the grid, whilst bottom-up blackstart sources will attempt
to restore load in the other three regions of the grid. It
may not be possible to restore 1800 MVA of generators
in the EWIC region before transferring to normal mode
Cigre Science & Engineering • N°14 June 2019
12
to allow synchronisation with the other regions, so it has
been recommended that a new study be made into the
short circuit level required to transfer to normal mode in a
blackstart scenario.
Secondly, small deviations in frequency when
synchronising the systems can cause a significant power
flow in both directions, so when synchronising EWIC
after a blackstart the frequency tolerance should be smaller
than the standard controlled synchronising settings.
Italy
The top-down restoration strategy implemented in Italy
uses the support of the 380 kV transmission system of
neighbouring power systems. Its purpose is to reach, as
fast as possible after a blackout occurs, the biggest thermal
power generators. The main difficulty in performing
such a strategy is the distance of these generators, which
are located in the South of Italy, far from the main part
of the interconnected European power system. The
execution time of the top-down strategy has been reduced
thanks to close coordination with neighbouring TSOs,
the installation of shunt reactors in 380 kV nodes for
voltage control, and to the online SCADA and WAMS
measurements.
Until now, the Italian TSO Terna performed 5 real top-
down restoration tests with the collaboration of RTE
(TSO of France) and Swissgrid (TSO of Switzerland), see
Table 2.
Table 2 Overview of Italian Restoration Tests
Test Year Cranking Path Cranking Path
# Length
1. 2012 France – Valmontone 940 km
(see Figure 4)
2. 2013 Switzerland – Presenzano 1000 km
3. 2014 France – Andria 1300 km
4. 2015 Switzerland – Brindisi 1400 km
5. 2017 France – Brindisi 1450 km
Real tests are very important in order to check the
technical feasibility of and evaluate the timing of the
preparation and execution of re-energization paths. To
this end, the training of operators plays a crucial role for a
successful restoration. Commitment among the involved
stakeholders is of utmost importance for restoration
tests in order to reproduce, as much as possible, credible
 Figure 4 France – Valmontone Top Down Restoration Test (940 km)
emergency conditions, assuring at the same time the
absence of impact on final users.
Before carrying out a real restoration test, it is necessary The electromechanical analyses are divided into different
to perform detailed analyses. The use of offline steady-
state and dynamic simulators of power network behaviour
proves itself to be crucial for verifying the effectiveness
of current and future plans. Electromechanical and
electromagnetic analysis are equally important in order
to analyse the performance of machines (including their
controls) on the one hand, and on the other hand to check
for possible critical effects due to for instance in-rush
currents, mutual couplings or resonance effects. Such
simulations require detailed dynamic models and careful
parameter settings. This work is carried out through
data sheet analysis and real measurements processing,
respectively before and after the actual restoration test on
field.
Italy-France Restoration Test
In 2012 Terna and RTE conducted the first European
top-down restoration real test on a path of about 940 km
starting from France (France to Valmontone, shown in
Figure 4. In about 25 minutes the total path was energized,
considering the synchronisation of 3 hydro units near the
French electrical area, with a total of 5 switching steps
implemented by Terna.
Electromagnetic analyses were conducted in advance
to verify possible critical aspects such as transient over-
voltages, circuit breaker transient recovery voltage and
rate of rise of restricting voltage values in case of fault
opening during the energization process. Also resonance
Cigre Science & Engineering • N°14 June 2019
13
effects were investigated by unbalanced load-flow and
frequency-domain impedance analysis.
steps. The first step is the collection of data. The Power
Network Simulator contains complete models of the
Italian and part of the French transmission system,
including dynamic models of the generators (including
capability curves, frequency and voltage regulation,
power system stabilizer, real load ramp behaviour and
synchronization devices) on the cranking path. For other
generators in Italy and neighbouring systems that are not
involved in the restoration test, the simulator contains
simplified models.
The second step is to verify, together with the involved
stakeholders, the feasibility of a real test. The third step
is the selection of the final test configuration. In this step
certain protection settings are altered in order to avoid
situations that can prevent the system from returning to
normal operation.
The full cranking path is modelled with the representation
of Marti’s model lines (frequency dependent line model),
of different tower line types, including mutual coupling
effects. Transmission lines that are not part of the cranking
path, but cause coupling effects with the cranking path are
also modelled. The tower type has been represented by
the geometrical parameters and physical characteristics
of phase and earth conductors (section, material, bundle
spacing, etc.) and of the ground (soil conductivity).
Attention has been given to phasing sequence
representation so that transposition can be considered.
 242
10:25 a.m.: CIRCUIT BREAKER
241
LACCHIARELLA-CHIGNOLO LINE CLOSED
240
10:27 a.m.: CIRCUIT BREAKER 10:47 a.m.: OUTAGE OF
239 MARTIGNONE-BARGI LINE CLOSED PRESENZANO GROUP
238

237

236
10:57 - 10: 59:
235 REMESHING ACTIONS

234

233

232
10:41 a.m.:
231 SYNCHRONIZATION OF
PRESENZANO GROUP
230 10:31 - 10:32: END OF THE TEST
229 VARIATION OF TAP
CHANGER LAVORGO
228 TRANSFORMER 10:34 a.m.: CIRCUIT
BREAKER SUVERETO-
227 MONTALTO LINE CLOSED
226 10:22 a.m.: CIRCUIT BREAKER 10:51 a.m.: CIRCUIT
MUSIGNANO-LAVORGO LINE CLOSED BREAKER SUVERETO-
225
STARING POINT OF THE TEST MONTALTO LINE OPENED
224

Phase-to-ground Voltage Phase-to-ground Voltage Phase-to-ground Voltage

Figure 5 Musignano Phase Voltages during Execution of Restoration Test

Non-linear characteristics of the shunt reactors and step- USA

up transformer magnetization were also represented. The
main results of the line energization were the line inrush
currents and the transient busbar overvoltage. At the end
of the lines, where a shunt reactor is located, in the 1st
cycle the peak voltage in one phase can reach up to twice
the nominal voltage.
Italy-Switzerland Restoration Test
Another top-down restoration test was performed in 2013,
this time by Terna and Swissgrid. The main differences
with respect to the previous test from France were:
• the use of only shunt reactors in order to reach the end
of the approx. 1000 km cranking path (Presenzano
station, south of Rome)
• first successful test of synchronization of a power
plant at 1000 km from Switzerland’s network
In this test, the path of 1000 km was energized in 13
minutes with 4 switching actions implemented by Terna.
It took another 4 minutes to synchronize the unit of
Presenzano. This was possible due to a pre-configured/
compensated transmission path divided in 4-5 main
sections. Each section consists of several substations and
related transmission lines which were energized once after
the other.
The measurement recordings confirmed transients fully in
line with offline studies; the protection and control system
were stable and fully in line with expectations. Figure 5
shows the phase-ground voltages for the entire execution
of the test as seen from Musignano (first Italian substation).
In the USA, restoration plans are coordinated and drilled
on with neighbouring SOs and other stakeholders (e.g.
impacted industries and governmental agencies). SOs
are expected to follow policies and procedures to ensure
the operational reliability of the interconnections. These
include reliability analyses and identifying special
operating procedures, analysing current operating
conditions, and implementing procedures to mitigate
operating limit violations on the transmission system.
Hurricanes and other weather related disturbances
have allowed SOs in the United States the opportunity
to practice system restoration plans on a small scale.
However, the United States northeast blackout of 2003 was
the most recent event that required SOs to implement their
restoration plans on a large scale. Following a disturbance
in the state of Ohio, equipment malfunctions and a loss of
situational awareness contributed to a cascading outage
that ended up impacting over 10 million people in Canada
and over 45 million people in 8 US states.
This northeast blackout and small scale restoration events
allow SOs to improve existing restoration plans. These
should take into account the fact that transportation
pathways that are normally available, may not be clear in
the event of a large scale interruption, thus impacting the
restoration time. Currently, many restoration plans still
utilize ground crews to physically assess the damage at
sites. However, with the development of new technologies,
many SOs are utilizing drones to patrol their transmission

Cigre Science & Engineering • N°14 June 2019

14
corridors in order to assess the damage. Using drone
technology has significantly decreased the amount of time
it takes to assess a SO’s system following a disturbance.
Other new technologies, such as smart meters, have
allowed SOs to determine the outage area in a shorter
amount of time. Similar to drone technology, smart meters
allow SOs to gain situational awareness of their systems
without having to dispatch personnel to customer sites to
determine whether they were impacted by the disturbance.
In the USA, SOs have identified transmission corridors
that are critical to restoration processes. Hardening of
these corridor structures against weather impacts greatly
reduces the impact to these critical corridors needed
to restore the system. Depending on the disturbance,
equipment can be damaged and requiring replacement.
Many SOs identify “staging sites” where they stage spare
equipment that may be needed to repair transmission
corridors. Staging sites should be reviewed to ensure they
are not susceptible to flooding or other weather related
events that would make them inaccessible.
Lastly, communications are critical during a restoration
event. While many SOs have radio systems to communicate
with their crews, the ability to communicate across a large
geographical area in the event of a communications failure
would prove to be very difficult. Continuing to investigate
new technologies that can communicate data and voice
across large geographical areas would be beneficial as a
backup communications source during a restoration event.
Summary
Different approaches exist for power system restoration.
These can be categorized into three main classes: the top-
down, the bottom-up and the hybrid approach. Table 3
summarizes the examples given in this section.
Table 3 Worldwide Strategies for Restoration
Country Approach Restoration: Top-down
Australia Hybrid AC interconnections from bordering states
Brazil Hybrid LCC HVDC + AC interconnections
India Hybrid AC interconnections from bordering states
Ireland Hybrid VSC HVDC + AC interconnection
Italy Top-down AC interconnections from bordering national power system
USA Top-down AC interconnections from bordering states
Cigre Science & Engineering • N°14 June 2019
15
2. Operator Training for
Restoration
Operator errors and remedies
Analysis of the recent emergencies and blackout
incidents suggests that mistakes by system operators are a
significant contributor, being just the second main cause
of these events (with natural disasters being the first) [5].
The consequences of system operator errors, i.e. their
extent and severity, depend not just on the current power
system state and available remedial actions, but also on
operator actions and the available and activated defence
measures (from predefined defence plans).
Generally, operational mistakes can be attributed to
several reasons, such as
• Insufficient situational awareness
• Mistakes in control decisions due to incomplete
mental model2 of the process or incomplete analysis
• Incorrect interpretation of constraints
• Incorrect interpretation of conditions under which
protection acts
• Misjudgement of the effects from the initiated control
actions
• Misunderstanding in communication
Based on the above listed reasons, available and related
defence measures can be identified. In [5] these measures
are divided in two main categories:
• Measures aimed to reduce the likelihood of such
events
• Measures aimed to reduce the impact of such events
-----------------------------------
2 - The mental model of the process operator is a part of expert operator
decision making model. It is resident in his long term memory. Additional
details and references can be found in the CIGRE TB No. 524, pp.27-29.
Restoration: Bottom-up
Hydro + pumped storage + gas turbines
Hydro
Hydro + gas turbines
Hydro + pumped storage + gas turbines
 Figure 6 Training goals frequently used, N=15 [9]
Figure 7 Training methods used, N=16 [9]
Restoration training goals, methods, content and
organisation
Operator training is a crucial element for the successful
execution of designed restoration plans. Figure 6 [9]
illustrates the frequently used training goals. The training
of the operators covers several operating conditions,
including blackstart and restoration procedures.
The training itself can be done in different ways, from
lectures over OTS to innovative E-learnings. Lectures,
site visits and on the job training are done by almost
all companies surveyed in [10], while OTS training is
performed with slightly lower frequency as shown in
Figure 7 [9].
E-Learning is an upcoming training method, owing this
to its flexibility with respect to time and place. All these
trainings methods have advantages and disadvantages,
therefore a mixture of different methods can provide a
good balance.
For effectively handling the emergency and restoration
power system operation states, an advanced set of
training modules is needed to raise the operator’s
knowledge and skills. Such a training module would
contain, for example:
• Principles of power system dynamics
Cigre Science & Engineering • N°14 June 2019
16
• Principles of power system stability controls
• Identification of possible emergency states
• Operation of the power system in the emergency state
• Handling power shortage situations (i.e. generation
adequacy)
• Principles of load shedding
• Power system restoration principles and procedures
The main tool for restoration training today is an operator
training simulator (OTS), also known as dispatcher
training simulator (DTS). The survey conducted in [10]
identified that 80%, 90% and 90% of those with an OTS
use it for training of conditions with normal power system
state, for emergency handling and system restoration,
respectively. After the major incidents that occurred
worldwide in the last 10-15 years, we can expect that
practically all companies that do use an OTS/DTS for
operators’ training, use it extensively for emergency
handling and restoration training.
An example of an OTS/DTS architecture, including
subsystems Power System Model (PSM), Control Centre
Model (CCM) and Instructional Subsystem (ISS), is
illustrated in Figure 8 [10].
The PSM is responsible for the realistic representation
of all basic power system elements (generation, network
 Figure 8 Traditional OTS/DTS generic architecture [10]
and consumers, together with their main control
and protection devices). Apart from modelling, this
subsystem also includes algorithms that simulate all
relevant dynamic behaviour of the power system.
The CCM is responsible for the exact representation of
the power system control centre equipment (typically
SCADA/EMS system) that operators use in their daily
work to monitor, analyse, support decision making and
finally control the system. The operator training on a
replica of their SCADA/EMS system yields the highest
possible training effect, though use of generic CCM is
possible too (mainly for improving understanding of the
power system reaction on the operator’s actions).
Finally, the ISS should enable the instructor to monitor
and control (start, stop, pause) a training session,
introduce events, emulate other non-modelled parties
and support creation, validation and maintenance of
training scenarios.
In the simulator training environment the operators can
train under task-relevant stress conditions, i.e. conditions
that characterize real life incidents/restorations should
also be recreated during the training session. Stress
conditions can for example be emulated with additional
(numerous) phone calls, by shortening of some time
constraints or resources, and/or by introduction of
unknown events.
3. Restoration in the Future
TSOs’ experience and knowledge are mainly based on a
“traditional” power system, where conventional generators
have an essential role in maintaining the system’s security.
It is now possible for system dynamics to be monitored
in real-time by control room operators or analysed in
the operational planning timeframes, depending on the
specific power system (e.g. level of renewable energy
sources (RES) penetration, size and stability conditions).
In the future, the power system is expected to become
more dynamic and less stable if no additional measures are
taken [11], [12]. It is expected that frequency stability will
Cigre Science & Engineering • N°14 June 2019
17
deteriorate as the system synchronous inertia decreases,
the short-circuit power levels across the network will
become more volatile, and there will be more scarce and
distributed voltage control resources. Currently, there
is very limited provision of ancillary services, such as
voltage control or balancing, by RES. Furthermore, it is
still not common practice to redispatch (e.g. curtail) RES
for congestion management. There is also no restoration
service capability yet in RES. It is also the case that TSOs
currently have little or no observability and controllability
of RES – both factors being a pre-requisite for coordinated
power system restoration.
This transition requires SOs to make a paradigm shift to a
more diverse and decentralised set of resources to actually
control and restore the power system, including vertical
(TSO/DSO) and horizontal (TSO/TSO) coordination,
together with a new set of awareness and decision support
tools, which include the measurement and analysis of
system dynamics (using WAMS and DSA). Moreover,
the increasing integration of power electronics into the
power system creates new possibilities to effectively use
the available technology.
Impact of Transition to RES on Restoration
The increasing penetration of RES, especially based on
wind and solar energy, introduces a volatile production
pattern in the generation mix due to its intermittent nature.
During system restoration, SOs have as common practice
to still use conventional power plants to restore the power
system, making the process stable and predictable. In a
future where less or no synchronous generators are going
to be available, it is important to rethink the restoration
strategies, using all technologies available in the power
system with the capability to provide support.
Due to the fact that the share of RES in distribution
networks is nowadays significant with the tendency to
grow, there is a need for TSO/DSO integrated restoration
plans, which will involve increased coordination,
information exchange, joint operator training, and most
likely common tools.
Impact of Distributed Generation penetration fast changing voltage profiles.
levels on restoration High DG-partly controllable active power
Although restoration tasks for TSOs and DSOs are
Here again, the DSO’s system might be overloaded, but
generally quite similar, since they are based on pre-
now it has its own operational possibilities, making the
defined operational criteria, their allocation between
restoration easier for the TSO:
participating control centres can differ significantly. This
• The DSO can solve congestions individually by
tends to be strongly dependent on the penetration level of
limiting power flows
decentralized generation. Different levels of decentralized
• TSO and DSO can agree on a band of exchange
generation require different involvement from the DSO,
power (active) between their systems
as explained next.
This band enables the TSO to plan for provision of
Passive DSO
control power, because up to a certain level active power
A DSO without any DG is categorized as a passive DSO.
fluctuations are compensated directly for by the DSO.
The main task for a passive DSO is to connect load. In
As the amount of information exchange rises, additional
this case, from the perspective of the DSO, all restoration
tools may be required to support the restoration process.
strategies will end up in a top-down scenario with the
In this case, controllability of voltage profiles still is poor.
main task being voltage control.
High DG-partly controllable active and reactive power
Low DG-not controllable
In this case the DSO can additionally control voltage
The DSOs with low DG penetration may also be treated
profiles with DG. This is usually challenging due to
as passive DSOs from an operational perspective. At the
the large quantity of DG. Here, an EMS application is
DSO level there is DG, but it may not be observable and/
required for calculating reactive power setpoints using
or controllable. In the restoration preparation phase all DG
e.g. an optimal power flow. Unfortunately, these systems
should be disconnected from the system, in so far as this
are unlikely to function correctly in the early stages of
is possible. This will lead to automatic resynchronisation
restoration.
and provision of reactive power by the remaining DG.
The TSO must for the first time consider DG infeed in Table 4 summarizes expanding DSO tasks with
its restoration strategy. After providing reactive power increasing level of DG and increasing observability and
to the DSO and connecting load, DG may automatically controllability.
synchronise and feed in maximum power. PE support for system restoration
High DG-not controllable In general power electronics interfaced devices (PEID)
In this case the DSO is still passive, operationally speaking have advanced control possibilities that can be used and
but the DG can be observed. From the DSO point of view further developed to enhance system operations and
this case is critical. The DG is needed for restoration, network security, such as continuous voltage control,
but cannot be controlled. The reversed power flows may synthetic inertia, system restoration services, power
lead to congestion in the DSO grid, both on lines and oscillation damping. Power electronics can be found in
transformers. The fast changing load on DSO lines lead to power electronics interfaced generators (PEIG), as well as
Table 4 Overview of Expanding DSO Tasks

Control of Control Congestion Voltage support

TSO-DSO exchange Observe
power DG DG handling with DG
Passive DSO TSO - - TSO -
Low DG - no control TSO TSO - TSO -
High DG- no control TSO TSO DSO - TSO DSO -
High DG - partly controllable active power TSO DSO TSO DSO DSO TSO DSO -
High DG - partly controllable active and reactive power TSO DSO TSO DSO DSO TSO DSO DSO

Cigre Science & Engineering • N°14 June 2019

18
in HVDC transmission systems, energy storage systems
and FACT devices (e.g. STATCOMS).
In an environment with high penetration of power
electronics, these devices will play a relevant role in
power system restoration.
HVDC links
The functionalities of HVDC links can be utilized in
order to aid the system restoration, including providing
active and reactive support during blackstart and building
of the cranking path. Two types of HVDC links can be
distinguished when installed in the power system: HVDC
interconnections between synchronous areas (SAs) and
embedded HVDC links within the same synchronous
area (from pure AC to hybrid AC/DC systems). From an
HVDC interconnection it is expected that it will provide
support from the “healthy” SA, whereas from the meshed
HVDC link it is expected that it will mainly provide
power flow control. Both types of links, depending on the is discussed. An operating strategy for BESS as both load
technology, may provide voltage control.
Whereas HVDC links are not commonly used for
providing restoration service, their participation is
expected to increase in the future. The use of the HVDC
in the restoration process depends on the technical
characteristics related to the type of technology as
described below [4].
Line Commutate Converter (LCC) HVDC
The LCC HVDC is a mature, thyristor-based technology
with the highest power rating and efficiency. LCC can
provide very fast active power regulation (primary
frequency regulation). For this, it needs a high level of
short circuit power. It is furthermore characterized by a
technical minimum of transmitted active power. Finally,
there is a relatively high risk of commutation failure.
Relating to restoration strategies, the LCC HVDC can
be used as a target “group/load” with a big regulation
capacity. In conclusion, LCC can be used in the bottom-
up strategy. For the operation of LCC HVDC a certain
short-circuit power level is required.
Voltage Sourced Converter (VSC) HVDC
The VSC HVDC is a younger, IGBT-based technology. On a smaller scale, retailers in South Australia are
VSC, similar to LCC, can provide a very fast active power
regulation. In addition, VSC doesn’t require a high short
circuit power level; it has a high reactive power control
capability, which is important during the restoration
Cigre Science & Engineering • N°14 June 2019
19
process. It has no technical minimum for active power
transmission and it does not have the risk of commutation
failure. Relating to restoration strategies, the VSC HVDC
can be used in the following situations:
• When the HVDC interconnects a healthy and a
blacked out system. In this case, the VSC can provide
active and reactive power regulating capability
• By using a nearby hydro or turbo-gas generation plant
or, alternatively, by using an internal source (e.g.
properly sized diesel generator), the VSC HVDC can
provide reactive power capability.
In conclusion, VSC HVDC can be used both for bottom-
up and top-down restoration strategies.
Energy Storage Systems
Battery Energy Storage Systems (BESS) can be used in
several ways for supporting the restoration process [13].
In [14] the participation of storage in the load restoration
and generator during restoration is proposed in [15]it is
critical to maintain system frequency stable to avoid any
further outage or cascading events. Load pickup is one of
the most important tasks to keep generation-load balance
for a stable system frequency. In current industry practice,
small loads are suggested to be picked up incrementally
to avoid any instable frequency dip. However, slow load
pickup may prolong the system restoration process. As
the fast response energy resources, battery storage system
(BSS. This strategy is highly dependent on the battery’s
charging and discharging capabilities.
In [16] the optimal size for batteries is determined, aiming
at providing frequency support during the restoration
process. In South Australia, there currently is a 100 MW
(129 MWh) BESS installation at Hornsdale, and a 30 MW
(8 MWh) BESS installation at Dalrymple. One of the aims
of these batteries is to provide frequency control ancillary
services. Successful operation was demonstrated on 14
December 2017. The battery in Hornsdale discharged
with millisecond response (see Figure 9) to immediately
arrest the frequency excursion following a trip of a 560
MW coal fired power plant.
implementing small scale Virtual Power Plants. Two
recent government launched schemes could see 90,000
new batteries with up to 400 MW of controllable storage
connected to the network at the distribution level.
 Figure 9 Hornsdale power reserve response to disturbance of 14 December 2017
Therefore, increased visibility is required of the low
voltage network, as well as greater coordination, e.g.
concerning voltage control.
Storage can also be used as blackstart sources that provide
the required power to non-blackstart generators. In [17]
for example, a case study of a hybrid system consisting
of a BESS and a STATCOM is presented. This hybrid in 2010 during the connection phase of Turkey to the
system was able to provide energizing power of 27 MW
for 15 minutes, which was considered sufficient for initial
restoration.
WAMS support for system restoration
Fast system dynamics resulting from the integration of
PEID adds complexity to power system operation. Wide
Area Monitoring Systems based on Phasor Measurement
Units (PMUs) are being used by System Operators
worldwide in the operational environment, giving the
control room information about the dynamic behaviour
of the network and, consequently increasing awareness
for system dynamics. Next to the improved situational
awareness and decision support, the synchrophasor
technology in the control room can definitely contribute to
power system restoration. When compared to traditional
SCADA measurements, synchrophasors have an added
value of synchronised voltage phase angle information
between areas that have to be re-energized and/or re-
connected, which can significantly benefit the restoration
process. In the preparation phase of the restoration
process, and when complemented with state estimation
data, the synchrophasors provide precise information of
the remaining system, its division in islands and available
components in the system. This information helps to
construct the restoration strategy. From a restoration
viewpoint, the restoration stage can be enhanced with
critical data such as synchrophasor measurements from
generating units and critical load. In the restorative
control module (as part of the SCADA), the algorithm
constituting of synchrophasor and state estimation data
can help in the automated process of re-energization
Cigre Science & Engineering • N°14 June 2019
20
of the shed load, resynchronisation of multiple islands
and automated building up of power generation. The
automated algorithm assists the operator in rebuilding the
system again with reduced time [18].
In 2004, WAMS were used during the resynchronisation
process of the first and second UCTE zone, and again
Continental European system. By doing so an on-line
dynamic assessment of the system stability during
the “hot” synchronisation phase was performed. The
policies, guidelines, as well as responsibilities during
resynchronisation for the ENTSO-E jurisdiction are
documented in Policy 5 of the Continental Europe
Operation Handbook [19].
In India, synchrophasors are being used in control
rooms for real time system monitoring and analysis
[20], [21]. Since PMUs send time synchronised signals
of frequency/voltage phasors, the synchronisation relay
available at substations could effectively be mimicked
at control centres. This helps in coordination between
generating station(s) in different islands and synchronising
substations for matching of electrical parameters and
faster synchronisation of islands. The above is being used
during synchronisation of islands during mock blackstart
exercises and is likely to prove beneficial during actual
restoration process.
The Italian practice, as highlighted in Section 1 is another
example of WAMS usage for restoration purposes.
DSA support for system restoration
A list of enhancements worth considering to further
improve the restoration effectiveness and efficiency
is presented in [4]. One of them is adopting advanced
SCADA and EMS functionalities, such as wide-
area security assessment or a fully integrated control
landscape including SCADA/EMS, PMU within Wide-
Area Control Systems and advanced analytics tools (e.g.
dynamic security analysis, DSA, and the like) to enhance
awareness and analytical capability to improve the
restoration process. In [1] a framework is proposed for
coordinated stability control. Using online DSA, control
decisions could be developed online (i.e. online control
decision planning). In this approach, a decision support
system may analyze available resources during normal
power system operation. Typical scenarios are selected
and restorative plans are generated in an online setting,
based on the actual conditions of the power system and its
operating environment. The advantage of this is that the
actual evolution of the power system is taken into account
in the decision planning process, therefore increasing its
effectiveness.
However, the implementation of such a scheme is not an
easy task. State of the art technologies for restoration have
barely achieved online applications [3]. There seems to
be a lack of common understanding and awareness about
tools and technologies for automation of online decision
planning for restoration control. In one application
(detailed in [1]), online DSA is used to verify the system
operator’s planned restoration controls, before any actions
are taken.
Within the European power system, an awareness system
called ENTSO-E Awareness System was set-up. It is in
fact not a DSA, but only a web-based monitoring system
where all TSOs share their main cross-border flow,
individual area control error, traffic light system with
respect to (n-1) security, tools availability, etc.
New requirements from OTS/DTS
Based on all the previously mentioned developments, USA were given.
today’s power system planning and operational
complexity have increased significantly, leading to a
situation where control decisions are much harder to
make and the consequences of inappropriate decisions,
in terms of their cost, can be even higher. This has
influenced many electricity regulators to require, not just
the certification of power system operator competences,
but also accreditation of their training programs.
Hence, training programs and training tools (OTS/DTS at
the first place) have to be changed accordingly to include
not just knowledge of new topics but also formation of
new skills and attitudes. This relates especially to the
restoration simulation capabilities of many of the already
existing simulators.
Cigre Science & Engineering • N°14 June 2019
21
New OTS/DTS power modelling capabilities must be
such that they realistically model:
• Renewable generation (including control) that is
often far from load centres
• HVDC, FACTS and protection systems
• Customer load modelling, demand-side
management and storage.
Analogously, new application functionalities that appear
in the TSO/ISO control centre should also be included
in the CCM with the goal to increase the reliability of
trainings. Examples of such functionalities are:
• DSA applications (including dynamic line and
transformer rating possibilities)
• Operator awareness systems (supported by
WAMS)
• Weather forecast system access
More details on OTS architecture requirements are
given in [10], while the current developments and
requirements on OTS, in light of ongoing changes in
the electricity industry and ICT technology, will be
prospected by the newly formed working group C2.39.
4. Conclusions
The aim of this Reference Paper is to shed some light
on worldwide practices of and expected future trends in
power system restoration. After a power system blackout,
it can be restored using either the top-down, bottom-up
or hybrid approach. Examples of implemented restoration
strategies in Australia, Brazil, India, Ireland, Italy and the
As was investigated by WG C2.21, human factors in
the control room are the second main cause of events
leading to blackouts, with natural disasters being the
first. This makes operator training one of the most crucial
activities for reducing the impact as well as the likelihood
of blackouts. Therefore, it is not a surprise that system
operators that are using operator/dispatcher training
simulators, use these intensively for emergency handling
and restoration training. This approach has shown to
be helpful in keeping all stakeholders prepared and in
reducing the time during actual restorations.
With the rapidly developing power system, changes and
innovations in different parts of the restoration process are
necessary. With increasing distributed energy resources,
the role of the distribution system operator in power
system restoration will become more important, where
coordination between different stakeholders will be
key. This increasing generation in distribution networks
demands an improved visibility. The use of WAMS can
greatly help to achieve this, especially during restoration
activities where situational awareness in the control room
is of utmost important.
Other innovative solutions for creating increased situation
awareness include utilisation of drones and smart meters.
System operators are using drones to inspect transmission
corridors in order to assess damage following e.g. natural
disasters. Drone technology has proved to significantly
decrease the amount of time it takes to assess the physical
condition of the power system after a disturbance. Smart
meters, on the other hand, have allowed system operators
to determine the outage area in a shorter amount of time.
Similar to drone technology, smart meters allow system
operators to gain situational awareness of their systems
without having to dispatch personnel.
With increasing power electronics interfaced devices in
the power system, it is also worth further investigating
how these can support the system operator. CIGRE
joint working group C2/B4.38 is investigating how the Development, 2018, pp. 1–9.
capabilities of power electronics interfaced devices can
be fully utilised to the benefit of the system operator.
CIGRE working group C2.26 will assess the required
changes in the current restoration process when systems
are becoming more and more dominated by power
electronics. This working group will also give insights
in the required simulation and modelling practices for
accurately simulating restoration processes.
Furthermore, an industry-academic research project on
restoration in future systems with high penetration of
power electronic converters is currently being drafted
as part of a European Union Horizon 2020 proposal.
Topics to be investigated in this project include real-time
interfaces for situational awareness of power systems in
the restoration state and the use of WAMS for supporting
a stable restoration process.
This all confirms that CIGRE, the industry as well as
academia acknowledge the need to further investigate
the topic of restoration. The aim is that these ongoing
activities will result in concrete practical guidelines for
system operators to be better prepared for operating a
non-conventional power system.
Cigre Science & Engineering • N°14 June 2019
22
References
[1] CIGRE JWG C2/C4.37, “Technical Brochure 742: A Proposed
Framework for Coordinated Power System Stability Control,” Paris,
2018.
[2] ENTSO-E, “Commission Regulation (EU) 2017/1485 Guideline on
Electricity Transmission System Operation,” Brussels, 2017.
[3] Y. Liu, R. Fan, and V. Terzija, “Power system restoration: a literature
review from 2006 to 2016,” J. Mod. Power Syst. Clean Energy, vol. 4,
no. 3, pp. 332–341, 2016.
[4] CIGRE WG C2.23, “Technical Brocure 712: System Restoration
Procedure and Practices,” CIGRE, Paris, 2017.
[5] CIGRE WG C2.21, “Technical Brochure 608: Lessons Learnt from
Recent Emergencies and Blackout Incidents,” CIGRE, Paris, 2015.
[6] AEMO, “Black System South Australia 28 September 2016,” 2017.
[7] Australian Energy Market Operator, “South Australia System Strength
Assessment,” 2017.
[8] Committee on Grid Disturbance, “Report of the Enquiry Committee
on Grid Disturbance in Nothern Region on 30 th July 2012 and in
Nothern, Eastern and North-Eastern region on 31 st July 2012,” 2012.
[9] CIGRE WG C2.35, “Technical Brochure 677: Power system operator
performance: corporate, operations and training goals and KPI’s
used,” Paris, 2017.
[10] CIGRE WG C2.33, “Technical Brochure 524: Control Centre Operator
Requirements, Selection, Training and Certification,” Paris, 2013.
[11] V. N. Sewdien et al., “Effects of Increasing Power Electronics on
System Stability: Results from MIGRATE Questionnaire,” in 2018
IEEE PES International Conference on Green Energy for Sustainable
[12] A. Halley et al., “Effects of Increasing Power Electronics based
Technology on Power System Stability: Performance and Operations,”
CIGRE Sci. Eng. J., vol. 11, pp. 5–17, 2018.
[13] D. K. Maina and N. K. C. Nair, “Recent advancements on power
system restoration,” in 2017 IEEE Innovative Smart Grid Technologies
- Asia: Smart Grid for Smart Community, ISGT-Asia 2017, 2018, pp.
1–5.
[14] J. Dang and R. G. Harley, “Islanded microgrids black start procedures
with wind power integration,” in IEEE Power and Energy Society
General Meeting, 2013, pp. 1–5.
[15] N. Kadel, W. Sun, and Q. Zhou, “On battery storage system for load
pickup in power system restoration,” in IEEE Power and Energy
Society General Meeting, 2014, vol. 2014–Octob, no. October, pp. 1–5.
[16] P. Mercier, R. Cherkaoui, S. Member, and A. Oudalov, “Optimizing
a Battery Energy Storage System for Frequency Control Application,”
IEEE Trans. Power Syst., vol. 24, no. 3, pp. 1469–1477, 2009.
[17] M. Bahrman and P. E. Bjorklund, “The new black start: System
restoration with help from voltage-sourced converters,” IEEE Power
and Energy Magazine, vol. 12, no. 1, IEEE, pp. 44–53, 2014.
[18] CIGRE WG C2.17, “Technical Brochure 750: Wide Area Monitoring
Systems – Support for Control Room Applications,” Paris, 2018.
[19] ENTSO-E, “Handbook Operation - P5 - Policy 5: Emergency
Operations,” Brussels, 2015.
[20] A. Singh et al., “Report on Low Frequency Oscillation in Indian Power
System,” New Delhi, 2016.
[21] POSOCO, “Synchrophasors - Initiative in India,” New Delhi, 2013.
Study Committee C4 – System technical
performance
The use of battery energy storage systems
or system integrity protection schemes in
South Australian power system
N. MODI*, B. BADRZADEH, S. GROGAN, C. NORONHA
Australian Energy Market Operator (AEMO)

Abstract Transient (EMT) model of the SA power system. For

The National Electricity Market (NEM) on the east coast
of Australia is comprised of five regions. Of these five
regions, the South Australia (SA) region is unique, in
that it has the highest ratio of asynchronous generation to
regional operational demand and routinely meets its total
demand by asynchronous generation alone. Additionally,
SA has around 950 MW of rooftop photovoltaic (PV)
installed, and, at the time of commissioning on 1 December
2017, the world’s largest lithium-ion battery, Hornsdale
Power Reserve (HPR). This battery energy storage system
(BESS) has a capacity of 100 MW and 129 MWh, with
more large-scale BESS capacity continuously being
added in the SA region and the NEM.
The SA power system has been islanded from the rest of
the NEM on occasions. One cause has been the operation
of the sole AC interconnector’s loss of synchronism
(LoS) protection relay. In September 2016, several factors
culminated in the operation of the interconnector’s LoS
relay, resulting in a state-wide blackout of SA. As a
recommendation from the investigation of this event, a
comprehensive system integrity protection scheme (SIPS)
was developed to improve SA power system security.
The SIPS scheme has three stages to its operation. The
intent is that they operate in sequence so each stage
defers or avoids activation of the subsequent stage. The
HPR battery is a key component of the SIPS, and this
paper examines the role of BESS in improving power
system security. The Australian Energy Market Operator
(AEMO) analysed a suite of system conditions which
were developed using a highly detailed Electromagnetic
the studied scenarios, AEMO identified that in certain
circumstances, the rapid injection of active power into
the SA power system from the BESS can avoid a LoS
condition occurring between SA and the rest of the NEM.
It has also been identified that this rapid injection of active
power from the BESS improves overall SA power system
security.
Introduction
The reduction in the cost of large-scale Battery Energy
Storage Systems (BESS) over a relatively short period
means that their deployment in power systems around
the world is rapidly increasing. Large scale BESS is
providing system security services such as frequency
and voltage control [1] [2]. BESS can also be used to
enhance performance of transmission grids and facilitate
more renewable integration [3][4][5]. In Australia, four
transmission connected large scale BESS are already
in operation, with more progressing through the grid
connection process.
The National Electricity Market (NEM) in Australia
covers one of the longest and least-interconnected power
systems in the world. It spans over 4,500 km and consists
of five power system regions across the east and south-
east of Australia. A map with geographical boundaries of
the NEM and SA power system is shown in the Appendix.
The South Australia (SA) region is at the fringe of
the NEM and has a relatively weak connection to its
neighbouring region via one double circuit 275 kV AC
interconnector (known as the Heywood Interconnector)
and one DC link (known as Murraylink). Existing transfer

* nilesh.modi@aemo.com.au

KEYWORDS
Battery Energy Storage Systems, Special Protection Scheme, Loss of Synchronism

Cigre Science & Engineering • N°14 June 2019

23
limit on Heywood interconnector is 500 MW from SA
towards Victoria (VIC) and 600 MW from VIC to SA.
The transfer limit on Murraylink is ± 220 MW.
Due to its unique location in the NEM and energy resource
mix, the SA region has distinct operational challenges
compared to other regions. Characteristics of the SA
power system include:
• A very high amount of asynchronous generation
(approximately 2,100 MW), compared to an
average operational demand of 1,000-1,200 MW.
The asynchronous generation is comprised of wind
generators, photovoltaic (PV) solar generators, and
grid-connected BESS.
• No coal-fired power stations remain in the region.
• The largest grid-connected lithium-ion battery in
the NEM, Hornsdale Power Reserve (HPR), with a
capacity of 100 MW/129 MWh.
• The majority of synchronous generation (gas-
powered) are very close to the metropolitan load
centre, while most of the wind and grid-connected
solar generators are electrically distant from the
metropolitan load centre.
• Around 950 MW of domestic and commercial
rooftop PV.
The SA region has recently seen a dramatic growth in
renewable energy penetration. Routinely, more than 100%
of the operational demand of SA has been provided by grid
connected asynchronous generation. The instantaneous
penetration of grid-connected asynchronous generation
has historically approached 140% of the operational
demand.
With a very high ratio of asynchronous to synchronous
generation, and remoteness from the rest of the
interconnected network, the SA power system is highly
reliant on continuous uninterrupted operation of its only
AC interconnector to maintain system stability. A key
characteristic of the network is that multiple generation
contingencies can cause unstable power swings resulting
in a loss of synchronism between SA and its neighbouring
region.
The SA region has experienced several such events which
have resulted in disconnection from the rest of the NEM.
In all cases except the South Australian blackout on 28
September 2016, a viable electrical island was formed.
Cigre Science & Engineering • N°14 June 2019
24
Once islanded, the SA power system has historically relied
on under-frequency load shedding (UFLS) to restore the
supply demand balance.
During a major storm on 28 September 2016, significant
clusters of wind turbines were unable to ride-through
multiple faults which occurred in a relatively short period
of time on the transmission system. This resulted in a
substantial reduction of wind generation, which was
the largest generation source in SA at the time. The
reduction of wind generation resulted in increased flows
on the Heywood AC interconnector towards SA to the
point where SA began to lose synchronism with rest
of the NEM. The Heywood AC Interconnector Loss of
Synchronism (LoS) protection relay detected this event
and subsequently tripped the AC interconnector. The large
generation loss within SA coupled with the subsequent
loss of interconnector import, resulted in a massive energy
imbalance.
AEMO investigated the South Australian blackout
event in detail and made several recommendations [6].
AEMO’s investigation indicated that before the time of
separation, system frequency did not drop sufficiently to
initiate UFLS action as the SA system frequency was the
same as that of the interconnected system. Immediately
after system separation, the frequency deteriorated very
rapidly, with a rate of approximately 6.25 Hz/s, such that
the UFLS scheme did not have sufficient time to respond
prior to system collapse.
One of the recommendations from AEMO’s investigations
was to develop a System Integrity Protection Scheme
(SIPS) to pre-emptively reduce load in SA while the
system is still interconnected. Doing so may either prevent
loss of synchronism or allow for islanding conditions that
would otherwise be unviable. The remainder of this paper
discusses this scheme in more detail.
2. System integrity protection
scheme
The SIPS is designed to rapidly identify conditions that
would otherwise lead to a LoS between SA and rest of the
NEM. The scheme has three stages, intended to operate
sequentially. Successful operation of stages 1 and 2 keeps
the SA region synchronously connected to the reminder
of the NEM, thereby providing more opportunity for the
 Figure 1 - Response from a thermal unit and battery
SA power system to recover to a stable operating state and
avoid a large frequency excursion. Operation of stage 3
would separate the SA region from the rest of the NEM.
However, to achieve stable operation after the activation
of stage 3, the SA region needs to rely on UFLS to restore
the supply-demand balance.
2.1. Stage 1: Energy discharge from BESS
On 1 December 2017, the world’s largest grid-connected
lithium-ion battery (often referred to as the “Tesla
Battery”), was commissioned. This BESS is officially
known as the HPR and is located approximately 200 km
north of Adelaide. It is rated for 100 MW / 129 MWh
capacity. Of the 100 MW instantaneous capacity, 70 MW
is reserved for power system security purposes, with the
remaining 30 MW available for commercial operation.
Operation of the BESS to date suggests that it can provide
a range of valuable power system security services,
including rapid and accurate responses to frequency
deviations and frequency control services [7]. The
BESS has been configured to provide post-contingency
frequency control support at all times.
Figure 1 shows a comparison between the response of
a thermal generating unit and battery during an under-
frequency and subsequent over-frequency event in the
NEM in 2018. The thermal unit and battery injected
additional active power of approximately 45 MW and
80 MW, respectively, before the frequency nadir. More
importantly, when the frequency turned around, the
battery immediately started absorbing power thereby
providing a greater contribution to arresting the frequency
peak than the thermal unit.
When this analysis was carried out, only the HPR was
integrated into the SIPS stage 1. Recently another
transmission connected battery near Dalrymple substation,
known as ESCRI, was also integrated into the SIPS stage
Cigre Science & Engineering • N°14 June 2019
25
1. However, this analysis does not include the effects of
this more recent addition. Further transmission connected
batteries are expected to be included in the scheme as
they come online. Stage 1 utilises the 70 MW of BESS
reserve from HPR, with the aim of avoiding LoS between
the SA power system and the rest of the NEM. Active
power is rapidly injected from the BESS while SA is still
synchronously connected. The stage 1 SIPS activation
signal to the BESS is initiated if active power flow into
SA over the Heywood interconnector either:
• Increases at a rate which is faster than a given
threshold, or
• Increases above a specified absolute value.
Once the trigger is received, the BESS will immediately
inject 70 MW into the SA power system, giving time for
either generator frequency control mechanisms to respond
or if unavailable, for the UFLS to operate.
Figure 2 shows a simplified schematic diagram that
illustrates the logic on which SIPS stage 1 operates.
This scheme was designed by ElectraNet who is the
transmission asset owner in the SA.
As shown in Figure 2, the operation of SIPS Stage 1 is
triggered by active power levels flowing into SA through
the Heywood interconnector. Based on operational
experience, thresholds for both the level and the rate-of-
change of active power have been determined.
2.2. Stage 2: Pre-emptive load shedding
The aim of stage 2 is to quickly detect power swings that
would otherwise lead to LoS if not rapidly managed.
The stage initiates pre-emptive load shedding (i.e. before
frequency reduces to the UFLS trigger levels) so as to
avoid a LoS condition between the SA power system and
the rest of the NEM.
The activation signal is initiated from a LoS relay set
 Import
Import >> Threshold
Threshold Import
Import
Stage 1 Trigger OR
OR power
power
Rate
Rate of
of change
change of
of into
into SA
SA
import
import >> Threshold
Threshold

SOUTH AUSTRALIA VICTORIA

LOAD
SHEDDING SOUTH EAST
LOS
LOS SUBSTATION
TRIGGER TO
RELAY
RELAY REMOTE
TAILEM BEND SUBSTATIONS
SUBSTATION
Stage 2 Trigger

Figure 2 - Stage 1 and 2 logic diagram of SIPS

to look for power swings that would otherwise lead to the outer and inner blinders within a specified timeframe,
instability. This LoS relay is installed at a substation deeper then the relay operates. An example characteristic of each
within the SA power system than the first interconnector relay is shown in Figure 3.
substation, but still on the primary path for inter-regional
power transfers (Tailem Bend substation in Figure 2). X
As further explained in Section 2.4, this substation is
Location of LoS
also the natural electrical centre of the SA power system protection
Location of LoS
protection (South
and the power system of the neighbouring NEM region (Electrical Center of
East)
SA – Tailem Bend)
(Victoria). This means that phenomena associated with
LoS (i.e. depressed voltage and increased power flow) are
likely to manifest at this location before other substations
in the region.
R

Upon detection of a power swing that would otherwise

lead to separation, the LoS relay at Tailem Bend activates
SIPS stage 2. This relay sends a signal to predetermined
circuit breakers around the SA region which disconnect
associated loads. This disconnection of load is outside of
the UFLS as this disconnection does not depend on the
frequency. This action does not account for how much
load is available to be shed on each feeder therefore it
can result in under- or over-disconnection (of load) for a
given disturbance. Figure 2 shows a simplified schematic
diagram that illustrates the logic on which SIPS stage 2
operates.
2.3. Stage 3: Out of step trip
This is the final stage in the SIPS. It is a traditional LoS
detection across the AC interconnector between SA and
its neighbouring region. Once a loss of synchronisms is
detected, a LoS protection relay on the SA side of the
interconnector opens the dual 275 kV circuits, electrically
islanding the state. Once islanded, the SA region relies on
traditional UFLS or over-frequency generation shedding
(OFGS) to restore the generation-demand balance.
2.4. LoS relays in SIPS stages 2 and 3
Currently, LoS relays are installed at two substations:
South East and Tailem Bend. Both relays are impedance-
based with double blinders. An impedance trajectory is
derived from the voltages and currents measured at each
substation. If the impedance locus transverses through both
Outer Blinder
Inner Blinder
Figure 3 - Loss of synchronism relay characteristics for Stage 2 and 3
A. South East LoS relay (Stage 3)
This substation is electrically close to the Heywood AC
interconnector. Upon detection of LoS, the relay opens
the two 275 kV interconnector circuits between South
East and Heywood substations (located in neighbouring
Victoria) resulting in disconnection of SA from the rest
of the NEM.
B. Tailem Bend LoS relay (Stage 2)
Following analysis from the SA blackout event, a second
LoS relay was installed at Tailem Bend substation. This
substation is between South East and Adelaide metro, at
the electrical centre between SA and rest of the NEM. and
Figure 5 show the frequency and voltage angle traces for
the blackout event which occurred on 28 September 2016
[6]. They demonstrate how the frequency and voltage
angle at the South East bus does not closely follow the
rest of the SA system just before LoS. This suggests that
the electrical centre between SA and rest of the NEM is
closer to Tailem Bend, which is in between South East
and the Adelaide metro region.

Cigre Science & Engineering • N°14 June 2019

26
 Figure 4 - Frequency traces for SA and Victoria - 28 September 2016 event

Figure 5 - Voltage angle for SA – 28 September 2016 event

The operation of this LoS relay is similar to that of the
South East LoS relay. However, the intent of this relay is
to pre-emptively disconnect loads within the SA region.
On detection of a power swing (i.e. loss of synchronism),
it sends a trip signal to predetermined circuit breakers to
disconnect load.
2.5. System conditions, study methodology and
assessment criteria
To assess the effectiveness of the BESS’s contribution
to the overall SIPS, a wide range of operating conditions
were simulated in a region-wide Electromagnetic
Transient (EMT) model of SA. For each of the system
conditions, a set of contingency events were investigated
as listed in Table 1.

Cigre Science & Engineering • N°14 June 2019

27
 Table 1 - Contingency events

Type of Generation Contingency

Scenario Single / Multiple Contingency Amount of Generation loss (MW)
(i.e. loss of generation)
A Synchronous generation Multiple 300 – 390
B Asynchronous generation Multiple 300 – 370
C Synchronous generation Multiple 200
D Asynchronous generation Single 270
E Synchronous and asynchronous generation Multiple 470

Table 2 - System conditions and contingencies

Asynchronous generation Synchronous generation level AC Interconnector im-

Scenario Contingency (MW)
level (MW) (MW) port (MW)
B1 1200 300 600 300
E1 1200 550 600 470

To assess the effectiveness of the BESS, an EMT model
of the SA region was developed. The model development
methodology, details about the models and their integration,
benchmarking of the models are available in [8]. As
the SA region has substantial asynchronous generation
compared to its demand, it was more appropriate to assess
the SIPS design using highly detailed EMT type models.
Once an EMT model of SA was established, a number of
scenarios to reflect various system operating conditions
were developed and analysed.
The effectiveness of stage 1 (BESS) was assessed based
on the following criteria:
• Effectiveness of stage 1 in avoiding a stage 2 trigger.
• Effectiveness of stage 1 in improving overall power
system security in SA.
3. Results and Discussion
This section presents key findings from the analysis
undertaken. In this paper, only results associated with key
scenarios are discussed. Table 2 provides detail of those
key scenarios.
3.1. Loss of multiple asynchronous generators –
Scenario B1
In this scenario, 300 MW of asynchronous generation
was tripped simultaneously. To investigate the ability of
each stage to defer the activation of the subsequent stage,
AEMO studied the following sensitivities.
3.1.1. All stages disabled
The intent of this sensitivity study was to confirm that a

Figure 6 - Voltage phase angles

Cigre Science & Engineering • N°14 June 2019

28
 Figure 7 - Impedance locus Stage 2 and 3

multiple contingency event that leads to a total generation
loss of 300 MW can trigger LoS between SA and the
rest of the NEM. It is apparent from Figure 6 and Figure
7 that such an event can trigger a LoS condition without
mitigating action from SIPS. Figure 6 shows how the
voltage phase angles between South East and the rest of SA
start to diverge, indicating an onset of LoS. This is further
confirmed in Figure 7, which shows the impedance loci
measured at both Tailem Bend and South East substations.
3.1.2. Stage 1 disabled, Stages 2 and 3 enabled
The intent of this sensitivity study was to test the
effectiveness of the SIPS stage 2 pre-emptive load shedding
in avoiding the onset of stage 3 (i.e. disconnection of SA
from the rest of the NEM) in the B1 scenario. Figure 8
shows the voltage phase angles and re-confirms the onset
of LoS between SA and its neighbouring region.
The following observations can be made regarding the
activation of stages 2 and 3 of the SIPS.
• Stage 2 of the SIPS (i.e. pre-emptive load shedding)
correctly triggers and disconnects load.
• The voltage phase angles stop diverging, as shown in
Figure 8.
• The trigger of SIPS stage 3 is avoided. Figure 9
shows that the impedance locus of stage 3 does not
cross the inner blinder of the LoS relay.
• The voltages throughout SA return to new steady-
state values as shown in Figure 10.
• Figure 11 shows the amount of load that was
disconnected. On disconnection of this load, the
SA power system regains synchronism with the rest
of the NEM and starts towards a new steady-state
operating point.

Figure 8 - Voltage phase angles

Cigre Science & Engineering • N°14 June 2019

29
 Figure 9 - Impedance locus of Stage 2 and 3

Figure 10 - Voltages at key high voltage substation in SA

Figure 11 - Load shed

Cigre Science & Engineering • N°14 June 2019

30
 Figure 12 - Voltage phase angles

3.1.3. All stages enabled
The intent of this sensitivity study was to assess the
ability of SIPS stage 1 to avoid triggering stage 2 in the
B1 scenario. The process by which SIPS stage 1 acts is
shown in Figure 2.
The following observations can be made for this scenario:
• Rapidly discharging BESS to the grid prevents the
development of unstable power swings in the SA
system. Voltages and voltage phase angles across the
region return to new steady state values relatively
quickly. This is shown in Figure 12 and Figure 14.
• Stage 1 (i.e. energy discharge from BESS) avoids the
trigger of Stage 2 (i.e. pre-emptive load shedding).
The impedance locus shown in Figure 13 confirms
the avoidance of a LoS condition. As shown in
Figure 13, the impedance locus feeding the Stage 2
trigger crosses the outer blinder, but does not cross
the inner blinder and trigger further SIPS action. The
locus returns beyond the outer blinder towards a new
steady state value.
3.2. Loss of multiple synchronous and asynchronous
generators – Scenario E1
Scenarios with multiple synchronous and asynchronous
generation disconnections were considered. The following
is an example of the loss of 470 MW of synchronous and
asynchronous generation.
3.2.1. Stage 1 disabled, Stages 2 and 3 enabled
The intent of this sensitivity study was to identify the
effectiveness of SIPS Stage 2 in deferring the activation of
Stage 3. Figure 15 and Figure 16 show that in the absence
of SIPS Stage 1 action, these multiple contingencies are

Figure 13 - Impedance locus of stage 2 and 3

Cigre Science & Engineering • N°14 June 2019

31
 Figure 14 - Voltages at key high voltage substation in SA

likely to result in a LoS between SA and the rest of the 4. Conclusions

NEM. This unstable condition is successfully detected by
SIPS Stage 2, which triggers pre-emptive load shedding in
SA. Although a possible recovery initially seems possible,
eventually a second LoS occurs and SIPS Stage 3 triggers.
3.2.2. All stages enabled
The intent of this sensitivity study was to identify the
effectiveness of SIPS stage 1 (BESS energy injection).
Figure 17 to Figure 19 demonstrate that for this scenario,
activation of SIPS stage 1 does not avoid activation
of Stage 2. However, stage 1 activation improves the
overall recovery of the SA power system to the multiple
contingency event and activation of the stage 2 pre-
emptive load shedding avoids activation of Stage 3.
Overall, activation of SIPS stage 1 and 2 collectively
defers the activation of stage 3, allowing the SA region to
remain connected to the rest of the NEM.
A SIPS has been developed for SA comprising fixed load
shedding and rapid injection by BESS. Prior to deployment,
the scheme was tested for a range of system conditions
using a large-scale detailed EMT model of SA. Stage 1
of the scheme uses the BESS energy and is activated on
both interconnector import level and rate of change of
power flow. Stage 2 of the scheme pre-emptively sheds
load in SA on detection of unstable power swings using
a LoS relay within the SA region. Stage 3 separates SA
from the rest of the NEM using a traditional LoS detection
mechanism on the AC interconnector.
The results highlight the effectiveness of the BESS’s
rapid active power injection in avoiding a LoS condition
between SA and rest of the NEM. It was demonstrated that
the inclusion of the BESS in the SIPS clearly improves the
performance of the SIPS as well as power system security
outcomes in the SA region overall.

Figure 15 - Voltages at key high voltage substation in SA

Cigre Science & Engineering • N°14 June 2019

32
 Figure 16 - Voltage phase angles

Figure 17 - Voltages at key high voltage substation in SA

Figure 18 - Voltage phase angles

Cigre Science & Engineering • N°14 June 2019

33
 Figure 19 - Impedance locus of Stage 2 and Stage 3

5. Acknowledgements Systems to Defer Distribution Capacity Upgrades on Utility, CIGRE

The authors would like to acknowledge the support
provided by ElectraNet and for the useful network insights
and discussion that occurred during the project.
6. References
[1] E.Sasano et.al. ‘Demonstration projects for providing ancillary
services using three different types of large-scale battery systems’,
CIGRE 2018, Paper No C2-112
[2] D. Stenclik et.al. ‘The Role of Fast Frequency Response in Low Inertia
Power Systems’, CIGRE 2018, Paper No C2-107
[3] N. Pahalawaththa et.al. ‘Battery Storage for Enhancing the
Performance of Transmission Grids’, CIGRE 2018, Paper No C4-118
[4] E. Paaso et.al. ‘Benefit Cost Analysis of Battery Energy Storage
2018, Paper No C6-201
[5] J.H. Lee et.al. ‘Application of ESS for Wind Power Connection to
Youngheung Wind Farm’, CIGRE 2018, Paper No C6-209
[6] Black system South Australia 28 November 2018, AEMO, Available
at: https://www.aemo.com.au/Media-Centre/AEMO-publishes-
final-report-into-the-South-Australian-state-wide-power-outage
[7] Initial operation of the Hornsdale Power Reserve Battery Energy
Storage System, AEMO, Available at: https://www.aemo.com.au/
Media-Centre/AEMO-Hornsdale-report https://www.aemo.com.au/
Media-Centre/AEMO-Hornsdale-reporthttps://www.aemo.com.au/
Media-Centre/AEMO-Hornsdale-report
[8] South Australia System Strength Assessment, AEMO, Available at:
http://www.aemo.com.au/Media-Centre/South-Australia-System-
Strength-Assessment

Cigre Science & Engineering • N°14 June 2019

34
7. Appendix

Regional boundaries of the National Electricity Market (NEM)

Cigre Science & Engineering • N°14 June 2019

35
 South Australian Power System

Cigre Science & Engineering • N°14 June 2019

36
8. Biographies
Dr. Nilesh Modi is a senior engineer at Australia Energy
Market Operators (AEMO). At AEMO, Nilesh has
worked across various teams including national planning,
electricity market operations and systems capability.
Nilesh has worked on various key projects/documents
published by AEMO, namely review of Black system
South Australia, system strength assessment for South
Australia, review of under-frequency load shedding
design for SA, system integrity protection scheme for
SA, system strength requirements and inertia requirement
methodology. He has been involved in investigating
system incidents that has resulted in large disruption
across the National Electricity Market (NEM). Nilesh
received his PhD from the University of Queensland for
which he has been awarded Dean’s excellence award for
RHD Excellence. Nilesh is a senior member of IEEE
and also adjunct research fellow at the University of
Queensland, Australia.
Dr. Babak Badrzadeh is the Manager of Operational
Analysis and Engineering team at AEMO, which develops
engineering solutions for a wide range of power system
security issues based on detailed power system modelling
and analysis. Babak is the convener of the Australian
Power System Modelling Reference Group, and has been
a key contributor to several CIGRE Technical brochures,
and IEC and Australian standards. He is currently the
Convener of CIGRE WG C2.26 and C4.56, and a recipient
of 2019 Energy Systems Integration Group’s Engineering
Excellence award. Babak holds BSc, MSc, and PhD
degrees all in the area of Power System Engineering. Prior
to joining AEMO in 2012, he was a Lead Power System
Cigre Science & Engineering • N°14 June 2019
37
Engineer with Vestas Technology R&D in Denmark
where he joined in 2010. Between 2007 and 2010 he
was with the Transmission and Distribution Division of
Mott MacDonald UK as a Senior Power System Analysis
Engineer
Sorrell Grogan is a senior engineer within the Operations
department of the Australian Energy Market Operator. He
works in the team determining the operational envelope
for running power systems with high levels of power
electronic based generation. He has played a key role in the
analysis of data captured during the 2016 South Australian
System Black event, and the development of large-scale
integrated electromagnetic transient simulation models
used to recreate extreme system events and to investigate
emerging dynamic phenomena.
Cheryl Noronha is a Senior Engineer at the Australian
Energy Market Operator (AEMO). Currently Cheryl
works in the Operational Analysis and Engineering team
at AEMO, primarily involved in assessment of system
restart, development of electromagnetic transient (EMT)
models of generators and large scale network models,
and conducting large scale EMT studies to investigate
extreme system events. Cheryl has notably been involved
in AEMO’s analysis and publication of the ‘South
Australian power system operation as a viable island’
report. Previously at AEMO, Cheryl was also involved
in assessing new and modified generator connections
utilising RMS and EMT simulation tools and determining
appropriate technical standards to apply for the new
connection. She is currently a member of the CIGRE
WG C2.26 (system restoration) and holds a Bachelor and
Masters of Electrical Engineering.
An aggregate dynamic model for distributed
energy resources for power system stability
studies
P. POURBEIK1, J. WEBER2, D. RAMASUBRAMANIAN3, J. SANCHEZ-GASCA4,
J. SENTHIL5, P. ZADKHAST6, J. C. BOEMER3, A. GAIKWAD3, I. GREEN7, S. TACKE8,
R. FAVELA9, S. WANG10, S.ZHU7 and M. TORGESEN11
Abstract:
As distributed generation penetration in power systems
around the world continues to increase there is a pressing
need for improved dynamic models for distributed
energy resources for use in large scale power system
simulation tools. This need has been heightened in the
past few years in North America. Thus, the Renewable
Energy Modeling Task Force of the Western Electricity
Coordinating Council embarked on this task in late 2016,
culminating with the development of a simple distributed
energy resource model called DER_A in 2018. This paper
describes this model in detail and demonstrates the testing
that was done to verify the implementation of the model in
several commercial software tools.
1. Introduction
The focus on expansion of renewable resources is a
continuing trend worldwide. Thus, wind and photovoltaic
(PV) generation continue to be deployed within the
power systems around the world at the transmission
level. However, another rapidly growing sector is the
deployment of distributed photovoltaic generation systems
at the residential and commercial level. In the case of large
utility scale wind and PV power plants, there has been
much effort in recent years to develop simple, generic and
publicly available dynamic models for simulating such
technologies in commercially available power system
simulation tools [1]. Also, with regards to wind generation,
there is an International Electrotechnical Commission
Working Group, nearing completion of an international
Corresponding author: pouyan@ieee.org
1- Power and Energy, Analysis, Consulting and Education, PLLC, Texas, USA
2- PowerWorld, Illinois, USA
3- Electric Power Research Institute, Tennessee, USA
4- General Electric, New York, USA
5- Siemens PTI, New York, USA
6- PowerTech Labs, British Columbia, Canada
7- California ISO, California, USA
KEYWORDS
Distributed Energy Resource Modeling
Cigre Science & Engineering • N°14 June 2019
38
standard set of public models for wind turbine generators
[2]. Also, CIGRE has recently published a Technical
Brochure on the subject of inverter-based generation
modeling [3].
With regards to the distributed energy resources (DER)
there has not been a simple dynamic model that is
available across multiple simulation platforms for use in
large power system simulation studies. One model was
developed many years ago called the PVD1 model [4].
This model, however, was not adopted by all commercial
software vendors in North America and had some known
limitations, from its inception, since at the time it was
developed the proliferation of DER was still limited and
the various standards related to DER were still under flux.
With the recent approval of the revised IEEE Standard
1547 in April, 2018 [5], and other similar standards such
as California Rule 2112, more functionalities like voltage
and frequency control are being proposed for DER. Hence,
within the Western Electricity Coordinating Council’s
(WECC) Renewable Energy Modeling Task Force
(REMTF), an effort was started to look at developing a new
model for modeling DER, to be ultimately incorporated
into the existing composite load model initially developed
in WECC13. This paper outlines the development of this
new model, gives a brief description of the model, and
provides a summary of the testing of the model in several
commercial power system simulation tools.
The remainder of this paper is organized as follows. In
section 2, is a brief outline of the model and its salient
features, as well as a description of how it is to be
8- Modesto Irrigation District, California, USA
9- El Paso Electric, Texas, USA
10- PacifiCorp, Oregon, USA
11- Western Electricity Coordinating Council, Utah, USA
12- http://www.cpuc.ca.gov/Rule21/
13- https://www.wecc.biz/_layouts/15/WopiFrame.aspx?sourcedoc=/Reliability/WECC%20
Composite%20Load%20Model%20Specifications%2001-27-2015.docx&action=default&D
efaultItemOpen=1
 1:T
Rfdr + j Xfdr
jXxf Bf1
Bss
DG
Figure 1: Location of DER plugging into the composite load model.
incorporated into the composite load model. In section 3,
a description is given of a detailed testing that has been
done of the model across four commercial software
platforms. Also, in section 3, a brief description is given
of running the model in large system studies, together
with the composite load model, to compare it with PVD1.
Finally, section 4 provides the conclusions and summary
as well as some brief comments on future work.
2. The DER_A Model
The ultimate purpose of the distributed energy resource
model version A (DER_A) is for it to be used to represent
the aggregated dynamic behavior of the DER in time-
domain positive-sequence stability studies. That is,
this model would represent the combined (aggregated)
dynamics behavior of many tens to hundreds of small
distributed inverter-based generators on the distribution
system on for example residential feeders – such as roof-
top photovoltaic generation. As such, it will eventually be
deployed as part of the composite load model, as shown
in Figure 1.
The concept behind development of the DER_A model was
to create a model that is able, to some extent, to emulate
the key dynamic performance that may be required from
such resources in the future, such as frequency and voltage
control. At first sight the 2nd generation generic renewable
energy source (RES) models [1], that were developed for
inverter-based generation, may seem appropriate to use to
model DER (i.e., repc_a + reec_a + regc_a). However,
there are two drawbacks with this approach (i) they
Cigre Science & Engineering • N°14 June 2019
M Motor A
Bf2 M Motor B
M Motor C
A/C Motor D –
represents 1-phase
residential a/c
Electronic
Load
Static
Load
constitute well over one-hundred parameters and so are
perhaps too complex for modeling aggregated DER, and
(ii) the 2nd generation models were developed for modeling
single large wind/PV and battery-energy storage plants and
so may not provide a simple means to represent aggregated
behavior across numerous distributed generators.
Thus, starting with the model structures of the 2nd
generation RES models (i.e., repc_a + reec_a + regc_a),
a significantly reduced version of the core functionality
was developed to form DER_A. Figure 2 shows the new
DER_A model. The complete parameter list for the model
is given in Table 1. The model has 48 parameters and 10
states, which is roughly 1/3 of the number of parameters
of the full large-scale 2nd generation RES generic models
[1]. Nonetheless, it preserves a significant number of those
features, namely frequency and voltage control emulation,
with asymmetric deadband. The voltage control only
allows for proportional control, while the model also
allows for constant power factor and constant Q-control.
It is possible, however, that both constant Q-control and
proportional voltage control or constant power-factor (pf)
control and voltage control are in effect simultaneously.
For example, assume the distributed generation is in
constant Q-control, holding a small lagging power factor
(or in constant pf-control at a small lagging pf), such that
it is generating 0.1 MVAr on a 2 MVA unit. Then assume
that Kqv = 10 (proportional voltage control gain) and
dbd1=dbd2=0.05 with Vref0 = 1.0. Now so long as the
voltage remains within 0.95 to 1.05 pu, the Q output of the
unit remains at 0.1 MVAr. If an event occurs to depress the
39
 dbd1, dbd2
_ Verr
Vt 1 Iqv
Kqv
1 + s Trv
Vt_filt
s0 +

Iql1

Iqh1
Vref0 (user defined)

pfaref tan
Iqmax
+
Pgen 1 + Iq
1 1 1
(Pord)
1 + s Tp 1 + s Tiq 1 + s Tg
Iqcmd

0.01
s1 Qref s2 s3
0 Iqmin
Pflag
Vt_filt Xe
(s0)

Vrfrac
~

Ed = V td o � iq X e
Eq = V tq o + id X e
+
Q priority Pqflag Current
1 j V
Iqmax = Imax 0 – Q priority Limit 0
Iqmin = -Imax 1 – P priority Logic
1
1 +
Ipmax = sqrt( Imax2 – Iqcmd2) 1 + s Tv
Vt_filt 0 1
P priority (s0) vl0 vl1 vh1 vh0 s4 Vtripflag
Ipmax = Imax
Iqmax = sqrt( Imax2 – Ipcmd2)
Iqmin = -Iqmax
Vt_filt Frequency Tripping
(s0) Logic See Details in

0.01
Model Description Id ( = Ip)
Freq_flag
0 Pmax rrpwr
Pref dPmax Ipmax
Pref
Pmax Ip
s7 1 1
+ femax
_ 1 + s Tpord Pord Ipcmd 1 + s Tg
1 Kpg + Kig
Pgen 1 + s Tp dPmin s8 s9
s Ipmin
(Pord) 1 -rrpwr
+ Pmin
femin s6
0 Pmin
fdbd1,fdbd2 Ddn
Frq_filt +
_
1
Freq
1 + s Trf
s5 + +
Dup
Freq_ref 0

0
To frequency relay model
1 1
If Vt (terminal voltage) ≤ Vpr then switch to
position 1, else position 0

Figure 2: The distributed energy resource model version A (DER_A).

voltage or raise the voltage outside of the deadband, then

the proportional control will act to increase/decrease Q 1
until the voltage comes back inside the deadband, at which
point Q drops back to its initial value. This is in keeping
with the main proposed concepts in IEEE Standard 1547,
A
though it is not an exact implementation of any specific
control strategy. There are of course other possible control
strategies, but this model being an implementation of
aggregated behavior, the group consensus was to keep it B
in the simplest format.
This model is intended primarily to be used as an aggregated
model of a large number of distributed generators. Thus,
the parameters Vrfrac, vl0, vl1, vh0, vh1, tvl0, tvl1, tvh0 and
tvh1 collectively allow for emulation of partial tripping 0
of the aggregated model. This is explained in detail, per
the pseudo code for this function in the appendix, and vl0 Vmin vl1
associated Figure 3. In this case the linear drop-off (shown Figure 3: Effect of the Vrfrac
in Figure 3) is intended to emulate the gradient of voltage (the code for this logic is provided in the Appendix)
along the feeder. The purpose of these timers is to allow
for the emulation of inverters disconnecting under low (or
high) voltage scenarios. For example, legacy technology integral controller (Kpg, Kig) is zero. The power reference
may disconnect quickly for a small voltage dip (i.e. one may Pref will initialize to Pord, and the frequency error is
set vl1 = 0.9 and tvl1 = 0.1 s) while part of the aggregate zero. Now if Freq_flag = 1 and a fault occurs nearby
model may be representing modern inverters that comply which results in partial tripping of the “aggregated” DER
with newer standards where it will not disconnect unless through the action of the Vrfrac logic, then the terminal
the voltage drops significantly for a longer duration (e.g. electrical power of DER_A will go down. Thus, if the
one may set vl0 = 0.5 and tvl0 = 1 s). Thus, this is to allow terminal electrical power is fedback, then the error into
for testing various aspects of standards such as IEEE Std the proportional-integral controller (Kpg, Kig) would
1547 requirements and California Rule 21. now become positive and Pord will increase until it hits
For the emulation of primary-frequency response in Pmax, or until the electrical power output of the model is
DER_A, the feedback signal (Pgen) is taken from the again equal to Pref. This is not appropriate, since there has
power-order (Pord) and not the terminal of the model (see been no system frequency deviation and also the model
Figure 2). This is because, in steady-state with frequency should not attempt to restore the power lost due to partial
at its nominal value, the error into the proportional- tripping effected by Vrfrac. Therefore, by taking the

Cigre Science & Engineering • N°14 June 2019

40
 Frq_filt
_
1
Freq
1 + s Trf
s5 +
Freq_ref
0 Frqm
1 1
If Vt (terminal voltage) ≤ Vpr then switch to
position 1, else position 0
Figure 4: Frequency tripping logic.
power feedback from the power-order (Pord) prior to the
Vrfrac block, this problem is avoided. This ensures that
Pord is always equal to Pref, which is what is desired.
Furthermore, the user should be allowed to set Tpord and
Tp to zero (0). By doing so and setting Kpg = 0 and using
a non-zero value of Kig, a simple proportional only droop-
control can be effected, since the closed loop around Pord
in this case creates a simple time-constant equal to 1/Kig.
In this case, i.e. when Tpord = Tp = 0, Kig cannot be set to
zero, it must be a positive number. For similar reasons, the
feedback to the power factor controller is also from Pord.
The frequency tripping is modeled in simple terms. If
frequency goes below fl for more than tfl seconds, then
the entire model will trip. If frequency goes above fh for
more than tfh seconds, then the entire model will trip. This
block is disabled, if voltage is below Vpr, to avoid tripping
on frequency spikes (as calculated in simulation) due to
sudden voltage drops. This is depicted in Figure 2, and
shown in more detail, in an expanded view, in Figure 4.
The model may also be used to emulate inverter-interfaced
distributed energy storage. This is achieved by allowing
the model to absorb, as well as generate, real power. The
additional flag, typeflag, achieves this. When typeflag is set
to 0, meaning the device is a generator, Ipmin (that is, the
minimum active current limit) is set to zero. When typeflag
is set to 1, meaning the device is a storage device, Ipmin is
set to -Ipmax. Need-less-to-say, for the sake of simplicity,
there is no attempt to model the storage mechanism (e.g.
charging/discharging of a battery) and so it is assumed that
the model would only be used for transient simulations
(e.g. 10 to 30 seconds) during which there would be no
appreciable effect of the storage mechanism.
A simple representation of the voltage source interface
that is employed by most equipment vendors (based on
[6]) is also modeled, since by far the majority of inverters
used for inverter-based resources are current-regulated
voltage-source converters.
The details of initializing the model can be found in the
Cigre Science & Engineering • N°14 June 2019
To controls
Frequency Trip Logic:
If Frqm < fl for more than tfl seconds then trip model
If Frqm > fh for more than tfh seconds then trip model
WECC model specification document [7] but the most
salient points are summarized here for completeness.
The values of Vt and Pgen are the voltage and electrical
power at the terminals of the DER_A model. Upon
initialization, Pref and Qref will be determined in software
to properly initialize the model. If Kqv is non-zero, then
upon initialization dbd1 < Vt - Vref0 < dbd2, where Vref0,
dbd1 and dbd2 are user defined value. If this condition
is not met, then the software tool will force Vref0 = Vt
and indicate this to the user in a warning message. If
dbd1=dbd2=0 (which should typically not be done, since
these distributed generation models are not intended to
tightly control voltage) and Kqv is non-zero, then the
program should give a warning/error message to the user
and indicate that Vref0 has been set to equal to Vt (to force
the error to zero and thus the output of the voltage leg
to zero); the initial Q from power flow is then initialized
off of the constant Q/pf leg. This is the simplest solution
in this case. Finally, during initialization, the software
program should check to ensure that the terminal voltage
(Vt) of the model initializes to a value that is greater than
vl1. Also, vl1 must be greater than or equal to vl0. If either
of these conditions are not met, the program will present
an error message to the user indicating that the value of
vl1 and vl0 are inappropriate, and thus the model will
ignore the Vrfrac block. A similar check should be made
on vh1 and vh0. Also, a check should be made to ensure
that tvl1, tvl0, tvh1and tvh0 are all greater than or equal to
zero. There is no limitation on which of these timer values
should be greater or smaller. The Vrfrac block is explained
in more detail in the appendix.
There is a possible control problem. If this model were
used to model a single large inverter-based device
connected to a weak grid point (i.e. low short-circuit ratio)
where the voltage is highly affected by this device, then
there could be a possibility for limit-cycling (i.e. voltage
goes outside deadband, device brings voltage inside
deadband by changing Q, Q drops to constant initial value
once voltage is within the deadband, voltage goes outside
41
 Figure 5: Benchmarking test case model.

deadband, etc.). For this, and other reasons, it is in general,
not recommended that this model be used to model large
plants.
Two other important notes should be made. First, that the
filtered value of voltage (Vt_filt) and frequency (Frq_filt)
is used in all the controls and timers. Second, that the
current limit is modeled as follows:
a. Q-priority: Iqmax = Imax; Iqmin = -Imax;
Ipmax = Imax 2 � Iqcmd 2 ; if typeflag = 0 then
Ipmin = 0, else Ipmin = - Ipmax
b. P-priority: Ipmax = Imax; Iqmax =
Imax 2 � Ipcmd 2 ; Iqmin = -Iqmax; if typeflag = 0
then Ipmin = 0, else Ipmin = - Ipmax
A final note is that the post-fault rate of recovery on active-
current (rrpwr) is also imposed (in the opposite direction)
when the model is being used to “emulate” charging of
an energy storage device. That is, when Pgen is negative,
then rrpwr is applied with its sign changed and it becomes
the ramp-rate at which charging power (power being
absorbed by the model) increases after a fault.
3. Testing the DER_A Model in
Four Commercial Simulation
Platforms
Once the model was defined and agreed to, four major
commercial software vendors in North America, decided
to graciously implement the model in their respective
software tools so that it could be tested. This testing was
done in two steps. First, a beta version of the model was
released by all the software vendors and a set of test
protocols were defined [8]. The tests aimed, to the extent
possible, to test all the features of the model and ensure
that they performed as expected, as well as to benchmark
the model across the four commercial platforms to ensure
consistent implementation and performance across all
the software tools. A simple test case was developed
for performing simulations in all the various software
platforms, it is shown in Figure 5. A complete list of
all the tests and results may be found in a report by the
Electric Power Research Institute (EPRI) [8]. These tests
were focused on testing the DER_A model as a standalone
model. The testing proved successful and so the model

Figure 6: Plots from one of the benchmarking tests – Test 1A – voltage sag with a ramped recover.

Cigre Science & Engineering • N°14 June 2019

42
 a) Voltages (pu) b) Frequency (Hz)

a) Net Load in the local area (MW) b) Total distributed generation in the area (MW)

Figure 7: Simulation results form a large system model (WECC system) where the distributed generation is represented within the composite load model first using
the old PVD1 model and then using the new DER_A model.

was approved and released on all the software tools.
Figure 6 shows one example of the many tests performed.
This testing was lead and performed by EPRI. During
the testing one case was found to have some small, but
noticeable, difference in response among the tools, where
two of the tools matched and the others had a slight
difference. This was for the case of playing into the model
a frequency wave-form. Upon closer investigation, it
was identified that the differences were due to numerical
precision of the integration schemes, and thus for now
this was not further investigated. Some of these subtle
differences in the frequency calculation can actually be
seen in the frequency traces in Figure 6.
The next step in testing the model was to incorporate it
into the composite load model (Figure 1) and to then test
it by using the model to simulate distributed generation
across a large system. This testing was done by one of
the task force members [9], by taking a WECC base case
and simulating some major system events in one region
with (i) all the distributed generation initially modeled
using the old PVD1 model [4], and (ii) by replacing all the
PVD1 models in the composite load model with the newly
developed DER_A model. Furthermore, some sensitivities
were performed on the model parameters. Example plots
from this work are shown in Figure 7. The conclusions
that may be drawn from this analysis, as seen in Figure 7,
are as follows:
1. The DER_A model seems to perform well in a large
system model.
2. If the parameters of the DER_A are properly adjusted,
it can be made to emulate the older, and much simpler,
PVD1 model – this can be seen by the fact that the
brown and blue lines in the simulations (Figure 7)
match for the total net load and distributed generation
in the area, which is driven by the performance of these
models.
3. Having the DER modeled as a part of the composite

Cigre Science & Engineering • N°14 June 2019

43
 load becomes more critical with the higher penetration
of the behind-the-meter distributed generation. The
earlier and simpler models for these resources may
potentially give less accurate and less realistic results
since they do not have the time constants and time-
dependent tripping logic in the DER_A. This can be
seen by the fact that in the simulations the green lines
(where the time constants in the DER_A model have
been varied to more closely mimic rule 21) show
significantly different total distributed generation
response as compared to the brown/blue lines (where
the PVD1, and DER_A made to mimic PVD1, model is
used).
4. Additional research is needed to better understand
how to parameterize the DER_A model, to perform
sensitivity studies to understand the sensitivity of
study results to the various parameters and thus which
parameters are most critical.
4. Conclusions and Future Work
A new proposed distributed energy resource model has
been presented here. This model has now been adopted by
several commercial software vendors in North America,
and more recently by at least one software vendor in
Europe. It has been shown, through testing the model
in four of the commercial software tools that consistent
and appropriate results can be obtained across the
software tools. Furthermore, initial simulations in a large
system model have shown reasonable results as well as
consistency with the older, and simpler models.
All of the above said, the task force that developed this
model fully realizes the following challenges and ongoing
research and development that is needed:
• To perform research on how to best/better parametrize
the model to suitably represent the aggregated
behavior of distributed generation in a system for both
(i) existing distributed generation, and (ii) for future
planed distributed generation.
• To look at the sensitivity of large system simulation
results to the various parameters of this aggregated
model in order to better understand the sensitivity of
system performance to the various model parameters.
Such work may lead to identifying aspects of the
model that need to be refined or changed in the future
in order to better model the actual aggregated behavior
of distributed generation. Thus, it is fully understood
Cigre Science & Engineering • N°14 June 2019
44
that the model as it stands may have limitations and
there may be aspects that require refinement as greater
experience is gained with distributed generation and as
the technologies evolve.
• To look more closely at the frequency response
capability modeling within the DER_A model. This
aspect has not been fully tested and may not be fully
representative of the aggregate response of distributed
generation. To date, to our knowledge, there is no
deployment of distributed generation in North America
with primary frequency response capabilities, and
so this will require further and future work to more
properly test and refine in the DER_A model.
5. References:
[1] P. Pourbeik, J. Sanchez-Gasca, J. Senthil, J. Weber, P. Zadehkhost,
Y. Kazachkov, S. Tacke and J. Wen, “Generic Dynamic Models for
Modeling Wind Power Plants and other Renewable Technologies
in Large Scale Power System Studies”, IEEE Trans. on Energy
Conversion, September 2017, available on-line: https://ieeexplore.
ieee.org/document/7782402/
[2] Ö. Göksu, P. Sørensen, J. Fortmann, A. Morales, S. Weigel, P.
Pourbeik, “Compatibility of IEC 61400-27-1 Ed 1 and WECC 2nd
Generation Wind Turbine Models”, Conference: 15th International
Workshop on Large-Scale Integration of Wind Power into Power
Systems as well as on Transmission Networks for Offshore Wind
Power Plants, November 2016.
[3] CIGRE TB 727, Modelling of inverter-based generation for
power system dynamic studies, May 2018. https://e-cigre.org/
publication/727-modelling-of-inverter-based-generation-for-
power-system-dynamic-studies
[4] WECC Generic Solar Photovoltaic System Dynamic Simulation
Model Specification, 2012 https://www.wecc.biz/Reliability/WECC-
Solar-PV-Dynamic-Model-Specification-September-2012.pdf
[5] 1547-2018 - IEEE Standard for Interconnection and Interoperability
of Distributed Energy Resources with Associated Electric Power
Systems Interfaces https://ieeexplore.ieee.org/document/8365917/
[6] D. Ramasubramanian, Z. Yu, R. Ayyanar, V. Vittal and J. M.
Undrill, “Converter Model for Representing Converter Interfaced
Generation in Large Scale Grid Simulations”, IEEE Trans. PWRS,
April 2016.
[7] DER_A Model Specification: https://www.wecc.biz/Reliability/
DER_A_Final.pdf
[8] EPRI Report, The New Aggregated Distributed Energy Resources
(der_a) Model for Transmission Planning Studies, Product
ID: 3002013498, May 8, 2018. https://www.epri.com/#/pages/
product/3002013498/
[9] I. Green, CMPLDWG Composite Model with Distributed
Generation DER_A, presentation at the WECC MVWG
Meeting, May, 2018 https://www.wecc.biz/Administrative/07%20
CMPLDWG%20Composite%20Model-%20Green.pdf
 Table 1: Model parameter list

Parameter Description
Trv transducer time constant (s) for voltage measurement
Trf transducer time constant (s) for frequency measurement (must be ≥ 0.02 s)
dbd1 lower voltage deadband ≤ 0 (pu)
dbd2 upper voltage deadband ≥ 0 (pu)
Kqv proportional voltage control gain (pu/pu)
Vref0 voltage reference set-point > 0 (pu)
Tp transducer time constant (s)
Tiq Q control time constant (s)
Ddn frequency control droop gain ≥ 0 (down-side) (pu/pu)
Dup frequency control droop gain ≥ 0 (up-side) (pu/pu)
fdbd1 lower frequency control deadband ≤ 0 (pu)
fdbd2 upper frequency control deadband ≥ 0 (pu)
femax frequency control maximum error ≥ 0 (pu)
femin frequency control minimum error ≤ 0 (pu)
Pmax Maximum power (pu)
Pmin Minimum power (pu)
dPmax Power ramp rate up > 0 (pu/s)
dPmin Power ramp rate down < 0 (pu/s)
Tpord Power order time constant (s)
Kpg active power control proportional gain (pu/pu)
Kig active power control integral gain (pu/pu/s)
Imax Maximum converter current (pu)
vl0 voltage break-point for low voltage cut-out of inverters (pu)
vl1 voltage break-point for low voltage cut-out of inverters (pu)
vh0 voltage break-point for high voltage cut-out of inverters (pu)
vh1 voltage break-point for high voltage cut-out of inverters (pu)
tvl0 timer for vl0 point (s)
tvl1 timer for vl1 point (s)
tvh0 timer for vh0 point (s)
tvh1 timer for vh1 point (s)
Vrfrac fraction of device that recovers after voltage comes back to within vl1 < V < vh1
fl frequency break-point for low frequency cut-out of inverters (pu)
fh frequency break-point for high frequency cut-out of inverters (pu)
tfl timer for fl (Tfl > Trf) (s)
tfh timer for fh (s)
Tg Current control time constant (s)
rrpwr Power rise ramp rate following a fault > 0 (pu/s)
Tv time constant on the output of the voltage/frequency cut-out (s)
Vpr voltage below which frequency tripping is disabled (pu)
Pflag 0 - for constant Q control, and 1 - constant power factor control
Pqflag 0 - Q priority, 1 - P priority for current limit
Freq_flag 0 - frequency control disabled, and 1 - frequency control enabled
Ftripflag 0 - frequency tripping disabled; 1 - frequency tripping enabled
Vtripflag 0 - voltage tripping disabled; 1 - voltage tripping enabled
typeflag 0 - the unit is a generator Ipmin = 0; 1 - the unit is a storage device and Ipmin = - Ipmax
Xe Source impedance reactive > 0 (pu)
Iqh1 Maximum limit of reactive current injection (pu)
Iql1 Minimum limit of reactive current injection (pu)

Cigre Science & Engineering • N°14 June 2019

45
6. Appendix: Pseudo code for tvl1 times out. That is, Vmin is the lowest point of Vt
during a simulation, but at the moment that timer tvl1
the Vrfrac block. times out it is set to (and kept at) the value of Vt at that
The block shown in Figure 3 is implemented consistent instant. This is done to avoid jumps in the response due to
with the existing PVD1 model, as described in a WECC movement (oscillations) in voltage. For example, consider
report [4]. However, the pseudo code and logic here is the following scenario. During an event Vt goes down to
quite different to that in [4], since we have added two (2) Vmin_a, then comes back up to Vt_2, and then goes again
time parameters tvl0 and tvl1, which determine when the down to Vmin_b, at which time the timer for tvl1 times
limits are imposed once the assigned time has lapsed. That out. Thus, it is the value of Vmin_b which we would like
is, the output of the block will always track the path of Vmin to be set to. This is depicted below in Figure 8.
the black line in Figure 3, unless certain conditions are
met. If the voltage stays below vl1 for a duration greater
than tvl1, then it will now always follow the path of the Vmin should initialize to the initial value of Vt or a default
red line when the voltage recovers. If the voltage stays value (e.g. 1.0).
below vl0 for greater than tvl0, then the output will always
Timer 1 = 0
remain at zero. In order to reduce this block back to that
implemented in PVD1, one would have to set Trv = 0 Timer 2 = 0
(eliminate the filtering of voltage) and set tvl0 = 999, tvl1
Counter 1 = 0
= 0.0, tvh0 = 999, and tvh1= 0.0.
Counter 2 = 0
1 If Vt < vl1 and Timer 1 = 0
Start Timer 1
elseif Vt > vl1 and Timer 1 started
A Reset Timer 1
end
If Vt < vl0 and Timer 2 = 0
B Start Timer 2
elseif Vt > vl0 and Timer 2 started
Reset Timer 2
end
if Vmin <= vl0
0
Vmin_a Vt_2 Vmin = vl0
end
vl0 Vmin = Vmin_b vl1
if Vt <= vl0 or Counter 2 = 1
Figure 8: Understanding how Vmin is determined. Multiplier = 0.0
Note that Vmin in Figure 3 is not an input parameter, it elseif Vt <= vl1 and Counter 1 = 0
is an internal software variable which is keeping track
Multiplier = (Vt – vl0) / (vl1 – vl0)
of the minimum voltage that the terminal of the model
reaches during a simulation, immediately after the timer elseif Vt <= vl1 and Counter 1 = 1

Cigre Science & Engineering • N°14 June 2019

46
Multiplier = ((Vmin – vl0) + Vrfrac * (Vt - Vmin)) / (vl1 7. Biographies
– vl0)
Pouyan Pourbeik received his BE and PhD in Electrical
elseif Vt >= vl1 and Counter 1 = 0 Engineering from the University of Adelaide, Australia in
Multiplier = 1 1993 and 1997, respectively. From 1997 to 2000 he was
with GE Power Systems. From 2000 to 2006 he was with
else
ABB Inc. From 2006 to March, 2016 he was with EPRI.
Multiplier = Vrfrac * ((vl1 - Vmin) / (vl1 – vl0)) + ((Vmin From April, 2016 he is with Power and Energy, Analysis,
– vl0)/(vl1 – vl0)) Consulting and Education, PLLC. He is an Honorary
Member of CIGRE and a Fellow of the IEEE, and a past
end
chairman of both CIGRE Study Committee C4 – System
if Counter 1 = 0 Technical Performance, and the IEEE PES Power System
if Timer1 > tvl1 Dynamics Performance Committee.

Counter 1=1 James Weber received the B.S. degree in electrical

Vmin = Vt
end
end
if Counter 2 = 0
if Timer2 > tvl0
Counter 2=1
end
end
The key here is that Counter 1 (2) get set only if the
condition of being below vl1 (vlo) is met for the given
time duration and once that condition is met the block
remains in that state indefinitely. Also, Vmin is set to the
value of Vt at the point when timer 1 (tvl1) times out.
The same logic is then implemented for Vt exceeding
vh1 while keeping track of the maximum voltage reached
during the simulation (Vmax) (see Vrfrac in Figure 2).
Note that if in a single simulation both a voltage dip and
a voltage rise is experienced, then the two arms of the
Vrfrac block simply multiply by each other. That is, for
example, if one first goes into a voltage dip, then coming
out of the dip the total magnitude of the block is affected
by Vrfrac, as determined by the results of the voltage dip.
This then becomes the value that goes into the voltage rise
scenario and is then affected by the Vrfrac determined by
the voltage rise logic.
engineering from the University of Wisconsin-Platteville
in 1995 and the M.S. and Ph.D. degrees in electrical
engineering from the University of Illinois at Urbana-
Champaign, USA, in 1997 and 1999. He is the Director
of Software Development at PowerWorld Corporation in
Champaign, IL, USA.
Deepak Ramasubramanian received the B. E. degree
from PESIT Bangalore, India in 2011, the M. Tech.
degree from Indian Institute of Technology Delhi, New
Delhi, India in 2013, and a Ph.D. degree from the Arizona
State University, Tempe, USA in 2017. He is presently
an Engineer/Scientist III with the Grid Operations and
Planning Group of Electric Power Research Institute,
Knoxville, USA.
Juan J. Sanchez-Gasca is a Technical Director at GE’s
Energy Consulting Department in Schenectady, NY,
where he has been involved in the study of power system
dynamics and in the development of dynamic models for
transient stability analyses. He holds a PhD in Electrical
Engineering from the University of Wisconsin-Madison.
He is an IEEE Fellow and a past Chair of the IEEE PES
Power System Dynamics Performance Committee.
Jay Senthil received his PhD in Electrical Engineering
from IIT Kanpur, India in 1992. From 1992 till 2001 he
worked for BHEL (1992-1995), the University of Western
Ontario, Canada (1995-1996), CAE Inc., Canada (1996-
1999), ABB (1999-2001). Currently (since 2001) he has
been with Siemens PTI in Schenectady, NY where he has
been responsible for the development, maintenance and
customer support for the dynamic simulation engine of
Siemens PTI PSS®E.

Cigre Science & Engineering • N°14 June 2019

47
Pouya Zadkhast is the Team Lead of Application Delivery
group at Powertech Labs and he has a Ph.D. in Electrical
and Computer Engineering from the University of British
Columbia. Dr. Zadkhast is the key developer of TSAT-
RTDSTM Interface (TRI), an advanced Co-Simulation
platform, and he has delivered multiple projects in the
area of renewable generator and FACTS modeling. His
area of interest includes dynamic simulation, small-signal
analysis, and online dynamic security assessment of
power systems.
Jens Boemer received the Diploma degree in electrical
engineering from the Technical University of Dortmund,
Dortmund, Germany, in 2005, and the Ph.D. degree from
the Delft University of Technology, Delft, the Netherlands,
in 2016. He is currently a Principal Technical Leader
with the Department of Grid Operations and Planning,
Modeling and Simulation, Electric Power Research
Institute (EPRI), Palo Alto, CA, USA. His field of interest
includes the grid integration of renewable and distributed
energy resources with the focus on power system stability.
Anish Gaikwad received his B.E. and M.S. degrees from
N.I.T. Nagpur and Mississippi State University in 1997
and 2002 respectively. He is currently a Senior Project
Manager in the Grid Operations & Planning research
group at EPRI, and leads the EPRI efforts on many
aspects of modeling and model validation research and
development.
Irina Green received here BE and MSEE in electrical
engineering from the Peter the Great St. Petersburg
Polytechnic University, Russia. She has extensive
experience in transmission planning and power system
analysis. She was a senior transmission planner with
Pacific Gas & Electric from 1997 to 1999. Since 1999 she
has been with the California ISO, and presently serves in
the role of Senior Advisor. In this role she helps to identify
and resolve engineering-related problems within assigned
areas of the CAISO-controlled grid, and extensively works
on the integration of renewable resources and distributed
energy resources.
Spencer Tacke received his BS and MS degrees in electrical
engineering from the University of California at Berkeley,
Cigre Science & Engineering • N°14 June 2019
48
and the University of Southern California, Los Angeles,
CA. He has held various power system engineering
positions with the U.S. Bureau of Reclamation (Hoover
Dam), General Electric, ITT-Jennings, Westinghouse,
Landis & Gyr Systems (Siemens), and the Modesto
Irrigation District.
Roberto Favela received his BS in Electrical Engineering
from California State Polytechnic University, Pomona in
1991. He is a Principal Engineer in the System Planning
and Interconnections department at El Paso Electric, serves
as the company’s expert in the transmission planning and
Interconnection Studies area and provides authoritative
consultation and recommendations to senior management.
He has held various power system engineering positions,
including superintendent of Distribution System
Engineering, with LADWP, Southern California Edison
and El Paso Electric.
Song Wang received his BS and MS in Electrical
Engineering from the Shenyang University of Technology,
China in 1993 and 2001, respectively. He is a Principal
Engineer in the transmission planning department
at PacifiCorp, serves as the company’s expert in the
transmission planning area and provides authoritative
consultation and recommendations to senior management.
He is chairman of the Western Electricity Coordinating
Council’s Modeling and Validation Working Group.
Songzhe Zhu received her BSEE from Xi’an Jiaotong
University, China in 1993 and MSEE from Nanjing
Automation Research Institute, China in 1996, and Ph.D.
from Iowa State University, USA in 2000. She is currently
a Senior Advisor with Regional Transmission at California
ISO.
Matthew Torgesen received his BS in Electrical
Engineering from Utah State University in 2013. He has
worked for several key industrial companies in a power
engineering capacity. From 2014 to 2018 he was with the
Western Electricity Coordinating Council (WECC) and
supported the Modeling and Validation Working Group.
He is currently an E&I specialist and Power Engineering
lead at Bayer’s phosphorus facility in Soda Springs Idaho.
Wire displacement of AAAC overhead line
conductors at wedge tension clamp

P. B. BUEHLMANNa,*, D. SCHENKa, E. MAZZAb,c , C. M. FRANCKa

High Voltage Laboratory Swiss Federal Institute of Technology ETH Zurich,

a

Institute of Mechanical Systems Swiss Federal Institute of Technology ETH Zurich,

b

c
Mechanical Integrity of Energy Systems Swiss Federal Laboratories for Materials Science and
Technology EMPA Duebendorf,
Switzerland

Abstract Symbols
α Total wedge bevel angle
Overhead line conductors are tensioned to fulfill sag
µ Friction coefficient
limitations. Free-span conditions of common conductors
C1 , C2 Constants for linear fitting function
are investigated and documented in literature. In this
φ Azimuth angle
paper, the structural changes of an all aluminum alloy
A Cross-section
overhead line conductor near a wedge tension clamp
D Conductor diameter
are investigated by x-ray tomography, digital image
d Wire diameter
correlation and finite element analysis. The results
F Tensile force
achieved with these three methods consistently show
l Length
that the inner layers of the conductor are slipping relative
lmech Mechanical transition length
to the outer layers of the conductor. Additionally, not
r Radial coordinate
all wires on the same layer are clamped equally. Thus,
uφ Azimuthal displacement in φ-direction
there is a mechanical transition zone of approximately
ur Radial displacement in r-direction
300 mm between the tension clamp and the free-span.
uz,ref Axial displacement relative to the axial
This mechanical transition zone and the associated
displacement of the reference point uzref.point |10kN
structural changes of the conductor are measured and
of wire 1 on layer 4 at z = 55 mm at 10 kN
modeled. The tools used in this study can be applied on
uz Axial displacement in z-direction
other conductors and clamps as well.
z Axial coordinate
Nomenclature
1. Introduction
Acronyms
1.1. Motivation
AAAC All aluminum alloy conductor
Reliable electric power transmission is essential for the
ACSR Aluminum conductor steel-reinforced
power grid of today and the future. This has evolved
ACT Adapted computed tomography
over many decades and is designed according to local
CCE Clamped conductor end
constraints and preferences. Thus, different kind of
CT Computed tomography
conductors are used.
DIC Digital image correlation
FCE Free-span conductor end On a global scale, aluminum conductor steel reinforced
FEA Finite element analysis (ACSR) are the most commonly used conductors
FIT Fitting curve for overhead lines (OHL). The steel core provides
LW Lower wedge more mechanical strength and is stiffer than the outer
OHL Overhead line aluminum layers. Thus, the conductor is less exposed to
UW Upper wedge elastic and inelastic deformation.

* pascalbuehlmann@outlook.com

KEYWORDS
All Aluminum Alloy Conductor (AAAC), Adapted Computed Tomography (ACT), overhead line, wedge tension
clamp.

Cigre Science & Engineering • N°14 June 2019

49
Less prevalent are all aluminum alloy conductors
(AAAC) which are used in Switzerland and other
countries. As these conductors are of fewer occurrence,
less investigations were performed compared to ACSR.
In comparison to steel-reinforced conductors, AAAC
are less susceptible to corrosion. But AAAC are more
affected by creep since all tensile load is applied on
the aluminum alloy wires. This creep behavior of
AAAC was investigated on single wires and the entire
conductor in [1, 2] mainly for conditions representative
of the free-span. These investigations were conducted
to better predict the lifetime of this conductor type. The
creep behavior of ACSR is documented in many other
papers, e.g. [3, 4].
A lot of knowledge and literature about the creep of the
different kind of conductors with different kind of alloys
exist. To additionally benefit from all this expertise, this
paper focuses on methods to determine the mechanical
behavior of individual wires near clamps. The goal of the
investigations presented here, is to develop a predictive
model that allows applying the know-how of creep
behavior to specific locations as clamps.
At suspension clamps, the conductor is exposed to high
tensile forces on both sides. Thus, the tensile force is
carried by the whole conductor cross-section. But in the
case of tension clamps, all tensile load is transferred from
the conductor to the clamp through the outermost layer
of the conductor. In the transition zone between free-
span with high tensile load and tension free conductor
behind the tension clamp, the tensile force is transferred
by friction from layer to layer and through the outermost
layer to the clamp. Since there is no form fit, wires of
the conductor can move relative to each other. Thus, it is
hypothesized that the inner layers can move relative to
the outer layers. This behavior could cause an increased
tensile load on the outermost wires which are in contact
with the clamp. Over time, these outermost wires creep
and thus elongate more than the ones in the inner layers.
Hence, the inner layers could be loaded more after time
which cause these wires to creep as well. This layer
staged loading is different to the conditions in the free-
span. Because of pitch angle and layer diameter, the
length of wire per unit length of conductor is larger for
the outer layers. Hence, in the free-span the inner layers
are generally loaded more than the outer ones. To enable
an investigation on this layer staged loading at tension
Cigre Science & Engineering • N°14 June 2019
50
clamps, determining the displacements on all wire layers
is desired. Therefore, three approaches to determine
wire displacements near tension clamps are presented in
this study. A finite element analysis (FEA) model was
developed and compared to two experimental methods
which can quantify the occurring relative displacement
of different wire layers and wires on the same layer.
The mechanical long-term behavior of the conductor
near the clamp can then be characterized by combining
the results of this study, the creep model of the specific
material or conductor, and the temperature data. Merging
results from these three outcomes is not part of this
paper. But the results presented here shall enable it.
The conductor temperature in the free-span can be
accurately determined by models summarized in [5].
From the temperature in the free-span, the conductor
temperature at the clamp can be computed by the model
presented in [6, 7]. With the temperature and tensile force
history or corresponding assumed scenarios, the lifetime
expectancy of the conductor near the tension clamp can
be obtained and compared to the situation in the free-
span. It could be quantified, how critical the situation at
the tension clamp can be for conductor creep.
Since AAAC are more exposed to creep than ACSR
and other reinforced aluminum conductors, the study is
performed on a AAAC at a wedge tension clamp, where
all tensile force is transferred from the conductor to the
clamp.
However, the here investigated slipping of inner wires,
the conductor bulging due to clamp compression, and the
unequally distributed wire stresses due to asymmetrical
clamping may occur on other conductors and clamps as
well. Especially on wedge type clamps but as well on
other kinds of tension clamps, these three phenomena
may occur, but in varied magnitude. Further testing
would be required for those applications. On other
clamp types such as suspension clamps, the investigated
behavior presumably differs significantly. Nevertheless,
the used techniques can be applied on many other cases.
In addition, these methods may enable to investigate
other critical situations of clamped rope structures.
Especially at suspension clamps, conductors are exposed
to fatigue and vibration issues [8–14]. This phenomena
is neither considered nor discussed in this paper.
 Figure 1: Simplified clamp assembly sketch. On the top, the cross-section
along the symmetry rz-plane is shown with the bevel angle α = 5° of the lower
wedge. The origin of the cylindrical coordinate system (r, φ, z) is located at the
conductor center at the ultimate point of full contact with both wedges. The
z-axis points towards the free-span.

1.2. Previous Relevant Studies as shown previously. In this paper, mechanical

1.2.1. X-ray Imaging of Conductors
Pouliot et al. [15] developed a method to test conductors
in the field for ruptured wires on inner layers near
suspension clamps. They used a mobile x-ray source
and digital detector, both mounted on a portable
system. With the chosen setup, broken strands can be
detected non-destructively. With further adaptions to
our requirements, this approach is used in our study
to measure wire displacement on the inner conductor
layers.
1.2.2. Digital Image Correlation on Conductors
Xiang et al. [16] presented how deformation of metallic
strands and wire ropes under axial load can be measured
with Digital Image Correlation (DIC) of speckle patterns
applied on the surface. This technique serves as an
additional measurement for detecting wire displacement
on the outermost layer of the conductor.
1.2.3. Finite Element Conductor Models
Finite element analyses on conductors were performed
with different approaches. One-dimensional beam
elements, volumetric three-dimensional elements as
well as a combination of both kind of element types
were used to discretize the wires of the conductor or
specific contact regions of it [1, 2, 12, 13, 17–20]. These
models consider free-span conditions or situations near
suspension clamps. The contact between suspension
clamp and a wire was modeled in more detail by [9].
2. Methods
The mechanical behavior in the free-span and near
suspension clamps is investigated in several studies
displacements near the tension clamp are analyzed to
better understand how critical this location can be.
All tensile force from the conductor is transferred to
the clamp through the outermost layer. Therefore, it is
assumed that the inner layers can move relative to the
outer layer as those are only held back by friction forces
and not by any form fit. This can cause higher loads on
the outermost layer.
To investigate the wire displacements near the wedge
tension clamp, one numerical and two experimental
approaches were taken. With all three methods, the wire
displacements in axial z-direction can be determined
and compared to each other. Two different experimental
measurement techniques are compared to the FEA
models, which help to better understand the mechanical
behavior of the conductor. The two experimental
procedures Adapted Computed Tomography (ACT)
with x-rays and optical measurement with Digital
Image Correlations (DIC) as well as the FEA model are
described in this chapter.
To measure the assumed relative movements between
individual wires and layers, a conventional strain
measurement with few strain gauges is not sufficient.
Furthermore, strain gauges would be destroyed by the
high inter-layer contact pressures.
2.1. Conductor, Clamp, and Test Rig
The three methods are applied on the same AAAC
and the same wedge tension clamp. The chosen parts
are commonly used on typical transmission OHLs in
Switzerland.

Cigre Science & Engineering • N°14 June 2019

51
To enable a direct comparison between the two
experimental methods, exactly the same conductor and
wedge tension clamp specimen is used first for the ACT
and sequentially for the DIC method. Thus, the markers
used for ACT on the outermost layer can be identified
on the DIC results. The DIC experiments were prepared
without disassembling the conductor and the clamp.
The diameter of all conductor wires is d = 3.39 mm.
Four layers are stranded crosswise on the core wire. The
pitch lengths for the layers 1 to 4 are: 133 mm, 213 mm,
266 mm, and 322 mm. This results in a total conductive
cross-section of A = 550 mm2 and an outer conductor
diameter D = 30.51 mm. Material tests on single wires
and the complete conductor were performed on specimen
from the same conductor from the same drum. The tests
are documented in [1, 2].
All three parts of the wedge tension clamp are made
of forged aluminum. The upper wedge is straight and
the lower wedge is beveled by α = 5° as illustrated in
Figure 1. The two wedge surfaces in contact with the
conductor are slightly grooved and thus cause a form
fit with the outermost layer of the conductor. Low
friction coating is applied on wedge surfaces which are
in contact with the outer clamp body. By steel straps,
the outer body is held to the insulator which is usually
connected to the tower. In the experiment, the insulator
is supported by the test rig.
Figure 2: Schematic representation of rotary table and test rig. The rotary table is
mounted on the very right side of the 7.5 m span, just next to the wedge clamp.
With this test rig, the conductor is stressed on a span
of approximately 7.5 m between the two wedge tension
clamps as shown in Figure 2. The tensile force on the
conductor can be controlled by motor, screw jack, and a
load cell. In this study, all presented measurements are
performed in an air-conditioned laboratory at 21.5 °C.
The conductor was at room temperature during all
presented experiments. In Switzerland, this conductor
Cigre Science & Engineering • N°14 June 2019
52
is typically loaded up to an effective tensile stress of
50 MPa which results in a tensile force of 27.5 kN. Thus,
a tensile force up to 30 kN is applied in this study.
2.2. Adapted Computer Tomography
2.2.1. Experimental Setup
To measure changes of the inner conductor structure,
x-ray imaging is suitable. If such structure changes occur,
it is assumed that not all wires of a layer behave the same.
Thus, x-ray projections from different angles are needed
to characterize these spatial structural changes. As x-ray
projections of a conductor on an image plate appear
very repetitive, the wires are difficult to distinguish and
it is not possible to re-allocate the same inner wire on
different projections. Therefore, precision steel balls are
used as markers. The position of a marker is changing
when the tensile force is varied. These displacements of
markers are used to analyze the structural changes of the
conductor at different loads. The method is explained in
more detail in the following.
Markers
The density of steel bearing balls provide enough contrast
on the x-ray image to make them clearly distinguishable
from the conductor aluminum alloy. As all the steel balls
have a diameter of precisely 1 mm, the shape and thus
the center of the sphere is clearly detectable and serves
as a well defined marker.
The steel ball diameter of 1 mm is chosen because it
is the smallest but still well manageable diameter. The
balls are glued into drilled holes of a depth and diameter
of 1 mm which enable to install the balls flush with the
wire surface. It is assumed that these holes do not have a
major impact on the structural behavior of the conductor.
Near the clamp, steel balls are installed with a spacing
of 10 mm over a length of 200 mm. From there on, the
spacing is increased to 20 mm. To keep the markers
distinguishable and keep the impact of the holes on
the wire displacement to a minimum, not all wires are
marked but only to a maximum of 3 wires per layer, as
listed in Table I.
To install the markers on all layers without damaging
the conductor, it was opened for approximately 3 m.
To prevent the conductor from complete de-stranding,
it was carefully kept in shape with temporary clamps.
After drilling the holes of 1 mm diameter and depth,
the precision steel balls were bonded into the holes
using epoxy adhesive. When all steel ball markers
were installed, the conductor was manually re-stranded
layer by layer. While re-stranding, the pitch angle of
every layer was measured to guarantee re-assembling
the structure to its original configuration. This kind of
manipulation enabled marking the conductor with a
minimal influence and no damage on the conductor
structure. The manual re-stranding process had a
minimal impact on the conductor structure. Thus, it is
assumed that the conductor behaviour is not influenced
by this process.
Figure 3: View from the top at φ = 0° on the conductor. The three marked wires
from the outermost layer are colored. Blue: Wire 1, Green: Wire 2, Red: Wire 3
Table I: Amount of steel ball marked wires per layer
Layer Marked wires Wires per layer
4 3 24
3 3 18
2 2 12
1 2 6
Core 1 1
Total 11 61
On the outermost layer, the wires to be marked are
chosen specifically. The first, wire 1 is the strand which
is clamped over the longest possible distance by the
upper wedge. The second, wire 2 is chosen to be marked
because it is in contact with the upper wedge until the
its symmetrical middle at φ = 0°. The third, wire 3 is
the wire which is just not clamped by the upper wedge
at z = 0 mm. Thus, wire 3 is the wire transitioning first.
These three marked wires are illustrated in Figure 3 and
Figure 5.
Perspectives
One x-ray image taken from one angle represent the
projection of a three-dimensional structure on a two-
dimensional image. This projection provides insufficient
Cigre Science & Engineering • N°14 June 2019
53
information to identify on which layer a certain marker
is located. Therefore, ten x-ray projections were
taken within an angle of approximately 140°. With
the information of angle and distances between x-ray
source, object, and image plate, the cross-section of the
conductor can be reconstructed as in the conventional
Computed Tomography (CT) method. Due to the well
defined markers and their center point, a complete cross-
section reconstruction is not required. Only the marker
center location reconstruction is necessary. Therefore,
this method is named Adapted Computed Tomography
(ACT).
Rotary Table with X-ray Source and Image Plate
The pulsed x-ray source1 and the image plate support
are mounted on a rotary table as shown in Figure 4. This
rotary table allows to keep constant distances between
source and conductor as well as between conductor
and image plate. Both distances, combined with the
projection angle are essential to re-construct the marker
locations. The rotary table is mounted on a support which
allows stepless variable angles of perspective around
the conductor as close to the wedge tension clamp as
possible, as represented in Figure 2.
The achievable resolution of the image plate scanner2
with a suitable image plate is a pixel size of 50 µm.
Figure 4: Schematic representation of x-ray source on rotary table with image
plate and centering device. All devices are on fixed locations to allow the
reconstruction of the marker locations.
2.2.2. Experimental Procedure and Evaluation
The procedure to create one batch of experimental data
is the following:
• Ten x-ray projections are evenly spaced within a total
angle of approximately 140°. After every projection,
_____________________
1
Golden Engineering, XRS-3
2
Duerr NDT, CR-35
 Figure 5: View from the free-span onto the FEA model at the initial position of analysis
step 1. The conductor is cut at z = 0 mm to better illustrate the three marked wires on the
outermost layer. Due to the radii of the wedge, the red wire 3 is not clamped at the front end
of the upper wedge.
the image plate is scanned, deleted and re-mounted in
its support on the rotary table, which in the meantime
is rotated by one angle increment. All ten projections
are performed at a constant tensile force.
• Semi-automatic image processing allows to detect
the steel ball markers and to determine the center of
each marker on each image. The processed images are
cut into sections along the longitudinal z-axis of the
conductor. The z-coordinate, the projection angle, and
the distances between x-ray source and image plate
allow to allocate each marker in a spatial cylindrical
coordinate system.
This procedure is repeated at 10 kN, 20 kN, and 30 kN,
to create three batches of data. The position of the marker
center points is changing for each tensile force. This
structural change can be represented as displacement of
each marker as a function of tensile force.
2.3. Digital Image Correlation
A second experimental method was required to verify the
developed ACT method. As strain gauges only measure
elongation but no displacement, the optical measurement
with Digital Image Correlation (DIC) was chosen. With
this method, speckle patterns on the conductor surface
are compared at varied load conditions. Thus, only the
outermost conductor layer can be analyzed. However,
the discretization is very fine and all surface wires can
be observed without any invasive preparation. The DIC
method and its application in this study is explained in
more detail in the following.
Speckle Pattern
The surface of the conductor is very repetitive and
light reflecting. Thus, it is not suitable to be analyzed
by DIC without any preparation. A white matt primer
was sprayed on the conductor. Black speckles were
applied by hand with an airbrush pistol. The resulting
Cigre Science & Engineering • N°14 June 2019
54
pattern is random and unique, which allow automatic
image processing. The speckles and the gaps in between
are both approximately 3 to 5 pixels large. This pattern
configuration enabled the best pixel subset to analyze
the pattern changes.
Camera Setup
To observe the spatial deformation of the conductor in
all three dimensions, a stereo camera setup is required.
Two cameras3 observe the same focus area from two
different angles. By calibrating the exact position of the
cameras, the spatial displacement and deformation of the
speckle can be determined.
To enlarge the focus area, several stereo camera systems
were used and the results were merged. To better
observe the variation of displacement on the conductor
circumference, the stereo camera systems are arranged
at the same longitudinal z-position at different φ-angles.
To extend the view along the mechanical transition
zone, the stereo camera systems are positioned next to
each other, at the same φ = 90° but with a spacing of
approximately 55 mm in z-direction.
Experimental Procedure and Evaluation
To make the measurement data comparable, the procedure
for the DIC measurements is similar to the one with the
ACT method. The same conductor and clamp specimen
are used for both cases. First, the speckle pattern was
applied on the conductor. Second, the cameras were
installed and their position was calibrated. The first
measurement is taken at a tensile force of F = 10 kN.
The tensile force was increased to 20 kN and 30 kN.
At both tensile loads another set of data was recorded.
Then, the camera position was changed from lateral to
circumferential configuration to take another three data
sets at 10 kN, 20 kN, and 30 kN.
____________________
3
Prosilica, GT3400
The recorded images of the stereo camera systems are
first processed by the provided software VIC-3D. The
data of the two stereo camera systems is assembled via
MATLAB.
2.4. Finite Element Analysis Model
The experimentally acquired data can only provide
information for certain cases and specific wires. To better
understand the mechanical behavior and to predict non-
measurable cases, a FEA model is used for this study.
Figure 6: Schematic representation FEA model of conductor with both rigid
body wedges. UW: Upper wedge, LW: Lower wedge, CCE: Clamped conductor
end, FCE: Free-span conductor end
The model is based on the previously developed model
from [1, 2]. This model is validated by single wire and
entire conductor experiments.
For the investigation done in [1, 2], a model length
of 30 mm was sufficient to describe the mechanical
behavior in the free-span. In the study presented in
this paper, the mechanical transition zone between the
wedge tension clamp and the free-span is investigated.
Therefore, a much longer FEA model is required.
The complete length of clamped conductor
lclamp = 275 mm plus additional conductor of 700 mm
including the transition zone are required. This results
in a model length of 1000 mm. This model is no longer
solvable with an implicit solver as done previously in
[1, 2]. Thus, an explicit quasi-static algorithm is chosen.
The explicit method was verified by comparing the
explicit and implicit approach on a load case solved by
both methods. As for the ACT and DIC experiments,
the axial displacements uz serve as comparison to the
explicit solver approach. The difference in displacement
uz between the explicit and implicit approach is in a
range of 3 %. This deviation is acceptable because the
implicit solver approach was confirmed in [1] to have
force-elongation characteristics in a range of less than
5 % difference compared to analytic models as well as
experimental measurements.
As shown in Figure 5, all 61 wires are discretized
Cigre Science & Engineering • N°14 June 2019
55
equally mainly by prism elements and by brick elements
in the wire center. Both kinds are extruded with an equal
length of 3 mm along the wire.
The upper wedge (UW) and lower wedge (LW) of
the clamp are modeled as discretized rigid surfaces.
The shape of the surfaces is designed according to the
blueprints of the manufacturer.
The FEA is conducted with ABAQUS [21]. The analysis
is divided in four consecutive steps:
Step 1: Both wedges are pressed to the conductor to
establish contact between both wedges and
the conductor.
Step 2: Tensile force F = 10 kN is applied.
Step 3: Tensile force is increased to F = 20 kN.
Step 4: Tensile force is increased to F = 30 kN.
Boundary Conditions, Restrictions, and Loads
The wedge movement on the α = 5° slope is restricted by
linear equations. To simplify the computation, the upper
wedge and lower wedge are symmetrically displacing on
a slope towards the conductor.
The grooved wedge surfaces cause a form fit between
wedges and conductor. Therefore, it is assumed that
established contacts between wedge surfaces and wire
surfaces cannot be released anymore.
The clamped conductor end (CCE) on the non-tensile
load side protrudes the wedges by approximately 25 mm
when clamped (see Figure 6). To prevent free-body
motion, all CCE nodes are constrained in r, φ, z-direction
during Step 1 of the analysis. From Step 2 on, all CCE
nodes are released in the longitudinal z-direction:
The displacement of all nodes on the free-span conductor
end (FCE) are restricted in circumferential φ-direction.
As at the CCE, this boundary condition is necessary to
prevent the conductor from untwisting.
The tensile force F is applied on a node at the FCE. To
distribute the load, a linear equation restricts all nodes at
FCE to displace equally in z-direction. The FCE nodes
can displace freely in radial r-direction.
To allow relative movements, the contacts between
wires can be open and close. A friction coefficient of
µ = 0.2 is assumed but varied to µ = 0.5 and µ = 0.05
for investigating its influence. A friction coefficient of
µ = 0.2 is rather low compared to literature for bending
load cases [18, 22], but it was representative for an axial
load case with the same conductor in [1].
3. Results
With the ACT method, only displacements can be
measured. Therefore, the displacement output from DIC
and FEA are plotted for comparison.
To make the results of the two measurement methods
ACT and DIC comparable with the FEA model, the
outcome of all three approaches are plotted using the
cylindrical coordinate system shown in Figure 1. To
transform all results into the same reference frame, the
first marker of the ACT on wire 1 of layer 4, which is
visible on all projections, serves as reference point. It
is located at zref = 55 mm. This marker is assumed to
displace the least, as it is the closest marker to the clamp,
which is on the wire that is clamped over the longest
possible distance. This marker position at zref , wire 1
of layer 4 can be identified in DIC and FEA results.
For all three methods, the specific axial displacement
uzref.point |10kN of the reference point at 10 kN of the
corresponding method is subtracted from all axial
displacements uz.
Thus, uz,ref is the relative axial displacement in reference to
the axial displacement uzref.point |10kN of the reference point
at 10 kN. To avoid confusion with relative displacements
between wires on the same or different layers, uz,ref is
referred only as displacement in the following.
Cigre Science & Engineering • N°14 June 2019
56
3.1. Outermost Layer
The optical measurements with DIC could only be
applied on the outermost layer 4. Thus, the two measuring
methods DIC and ACT are first compared to the FEA
model on the outermost layer. In Figure 7, the resulting
displacements of the three marked wires are shown in
separate graphs 7a to 7c.
Due to the position of the stereo camera systems, only
wire 3 can be observed over a distance of 200 mm. The
other two wires are out of sight at a closer point to the
clamp. The DIC results are evaluated for pixel subsets
near the markers of the ACT method.
In Figure 7 it can be seen that at most positions, the
measured displacements of the two experimental
methods are similar. But the results of the FEA can be
off-set at certain locations. Specifically in sub-figure
7a and 7b, the displacement value between z = 150 mm
and z = 200 mm is off-set by a factor of approximately
3. Possible reasons for this substantial difference are
discussed in section 4.3. However, the slope and thus the
stiffness is similar for the FEA model and the ACT result
in this sector.
3.2. All Layers
The displacements from the FEA model are plotted for
each layer in the corresponding part of Figure 8. The
solid lines are for the load case of 30 kN. The dashed
lines represent the displacements at a tensile load of
20 kN.
To ease the readability, ACT results are only shown for
the 30 kN load case in Figure 9. The comparison of the
results between ACT and FEA are the same as for the
outermost layer 4 but for all marked wires.
Based on the FEA results, it can be stated that there is
a transition from the conductor being clamped up to
the point in the free-span, where the influence of the
wedge tension clamp on the conductor is negligible.
This transition differs for each layer and additionally for
different wires on the same layer. The layer dependent
transition can be further distinguished into the varied
transition length and slip behavior.
The slip behavior of the inner layers can be observed by
two phenomena. Firstly, the displacement at z = 0 mm
and right after is larger for the inner layers compared to
the outer layers. Additionally, the displacement curvature
of the core and first layer is negative. Secondly, the
displacement is increasing for the inner layers within the
clamped zone for z < 0 mm.
Figure 7: Displacement uz,ref at 30 kN of ACT, DIC, and FEA of wire 1, 2,
and 3 on layer 4.
In Figure 10, the FEA results with friction coefficients
µ = 0.05, µ = 0.2, and µ = 0.5 are presented. The results with
µ = 0.2 serve as a reference and are identical with the results
shown in Figure 8 and Figure 9. For better readability, the
results with friction coefficients µ = 0.05 and µ = 0.5 are
shifted by uz,ref − 0.1 mm and uz,ref + 0.1 mm, respectively.
Cigre Science & Engineering • N°14 June 2019
57
The results show that the described slip behavior
increases with smaller friction coefficient. The increased
displacement at z = 0 mm and right after as well as the
increasing displacement within the clamped zone for
z < 0 mm, are both more pronounced on the inner layers
for smaller friction coefficients.
The variation of transitions for wires on the same
layer is depending on the circumferential position φ at
z = 0 mm for each wire. Due to the cylindrical asymmetric
clamping, some wires are not in contact with the clamp
wedge at z = 0 mm. Thus, the transition zone initiates at
z < 0 mm for those wires.
This variation on the same layer is more significant for
wires on the outer layers, as these layers are restricted
by no or less contact surfaces than the inner layers. This
characteristic can be well observed in Figure 8a. The
displacement uz of wire 3 is increasing at z < 0 mm. In
contrast, uz of wire 1 and 2 are increasing at the ultimate
contact between the conductor and the upper wedge at
z = 10 mm.
A side effect of a clamp with two wedges is the squeezing
of the conductor. Due to the geometry of the chosen
wedge tension clamp, the conductor is compressed
in vertical direction. Thus, the conductor bulges in
horizontal direction. This created free space can be
occupied by wires on the penultimate layer 3 as it can be
seen for wire 2 in Figure 8b. Coincidentally, this wire is
not clamped up to z = 0 mm due to the bulge created by
the asymmetric clamping.
The described phenomena are the same but scaled for
the 20 kN and 30 kN load case. However, the length of
the transition zone is not influenced by the tensile force
as it can be seen in Figure 8.
Compared to the FEA model results, the variation of
displacements along the conductor is more significant
for the measured values of the ACT method as it can be
seen in Figure 9. Firstly, the amplitude of the variation
is larger. Secondly, the shape of the variation is shifted
towards the free-span for the ACT results. Even though
the location and the amplitude of the variation is different
between ACT and FEA results, they are qualitatively
similar in both methods.
4. Discussion
4.1. Transition Zone
The mechanical transition zone between the wedge
tension clamp and the position where the influence on
the conductor is negligible depends on several factors,
as discussed in the following.
4.1.1. Slipping of Inner Layers
Due to layer diameter and pitch angle, the length of
wire per unit length of the conductor is shorter for the
inner layers. Thus, the longitudinal strains and stresses
are higher for the inner layers at free-span conditions.
At the back-end of the tension clamp, the tensile force is
negligible. Thus, besides friction forces, there is nothing
holding the inner layers in longitudinal direction. Due to
the lack of form fit between the layers, slip behavior of
the inner layers occurs.
The entire tensile force is transferred from the clamp
to the conductor by friction. The wedge surfaces in
contact with the conductor are slightly grooved in the
chosen clamp. These grooves provide a form fit with the
outermost layers. Thus, slipping is not occurring at this
location. This is not the case for other kinds of clamps,
where these surfaces are missing a rough structure.
The friction coefficient µ was changed for the FEA
model to analyze its influence. The inner layers are
slipping less with higher friction coefficient. However,
in the Figures 7 and 9 it can be seen that the slope of
measured displacements on the outer layers is steeper
than the one of the FEA. This larger displacement
gradient can be caused by the enhanced loading of
the outer layers, due to increased slipping of the inner
layers. Hence, the contact between wires would need to
be modeled with a friction coefficient µ < 0.2. However,
for the investigated friction coefficients, the highest
loading of the outermost layer was achieved for µ = 0.2.
Compared to bending load cases in literature [13, 22],
a friction coefficient of µ = 0.2 is rather small, but it
was representative for an axial load case with the same
conductor in [1, 2]. Nevertheless, for a realistic range of
values of µ, slippage occurs and generates a transition
zone comparable to the results shown in Figures 8, 9, Figure 8: Displacement uz,ref at 30 kN (solid line) and 20 kN (dashed line) of
and 10. FEA of all marked wires of all layers.

Cigre Science & Engineering • N°14 June 2019

58
Figure 9: Displacement uz,ref at 30 kN of ACT (asterisk) and FEA (solid line) of Figure 10: Displacement uz,ref at 30 kN of FEA for varied friction coefficients. The
all marked wires of all layers. results with µ = 0.2 serve as a reference and are the same as shown previously
in Figures 8 and 9. For better readability, the results with friction coefficients
µ = 0.05 and µ = 0.5 are shifted by uz,ref − 0.1 mm and uz,ref + 0.1 mm, respectively.

4.1.2. Asymmetric Clamping

The chosen wedge tension clamp is compressing the and clamped equally. Thus, not all wires are exposed to
conductor from two sides. This causes the conductor to the same stresses and strains. Asymmetric clamping is
bulge to the other two sides which are lacking wedges. causing the fluctuating displacements within the clamped
Therefore, not all wires on the same layer are compressed zone. This can be seen in the FEA results in Figures 8.

Cigre Science & Engineering • N°14 June 2019

59
4.1.3. Length of Transition Zone
The length of the mechanical transition zone lmech is
the distance from z = 0 mm up to the point where the
influence of the clamp can be neglected. Beyond this
zone, the displacements follow a linear trend.
To evaluate the transition length based on the FEA results
with a friction coefficient of µ = 0.2, the displacements
uz,ref between z = 400 mm and 600 mm are fitted with the
linear function
with the constants C1 and C2 .
Beyond the position where the difference between the
fitting curve uzFIT (z) and simulated displacement
is 5 %, the influence of the clamp can be neglected and
thus lmech is reached.
(11)
The resulting lmech of each wire is shown in Table II as
a range for each layer for both load cases of F = 20 kN
and 30 kN with a friction coefficient of µ = 0.2. The
transition length varies significantly between different
layers and between different wires of the same layer.
By comparing the two load cases, lmech of the first and
second layer does not change as much as of the other
layers. In layer two, no variation between load cases
and between wires in the same layer were detected. For
the layer four, the transition length became shorter at
higher tensile force, whereas the transition length for the
core became longer. For the layers three and one, the
variation between wire on the same layer is smaller at
higher tensile force. Independent of the tensile force, the
transition length is less than 300 mm for the FEA results.
Results from ACT are only available up to approximately
z = 300 mm. Beyond z = 200 mm, all ACT results
approach the same linear trend. Towards z = 300 mm,
most of these displacements reach uz,ref ≈ 0.3 mm. No
clear statement for the length of the transition zone can
be made from the presented ACT results.
Cigre Science & Engineering • N°14 June 2019
60
Table II: Range of mechanical transition length lmech per layer for the 30 kN and
20 kN load case with a friction coefficient of µ = 0.2, calculated according to
equation (10) and (11).
Layer lmech lmech
F = 30 kN F = 20 kN
mm mm
4 157-271 262-280
3 178-286 82-295
2 172 172
1 196-202 187-211
Core 265 208
4.1.4. Tensile Force Variation
Variation of tensile force is changing the resulting axial
displacements as well as the compressing forces. Hence,
variation between wires depends on different tensile
forces, too. But the length of the transition zone is not
affected at the two investigated cases.
4.2. Consequences for Creep
Due to slipping and asymmetric clamping, the wires on
the outermost layer clamped up to the last point of full
contact with the wedges at z = 0 are exposed to higher
tensile stresses than other wires. Initially, these wires are
expected to creep more. With time, other wires are going
to be exposed to higher tensile stresses as well, since
the wires with initially high stresses are deformed more
due to creep. Thus, the wires with initially low tensile
stress are stressed more after time. The tensile loading
becomes more homogeneous with time. However, the
initially high loaded wires experienced more creep and
are thus more damaged. In certain circumstances, this
staged creeping phenomena could lead to a localization
effect with creep failure.
More detailed characterization of the creep behavior in
the transition zone can be achieved by implementing
material creep equations into the presented finite element
model and by considering the temperature profiles in the
thermal transition zone [7].
4.3. Comparing FEA Model to Measurements
In Figure 7, the results of the two measurement methods
are in accordance for most locations along the conductor.
Compared to ACT, the DIC results represent a slightly
stiffer conductor. This can be caused by the sequence
of the experiments. On the same conductor and clamp
specimens, the DIC method was applied after the ACT
method. Thus, the load cycles of the ACT measurements
could have caused plastic deformations on certain
locations that led to a stiffer conductor behavior. Even
though the sequential testing with ACT and DIC method
might have had an influence on the conductor behavior,
the benefit to enable a direct comparison between the
two methods was assessed with higher importance. With
different specimen, not all wires would be placed in the
same position. Thus, all contact points and the position of
the wires on the inner layers would be different between
the two experimental methods.
The difference between measured and simulated results
can differ significantly within one order of magnitude,
as shown in Figure 9. This can be explained by the
simplification in the FEA model and the accuracy level
of the measurement method.
In the FEA model, contacts between wedge surfaces and
wire surfaces are locked as soon as established. Wire
nodes that are in contact with any of the two wedges can
not move relative to the wedge in any direction. This
behavior is more restrictive than the real case, where at
all ends of the wedges, wires might be slightly sliding
on the wedge surfaces. This can be seen in Figure 11,
where the deformation of the FEA model is exaggerated
by a factor of 15 for the 30 kN load case. In reality, the
response is less stiff than assumed by locked contacts
in the FEA model. This allows the wires to displace
more than in the model. Furthermore, the conductor is
in contact with the upper wedge up to z = 10 mm and
with the lower wedge up to z = 0 mm. In reality, the
extended 10 mm contact with the upper wedge does not
significantly influence the conductor, as the counteracting
force from the lower wedge is missing. The FEA model
lacks this counteracting force as well. But since the
modeled contact between conductor and wedge is
locked, the upper wedge is influencing the outermost
wires for additional 10 mm. The locked wedge-wire
contacts and the unequal wedge lengths lead to a more
restrictive and thus stiffer behavior of individual wires
in the FEA model.
The chosen x-ray imaging method with markers enables
to measure displacements of the inner conductor
structure. The chosen setup consists of many individual
parts. The guidance of the rotary table relative to the test
rig was managed as well as possible. But an industrially
proven CT-scanner with better device guidance would
improve the accuracy of the results more than a higher
pixel density of the image plate and scanner combination.
Figure 11: Illustration of outermost layer with a deformation scale factor 15.
To better distinguish the wires, the colors are alternated between light grey
and dark grey. The marked wires follow the previous color code. Blue: Wire 1,
Green: Wire 2, Red: Wire 3
5. Conclusion
The shown methods enable to achieve qualitative and
quantitative results of wire displacements of an OHL
conductor near a wedge tension clamp. The transition
zone and occurring phenomena can be described and
explained consistently with all three approaches. The
results lead to a better understanding of the mechanical
behavior of OHL conductors near tension clamps in
general.
The developed FEA model enables to simulate
long clamped conductors. It was compared to two
displacement measuring techniques DIC and ACT
specifically adapted to this application. The developed
ACT method with x-rays allows to measure wire
displacement of all conductor layers.
All presented tools can be used for further in depth
analysis and can be applied to other cases.
The key results of the investigation in this paper are
summarized in the following list:

Cigre Science & Engineering • N°14 June 2019

61
• The mechanical transition zone between the wedge
tension clamp and the position in the free-span
where the influence of the clamp can be neglected is
approximately 300 mm long in the investigated FEA
cases.
• Due to the lack of form fit, the entire tensile force
is transferred from the tension clamp to the inner
conductor structure by friction. This results in
slipping of the inner conductor layers towards the
free-span.
• The studied wedge tension clamp is compressing the
conductor from two sides, which results in unequally
clamped wires on the same layer. First, not all
wires of the outermost layer are clamped up to the
end of the wedges. Hence, wires on the outermost
layer start to transition before the end of the clamp.
Second, compressing the conductor from two sides
results in a bulge. Hence, the free space created
by the bulge leads to unequally clamped wires on
the inner layers as well. Thus, we recommend to
compress the conductor on all its circumference with
an homogeneously distributed force.
• Measuring wire displacement of inner conductor
layers can be achieved with the ACT method. With
further improvements of the device guidance system,
this method can be applied for other investigations
to enhance the understanding of changes within
conductors and similar complex structures.
• The explicitly solved 1 m long FEA model enabled
to simulate the conductor behavior in the mechanical
transition zone. The model represents a too restrictive
wire displacement and the resulting outcome values
differ from the measurements. However, all measured
phenomena could be qualitatively reproduced by the
FEA model.
Acknowledgment
The authors thank Marco Frigerio, Konstantin Papailiou,
Stuart Holdsworth, Kilian Schillai and Martin Baltzinger
for their contributions to this study. Financial support
of the Competence Center for Energy and Mobility
(CCEM) is acknowledged.
References
[1] M. Frigerio, P. B. Buehlmann, J. Buchheim, S. R. Holdsworth, S.
Dinser, C. M. Franck, K. Papailiou, E. Mazza, Analysis of the
tensile response of a stranded conductor using a 3D finite element
Cigre Science & Engineering • N°14 June 2019
62
model, International Journal of Mechanical Sciences 106 (2016)
176–183.
[2] M. Frigerio, Quantifying the Effect of Time and Temperature on
the Mechanical Behaviour of Aldrey Overhead Line Conductors,
Phd thesis, ETH Zurich (2016).
[3] J. Harvey, Creep of transmission line conductors, IEEE
Transactions on Power Apparatus and Systems PAS-88 (4) (1969)
281–286.
[4] J. Harvey, R. E. Larson, Use of Elevated-Temperature Creep
Data in SagTension Calculations, IEEE Transactions on Power
Apparatus and Systems PAS-89 (3) (1970) 380–386.
[5] Cigré WG B2.43, Technical Brochure 601: Guide for thermal
rating calculations of overhead lines, Cigré, 2014.
[6] P. B. Buehlmann, M. Meier, M. Vifian, E. Mazza, C. M. Franck,
Temperature Profile along an Overhead Line Conductor in and
near the Tension Clamp, Cigré Session 46 (B2-305).
[7] P. B. Buehlmann, C. M. Franck, Temperature Profiles of All-
AluminumAlloy Conductors near Wedge Tension Clamps, Cigré
Science & Engineering 10 (February) (2018) 66–76.
[8] Cigré WG B2.30, Technical Brochure 429: Engineering Guidelines
Relating to Fatigue Endurance Capability of Conductor/Clamp
Systems, Cigré, 2010.
[9] F. Lévesque, S. Goudreau, L. Cloutier, A. Cardou, Finite element
model of the contact between a vibrating conductor and a
suspension clamp, Tribology International 44 (9) (2011) 1014–
1023.
[10] A. Cardou, A. Leblond, L. Cloutier, Suspension clamp and
ACSR electrical conductor contact conditions, Journal of Energy
Engineering 119 (1) (1993) 19–31.
[11] F. Levesque, S. Goudreau, A. Cardou, L. Cloutier, Strain
Measurements on ACSR Conductors During Fatigue Tests
I—Experimental Method and Data, Power Delivery, IEEE
Transactions on 25 (4) (2010) 2825–2834.
[12] Y. D. Kubelwa, R. C. Loubser, P. Moodley, Experimental
investigations of bending stresses of ACSR conductors due to
aeolian vibrations, Cigré Science & Engineering 9 (October)
(2017) 17–26.
[13] S. Lalonde, R. Guilbault, S. S. Langlois, Modeling multilayered
wire strands, a strategy based on 3D finite element beam-to-
beam contacts Part II: Application to overhead conductors wind
induced vibrations, International Journal of Mechanical Sciences
126 (2017) 297–307.
[14] Cigré Green Book: Overhead Lines, Study Commitee B2, 2014.
[15] N. Pouliot, G. Rousseau, A. Leblond, S. Montambault, Portable
X-ray system for in situ detection of broken ACSR strands at
suspension clamps: field results and equipping the Line Scout
robot, Cigré Session 46 (B2-206).
[16] L. Xiang, H. Y. Wang, Y. Chen, Y. J. Guan, L. H. Dai, Elastic-
plastic modeling of metallic strands and wire ropes under axial
tension and torsion loads, International Journal of Solids and
Structures 129 (2017) 103–118.
[17] S. R. Ghoreishi, T. Messager, P. Cartraud, P. Davies, Validity and
limitations of linear analytical models for steel wire strands under
axial loading, using a 3D FE model, International Journal of
Mechanical Sciences 49 (2007) 1251–1261.
[18] S. Lalonde, R. Guilbault, F. Légeron, Modeling multilayered wire
 strands, a strategy based on 3D finite element beam-to-beam
contacts Part I: Model formulation and validation, International
Journal of Mechanical Sciences 126 (2017) 281–296.
[19] I. Argatov, Response of a wire rope strand to axial and torsional
loads: Asymptotic modeling of the effect of interwire contact
deformations, International Journal of Solids and Structures 48
(2011) 1413–1423.
[20] R. Baumann, P. Novak, Efficient computation and experimental
validation of ACSR overhead line conductors under tension and
bending, Cigré Science & Engineering 9 (October) (2017) 5–16.
[21] Dassault Systemes, Simulia, Abaqus 2016.
[22] K. O. Papailiou, Die Seilbiegung mit einer durch die innere
Reibung, die Zugkraft und die Seilkrümmung veränderlichen
Biegesteifigkeit, PhD Thesis, ETH Zurich (1995).

Cigre Science & Engineering • N°14 June 2019

63
Seismic response of
high-voltage cables in ducts –
A preliminary study

S. CHERUKUPALLI*, Principal Engineer, Manager Transmission Engineering

S. LIMA, Senior Engineer, Transmission Engineering,
B.C. HYDRO, Vancouver, British Columbia Canada

Abstract This study is mainly concerned with numerical modelling

The goal of this study is to develop a reliable method
to analyze the dynamic response of high-voltage
transmission cables installed in horizontal directional
drilled (HDD) pipes under seismic events and assess the
tensile forces that a cable joint may experience at the ends.
The seismic ground shaking and resulting movement of
cables within the ducts and manholes is modelled with
site-specific earthquake time histories compiled by BC
Hydro dam safety department. The maximum slope
angle for a cable crossing was previously assumed to be
between 10 and 13 degrees using a simplified numerical
model. Based on the improvements made to the seismic
response models, it was shown that the approach angles
at each end can be theoretically increased by up to 5
degrees. For an 850 m crossing resulting in shortening
of the approach anagles and considerable savings on
material and construction activities. This study is based
on a recently installed BC Hydro 230 kV XLPE cable
systems with corrugated aluminum sheath.
1. Introduction
With the complexities of incorporating transmission
cable corridors in busy streets and crossing main arterial
roads and highways quite often these portions of the
underground transmission cables in ducts have to be
installed in a tunnel or horizontal directional drilled (HDD)
carrier pipe. The gaps between the carrier pipe and ducts
are generally filled with fluidized thermal backfill (FTB)
or similar material to improve the thermal performance of
the installed cable systems.
of seismic ground shaking and resulting forces and
displacements of cables that can occur within the ducts
and manholes. It is assumed that the carrier pipe and
ducts move as one unit and there is no slippage relative
to the surrounding ground motion due to the grouting
operation described above. Hence, large permanent
ground deformations (PGD) and the effects of differential
settlements have not been considered in this study and are
more appropriately assessed by a geotechnical engineer.
At locations where HDD pipes are required, the cables
typically have to be angled downwards to go into a
deep underground crossing. A sharper entry angle will
reduce the construction costs and the overall footprint.
However, this may have a serious adverse effect on the
performance of cable and adjacent joints under thermo-
mechanical or seismic loading. The optimum angle of the
slope has to be determined by investigating the forces and
displacements induced in the cables and joints for these
various alignments.
To fully assess the approach angles under seismic events
specialized software used by the offshore industry for
dynamic response analysis of subsea power cables is used.
This software is typically used to model cable interactions
with seabed and subsea structures, as well as pipe-in-
pipe systems. Verification of this software for seismic
analysis of high-voltage cables in ducts was carried out by
comparing to a model which was validated by examining
the stresses that caused pipe-type cables to fail during
Kobe Earthquake of January 1995 in Japan, Ref. /1/.

* sudhakar.Cherukupalli@bchydro.com

KEYWORDS
HDD, Cables in Ducts, Horizontal Directional Drilling, Seismic Performance of Cables, Dynamic Analysis

Cigre Science & Engineering • N°14 June 2019

64
This paper will discuss the following items;
• Development of a single duct seismic model for HDD
crossing with cable slack at each end.
• Slope angle assessment by altering the slope angles up
to 18°.
• Development of a local manhole model considering
an actual cable configuration inside a manhole,
including cable joints, guide posts and duct entry into
the concrete manhole.
• Thermo-mechanical analysis using the manhole and
duct model and applying cyclic thermal expansion and
contraction loading to investigate cable behavior and
anchoring arrangement inside the manhole.
• Dynamic analysis using global seismic x, y, and z
forced displacements at duct sections.
• Determine cable residual loads under no circuit
loading conditions and perform pull-in analysis to
estimate cable tensions and side wall pressures during
installation.
Figure 1a shows a plan and profile of a typical HDD under
a body of water. High Density Polyethylene (HDPE) pipes
are pulled into the drill hole and grouted. Subsequently,
transmission cables are pulled into the empty ducts. Due to
the considerable expense of such a construction additional
HDPE pipes are typically included in the bundle to allow
for additional cable circuits in the future.
2. Assumptions made in the Study
The following assumptions were made in this seismic
analysis study:
• Only cyclic seismic loadings are considered in
assessment of mechanical integrity of the cables.
Hence, it is assumed that large permanent ground
deformations do not occur during the seismic event.
Table 1: Cable mechanical properties
Parameter
OD
Weight per unit length in air
Weight per unit length in water (assumed operating condition)
Axial stiffness
Cigre Science & Engineering • N°14 June 2019
65
• The dynamic response of the carrier pipes and internal
ducts follow the local seismic ground motion without
any attenuation.
• The cables are placed inside the ducts and are allowed
to slide axially when the inertia loading overcomes the
frictional forces.
• It is assumed that the cables are free to move axially
until they reach the pre-determined slack values of 100
mm or 1.0m at the manhole interfaces.
• All load cases were assessed for cable-to-duct friction
coefficients between 0.18 and 0.23 for comparative
purposes, assuming cables immersed in water as per
previous study, Ref. /1/.
• The nominal time step used in the dynamic analyses is
0.01s.
• Unless noted otherwise, the arrival time of the seismic
wave along the cable is based on a wave propagation
velocity of 1,500 m/s. A sensitivity analysis is included
to show seismic response under a wave propagation
velocity of 500 m/s.
• Mechanical properties of a recently installed BC Hydro
230kV 1,600 mm2 copper conductor cable with XLPE
insulation and an extruded corrugated aluminum
sheath were assigned to the cable model, see Table 1.
Three axial cable stiffness values were investigated to
take into account uncertainties in values provided by
the manufacturer and variability in ambient conditions.
The bending and torsional stiffness values were
estimated based on data from similarly sized cables,
Ref. /5/. However, since both ends are free to move due
to cable slack, these properties are assumed to have less
impact on the results.
Figure 2 shows the profile of a typical HDD crossing that
was used in Vancouver.
The global model assumes that the dynamic response of
Value Units
141 mm
27 kg/m
11.4 kg/m
40, 80 & 120 MN
 (a)

(b)
Figure 1a) – Plan and profile of a typical HDD under a body of water b) Cross-section of this horizontal direction drill

the cable inside the duct and manholes at each end are de- with the guide posts and entry/exit ducts is modelled by
coupled until the pre-determined slack of 100 mm or 1.0 m assigning contact elements. The straight joint and cable
is reached. Once the effect of the cable slack configuration duct-to-manhole interface are modelled so that interface
is investigated, this can be inserted as a feed-in to the local loads can be calculated.
manhole model as shown in Figure 3. The cable interaction

Fig. 2: Assumed geometric profile at the North and South ends of the HDD crossing

Cigre Science & Engineering • N°14 June 2019

66
 Fig. 3: Cable configuration in a 12 m x 3 m concrete manhole

A. Seismic Data
It is estimated that the velocity of the seismic wave
propagation for compact to dense sand/silt stratum and
weak sandstone bedrock is 1,500 m/s. The arrival times
for the seismic wave along the cable are defined in Table 2
and compared to a wave propagation velocity of 500 m/s.
Table 2: Arrival time of seismic wave along cable for different wave
propagation

Time Delay [s]

Distance from
South end of cable [m] 500 m/s wave 1,500 m/s wave
propagation propagation

0 0 0 Figure 4: Seismic time history used in dynamic analysis

120 0.24 0.08 Typically, cables can experience high tensions during
installation, which reduces with thermo-mechanical
240 0.48 0.16 relaxation forces during cyclic loading within the duct.
360 0.72 0.24 Therefore, as part of the dynamic analysis, the expected
initial conditions of the cable have to be determined. This
480 0.96 0.32 can be done by pulling the cable through the ducts and,
subsequently, artificially expanding and contracting the
600 1.20 0.40
cable to simulate the relaxation in the cables once put in
720 1.44 0.48 operation. Previous studies did not investigate different
initial residual tensions of the cable system prior to arrival
For this study five earthquake time histories were evaluated. of seismic wave. Thermo-mechanical loading was also not
The time histories were selected based on the obtained considered in the original study.
Uniform Hazard Spectra (UHS) curve for 10% damping.
B. Cable Modeling Methodology
These were scaled to match the spectral acceleration in
the range of estimated fundamental periods of the cable The 3D model in this study consists of 47 contact
model of 0.15 to 0.2 s. An example of one of the applied elements to simulate the seismic motions of the duct
earthquake time histories is shown in Figure 4. system. Each duct section has an imposed displacement

Cigre Science & Engineering • N°14 June 2019

67
 Figure 5: Global cable and duct model

that takes into account the arrival time of the seismic
wave. The cable is placed inside the duct sections and
consists of 1720 line elements. Springs are attached at
each end to model the cable slack and interaction with
the manholes, see Figure 5.
The cable is modelled as line elements as shown in Figure
6. Each line consists of a straight massless segment with
axial and torsional properties. The mass, forces and
moments are placed at the nodes.
The axial stiffness and damping of the line are modelled
by the axial spring-damper at the center of each segment,
which applies an equal and opposite effective tension
force to the nodes at each end of the segment. The bending
properties are represented by rotational spring-dampers on
either side of the node, spanning between the node’s axial
direction Nz and the segment’s axial direction Sz.
C. Dynamic Analysis
The dynamic response of the cables is determined by
solving the equation of motion in the time domain
according to the following expression:

Figure 6: Cable line model

Cigre Science & Engineering • N°14 June 2019

68
M(x, a) + K(x) + C(x, v) = F(x, v, t)
where
M is the inertia load physical high frequencies. For cases where snatch loads on
K is the stiffness load
C is the damping load
F is the excitation loading
x, v and a are the position, velocity and acceleration
vectors, respectively.
t is the simulation time
The excitation loading, F, is applied as a forced
displacement, x, at the duct sections which represent the
seismic ground motion. For the cables, which are placed
inside ducts, the inertia loading will have to overcome the
frictional forces between cable and duct before it starts to
slide relative to the duct. Otherwise, the cable will see the
same motion as the carrier pipe.
As the cables are assumed to be submerged in water
(introduced into the pipes to ease cable pulling), the
damping effects of the surrounding fluid would be
dominant. Since the software is specifically designed
to analyse the dynamic response of subsea cables, the
hydrodynamic loads are also taken into account when
solving the equation of motion.
All dynamic analyses are based on the implicit generalized
α-integration scheme as defined by Chung and Hulbert,
Ref. /2/. This method includes numerical damping of non-
the cable are anticipated, additional runs should be carried
out using the explicit integration scheme to ensure that all
non-linear effects have been included.
A total of 60 load cases were defined. Each run was
designated a name based on the following parameters:
Cable slack length at manhole (100 or 1000 mm); axial
cable stiffness (40, 80 or 120 MN), coefficient of friction
(0.18 or 0.23), and earthquake time history (five different
time series). The duration of each run is set at 40 seconds
to match the input time histories.
3. Results
An example of the results from a single dynamic analysis
load case is shown in Figure 7 below and identifies the
peak tension and compression values reported along the
cable length. Figure 8 presents the peak values extracted
from each load case with 100 mm slack at the manhole
ends. Comparison is made between the two slope angle
combinations namely 10°/13° and the upper limit of
15°/18°.

Figure 7: Typical maximum tension/compression loads reported along cable length

Cigre Science & Engineering • N°14 June 2019

69
 Figure 8: Dynamic maximum tension/compression for 100mm slack and a 1500m/s seismic wave velocity for different entry and exit angles
(15°/18° or 10°/13°)

In general, Figure 8 shows that the peak tensions reported
along the cable are between -20 kN and 20 kN for the
100 mm slack load cases. Furthermore, it is shown that
there is not a significant difference between the peak
tensions reported for the two slope configurations. On design of the cable system it is therefore important to also
average, the maximum reported tensions along the cable
lengths increase by 5% when the angles increase by 5°.
The difference between the peak compression loads is
typically less than 1%. A similar comparison for load cases
with 1.0 m cable slack at manholes showed similar results,
albeit with peak forces computed along the cable typically
in the -20 kN to 10 kN range.
Additional sensitivity analyses show that the distribution
of the tension is also affected by the direction of seismic
wave propagation. In general, higher tensions are found
at the downstream end of the seismic wave. During detail
include directional effects – either by expanding the load
case matrix or running several sensitivity cases that take
into account the potential approach angles of the seismic
waves
To investigate the sensitivity of the dynamic response of the
cables to seismic wave propagation velocity, the 1,500 m/s
is compared to a lower bound value of 500 m/s. Using the

Figure 9: Comparisons of tensile and compressive forces for the two different seismic propagation velocities

Cigre Science & Engineering • N°14 June 2019

70
 Figure 10: Axial displacements at cable end
model developed for the 18°/15° slope configuration,
the results are summarized in Figure 9.
From this sensitivity check it is shown that the arrival time
(hence the seismic wave velocity) of the seismic wave
has a more significant impact on the peak tension results
compared to increasing the slope angles of 10° and 13°
by 5 degrees.
From Figure 10 it is shown that the cable axial movement
at the cable end follows the ground motion closely. The
graph illustrates how the cable slack of 100 mm is reached
after approximately 4 seconds from arrival of the seismic
wave and, after some initial slippage, follows the general
ground motion.
The original analysis was based on a horizontal axial seismic
excitation only. In order to assess the full dynamic response
of the cable system, vertical and lateral components need
to be added to the time series. By including combined
axial and vertical motions, where the vertical component
represents 80% of the horizontal motion, the peak forces
are found to increase by approximately 25% and 18%
under tension and compression, respectively.
Conclusions and recommendations
By performing numerical modelling of the seismic
response of transmission cables and accessories, the
anticipated tension and compressive forces experienced
in the cable and joint can be better estimated. This may
be used to appropriately tailor and optimize the seismic
testing protocol and qualification of a cable system.
Based on the improvements made to the seismic response
models, it is shown that the approach angles at each end
could be increased by up to 5 degrees. For an 850 m
crossing this would result in a horizontal shortening of the
Cigre Science & Engineering • N°14 June 2019
71
approach slopes and considerable savings on materials and
construction activities and possible improved construction
schedule. Discussion will need to be held with the drilling
Contractor to confirm that such a drill profile can actually
be achieved practically. Simultaneous consultations with
a qualified Geotech engineer have also to be considered
before a final decision can be made.
By comparing the results from the 10°/ 13° and 15°/ 18°
slope angles, it is shown that there is not a significant
difference between the peak tensions in this 45 m deep
crossing. On average, the maximum tensions along the
cable lengths increase by 5% when the angles increase
by 5° under a coefficient of friction range of 0.18 to 0.23.
When the installation is new the expectation is that µ; the
coefficient of friction will be low but with the passage of
time, introduction of sediments into the pipe will tend to
increase the value for µ. The values selected were based on
a separate study conducted to measure static and dynamic
cable friction in pipes with and without water.
From a sensitivity study of the seismic wave propagation
velocity, it is shown that the arrival time of the seismic
wave along the cable length can have a more significant
impact on the peak tension results compared to the 5°
change in slope angle.
By subjecting the cables to thermal expansion and
contraction prior to running the seismic events, it is shown
that the load cases that represent the lower bound values of
friction and axial stiffness are highly sensitive to the initial
residual tension conditions.
The original seismic study only considered axial seismic
ground motions. By including a vertical component
that represents 80% of the horizontal motion, the peak
maximum and minimum forces change by +25% and
-18%, respectively.
Sensitivity analyses show that the distribution of the
tension is also affected by the direction of seismic wave
propagation. In general, higher tensions are found at the
downstream end of the seismic wave.
Future studies are planned to include the dynamic seismic
response of cables under different circuit loading conditions
to address how hot and cold cable systems behave under a
seismic event.
Since the lower bound friction and lower bound axial
stiffness load cases were found to be more critical under
dynamic loading of the cables, further investigation into
appropriate lower bound values is required to ensure
the basic load case matrix covers worst-case operational
conditions.
It is also important to dedicate resources to obtain detailed
information on cable mechanical properties, newer cable
sheath constructions, and site-specific seismic data as
input for the dynamic analyses. To obtain realistic bending
and torsional properties of transmission cables, it may be
required to conduct experiments on actual cables to be
installed.
References
[1] Dr. M. Lee, File No. CBL0904-1 , “Seismic Analysis of Cable in HDD
Duct using Drain2DX, Mount Pleasant-False Creek”, May 14, 2009
[2] Chung, J. and Hulbert, G. M., A time integration algorithm
for structural dynamics with improved numerical dissipation:
Generalized-a method, ASME Journal of Applied Mechanics, 60,
1993
[3] BC Hydro Technical Report, Verification, “Seismic Analysis of
High-Voltage Cables in HDD/Tunnel”, Prepared by S. Lima (BC
Hydro), November, 2017
[4] Powertech Technical Report, “Frictional Properties of HDPE
Conduit and PE Cable”, Prepared by K. Fitzpatrick (Powertech)
for T. Kojima (BC Hydro), May 4, 2009
[5] BC Hydro Technical Report, “Recommended Mechanical Properties
of Transmission Cables and Other Properties Required for Seismic
Performance Analyses”, TD2016-000.500, January 2017
Acknowledgements
The authors thank BC Hydro Management for the interest Lead Engineer in the Subsea department of Wood (former
and ongoing support during a project and in preparation of
this paper. The authors also thank the guidance provided by
Takashi Kojima throughout this study and sharing the data
compiled on the mechanical properties of transmission
cables as well as Dr. Michael Lee for undertaking the early American Society of Civil Engineers and the Norwegian
analysis in the past. The support received from BC Hydro
Cigre Science & Engineering • N°14 June 2019
72
Project Manager Antigone Dixon-Warren and Project
Engineer Robert Smith is gratefully acknowledged. The
authors also thank colleagues Anthony Lam, and Dr. Kofi
Addo for their review and suggestions of this manuscript.
Biographies
Dr. Sudhakar Cherukupalli worked as an R&D Engineer
at Brown Boveri, India, for seven years, after completing
his Master’s degree. There he was responsible for 420kV
switchgear design. In 1987, he completed his Ph.D from
the University of British Columbia and then joined BC
Hydro R&D, which later became Powertech Labs. He
is currently a Principal Engineer and Manager in BC
Hydro’s Transmission Cable Design department. He has
extensive experience in design, installation, and testing
of transmission and distribution cables, and accessories.
He has authored and co-authored over 40 technical
publications. He has contributed to several CIGRE Working
Groups such as the “Impact of EMF on Current Ratings
and Cable Systems”. He has served as the Convener of
CIGRE D1.34 titled “Dissolved Gas-in-Oil Interpretation
for Transmission Cables”. He has been a contributor to
several IEEE Standards on Voltage Endurance Testing and
Thermal cycling testing of large hydro generator bars and
coils, Distributed Temperature Systems. He is a Senior
Member of the IEEE, serves as Chair of Subcommittee
F –Field Testing and Diagnostics of the IEEE-Insulated
Conductors Committee and is a Professional Engineer
in the Province of British Columbia, Canada. He can be
contacted at s.cherukupalli@ieee.org.
Scott Lima graduated from Heriot-Watt University, UK in
1995 with a degree in Offshore Mechanical Engineering.
Working at Schlumberger and Worley Parsons Canada
he was responsible for the design and analysis of a wide
range of fixed and floating marine structures. In 2007 he
joined Subsea 7 in Norway to specialize in installation
and design of subsea flexibles and then held a position as
JP Kenny). He returned to Canada in 2016 and is now
working as a Senior Engineer at BC Hydro.
Scott is a registered Professional Engineer in the Province
of British Columbia, Canada. He is also registered with the
Society of Chartered Engineers.
A novel Wavelet selection scheme for
partial discharge signal detection under
low SNR condition

J. LIU1*, W.H. SIEW1, J.J. SORAGHAN1, X. HU2, X. PENG3, and E. A. MORRIS1

1
Department of Electronics and Electrical Engineering,
University of Strathclyde, Glasgow, G1 1XW, UK
2
Department of Electrical Engineering, Guizhou University, Guiyang, 550025, China
3
School of Electrical and Electronic Engineering, Huazhong University of Science and Technology,
Wuhan, 430074, China

Abstract: signal information both in time and frequency domains.

Wavelet-based techniques have been widely used
to extract partial discharge (PD) signals from noisy
signals. Generally, the procedure consists of 3 steps:
wavelet selection, decomposition scale determination,
and noise estimation. Wavelet selection is the first and
most important step for its successful application in PD
denoising. However, despite many variants of techniques
deployed, the success rate is not generally good especially
when the signal to noise ratio is unity or less. This paper
discusses a novel technique that addresses this issue. The
technique is inspired by the concept of Shannon entropy
and the associated information cost functions (ICF) in
information theory. It is adaptive to the detected PD signals.
The paper demonstrates that the proposed technique is
effective when applied to PD signals obtained through
laboratory experiments and on-site measurements. When
this technique is applied to cable diagnostics, it should
have the potential to extend the range of PD detection
from cables.
1. Introduction
Partial discharge (PD) measurement is an effective
technique for the monitoring of electrical insulation.
However, PD signals are normally contaminated by noise
from the environment, which increases the difficulty of
their detection. To effectively extract PD signal from
noisy signals, various denoising techniques, such as
adaptive filter [1], [2], and the wavelet-based technique
[3]–[7] have been adopted to remove noise. The wavelet-
based technique has been widely used in recent years
since wavelet transform can simultaneously provide
This advantage is particularly useful for the processing of
non-stationary signals, e.g., PD signals.
It is a desirable property that a denoising technique can
maximize the elimination of the noise while minimizing
the loss of the information-bearing signal of interest.
In the case of wavelet-based denoising technique, a
noisy signal is decomposed into multi-scale wavelet
coefficients by a selected basis function. Those wavelet
coefficients associated with noise are processed by an
estimated threshold, and thus, the significant features of
the signal of interest are retained. Reconstruction is then
performed to build the denoised signal. Based on the
processes of decomposition and reconstruction, wavelet-
based denoising is determined by three aspects: the choice
of wavelet, decomposition scale, and noise or threshold
estimation. The choice of a suitable wavelet basis function
is the first, and most, significant step for the application of
wavelet-based denoising as the wavelet can be translated
and scaled to represent the signal of interest as effectively
as possible. In practice, different denoised signals are
obtained by using different wavelet basis functions [8].
This technique is also applied to the field of wavelet-based
PD denoising. As such, the investigation of an appropriate
wavelet basis function for wavelet-based PD denoising
has been performed in [9], [10].
A wavelet selection scheme was introduced in [9] based
on the correlation coefficient between a known PD
signal and wavelet waveform. This scheme is termed
correlation-based wavelet selection scheme (CBWSS).
The optimal wavelet is desired to generate the highest
wavelet coefficients in wavelet analysis of PD signals,

* jiajia.liu@strath.ac.uk

KEYWORDS
Denoising, detection, partial discharge, wavelet selection, wavelet entropy

Cigre Science & Engineering • N°14 June 2019

73
and thus, the essence of PD signal of interest can be
effectively preserved. Based on this, the wavelet that
can maximize the correlation coefficient is selected as
the most appropriate wavelet for PD denoising. This
approach for best wavelet selection, however, has an
inherent limitation, it requires prior knowledge of PD
waveforms. The waveform of PD signals depends on: the
type and location of PD sources, propagating medium
and path, and the detecting circuit. The variability of
PD waveforms impedes the application of CBWSS for
online PD monitoring systems. Also, it is not a scale-
dependent wavelet selection scheme. The denoised
PD signal may not be as good as expected. The most
significant drawback, however, is that the PD signal is
normally corrupted by the noise in the environment,
which can lead to the selected wavelet being a match
of the noisy PD signal rather than the pure PD signal,
especially when the signal to noise ratio (SNR) is very
low. In an attempt to overcome the limitation mentioned
above in CBWSS, a scale-dependent energy-based
wavelet selection scheme (EBWSS) was presented in
[10]. The wavelet that can maximize the energy ratio of
approximation coefficients at each decomposition scale
is selected as the best wavelet. It has been demonstrated
to outperform CBWSS [10]. In EBWSS, two typical
PD waveforms, damped exponential PD pulse (DEP)
and damped oscillating PD pulse (DOP), were used to
demonstrate the energy criterion for the optimal wavelet
selection. With further exploration in details of EBWSS,
it has been found that the criterion is not strictly true for
DOP signals, particularly when the decomposition scale
increases. The motivation of this paper is therefore to
provide an automated and data-driven selection scheme
for the best wavelet selection in the context of PD
denoising.
In this paper, the new wavelet selection scheme is
inspired by the concept of Shannon Entropy [11], and
the associated information cost functions (ICF) in
information theory [12]–[14]. An ICF can select the best
wavelet to expand a signal in wavelet domain. Wavelet
entropy, derived from Shannon Entropy, can measure
the randomness of the wavelet coefficients at each
decomposition scale. The smaller the wavelet entropy,
the lower the randomness of the wavelet coefficients.
As such, the new selection scheme is proposed with the
combination of ICF and wavelet entropy, and termed
Cigre Science & Engineering • N°14 June 2019
74
wavelet entropy-based wavelet selection scheme
(WEBWSS). Simulated PD signals, i.e., DEP and DOP,
PD signals obtained through laboratory experiment
using test samples with artificial defects and on-site PD
measurements are used to demonstrate the performance
of this novel wavelet selection scheme. Results show that
it is a promising wavelet selection scheme to improve
the effectiveness of PD denoising.
2. Wavelet-Based Technique
2.1 Wavelet Theory
Wavelet transform (WT) is an alternative approach to
traditional methods, e.g. Fourier Transform, in signal
processing. The major advantage of WT is that it can
map a signal in the time-frequency plane. Due to this
advantage, WT is a promising technique in the analysis
of variations in signals or images with the requirements
of both time and frequency information. WT can be
interpreted as an expansion or decomposition of signals
or images in terms of a wavelet that can be scaled in an
auto-similar way [15]. The orthogonal property of the
wavelet used for the expansion or decomposition is the
essence in WT. Generally, WT is achieved through the
application of continuous wavelet transform (CWT) or
discrete wavelet transform (DWT). DWT is preferable
due to its representation of signals or images through
its DWT coefficients without redundancy and, thus, is
using less computational time. In this paper, the wavelet-
based technique referred to is the DWT.
In [15], the wavelet expansion of a signal x can be
expressed as
(1)
where both i and j are integer, i is the time-delay index
and j is the scale index. is the expansion set of
wavelet basis functions, and is the set of expansion
coefficients, or wavelet coefficients, which is called the
discrete wavelet transform of x. The expansion in (1)
is the inverse discrete wavelet transform (IDWT). The
scheme of DWT for signal decomposition is depicted in
Figure 1. A signal is convolved with the low- and high-
pass filters and followed by a downsampling operation
by 2. In signal processing terminology, the outputs of
the low- and high-pass filters are termed approximation
and detail coefficient respectively. The approximation can be seen that the frequency band of low-pass filter is
coefficient is used as the input signal for next-scale 0 ~ fs/2J+1 and the frequency band of high-pass filters is
decomposition. This decomposition is iterated until the fs/2J+1 ~ fs/2 for a full-scale decomposition.
predefined scale, J, reaches. It is important to note that HJ(w) GJ(w)
the maximum decomposition scale Jmax is defined as
log2(N), where N is the length of the input signal. The G3(w) G2(w) G1(w)

reconstruction of the input signal, i.e., inverse DWT 0 fs fs fs fs fs fs

2J+1 2J 16 8 4 2
(IDWT), is a reverse operation as shown in Figure
1. Instead of downsampling in DWT, upsampling is Figure 3. The frequency bands of filters at each decomposition scale
involved in IDWT. Figure 2 shows the processes of 2.2 Wavelet-based Denoising
IDWT for signal reconstruction. To obtain perfect
signal reconstruction, the low- and high-pass filters are The wavelet denoising theory is dependent on the
designed as quadrature mirror filters (QMFs). fundamental idea that the energy of a signal is often
concentrated in only a few coefficients while the energy
of noise is widely spread among all the coefficients in
Scale 2

h
Scale 1 2 the wavelet domain [9], [16], [17]. General procedures
h Approx. cfs
2
Low-pass filter
Downsampling
for the wavelet-based denoising of a signal are presented
as follows:
g
Input data Low-pass filter
Downsampling 2
g
2
High-pass filter 1) Apply DWT to decompose the noisy signal s with a
Detail cfs
High-pass filter selected wavelet to a predefined scale J, and obtain
Figure 1. The implementation of DWT in signal decomposition approximation coefficients aJ at the final scale J and
detail coefficients dj at decomposition scale j, where
Scale 2
~
h
j = 1, 2, ..., J.
2 Scale 1
2) Estimate the threshold through a noise estimation
Approx. cfs ~
h
2 technique and apply this threshold to detail
Upsampling Low-pass filter

~ Output data
coefficients, dj, at decomposition scale j using hard
g Low-pass filter
2 Upsampling
~
High-pass filter
Detail cfs
2
g
or soft thresholding scheme.
High-pass filter 3) Apply IDWT to the approximation coefficients aJ
Figure 2. The implementation of IDWT in signal reconstruction and the processed detail coefficients d'j to reconstruct
the denoised signal s'.
Equally, the implementation of DWT in signal
Based on the noise estimation technique proposed in [9],
decomposition also can be interpreted in the frequency
[18], the scale-dependent threshold used in this paper is
domain. As shown in Figure 1, the DWT process is
estimated by
equivalent to filtering the signal by the filter pairs, the
low-pass filter h and the high-pass filter g. Ideally, these (2)
filter pairs halve the frequency band with the increase
of scale. Let fs be the sampling frequency of the input
where MAD|·| is the median absolute deviation of the
signal, the frequency band, G1(w), of the output of the
detail coefficients dj at decomposition scale j, and nj
high-pass filter is fs/4 ~ fs/2, while the frequency band,
is the length of dj . For the thresholding scheme, soft
H1(w), of the output of the low-pass filter is 0 ~ fs/4.
thresholding in [18] is used in this paper, the function
For next scale, H1(w) is further split into G2(w) and
is given by
H2(w), which are fs/8 ~ fs/4 and 0 ~ fs/8, respectively. The
frequency band is iteratively halved in the subsequent
(3)
decomposition in the same manner until the predefined
scale reaches. For a J-scale DWT, the distribution of the
corresponding frequency bands is shown in Figure 3. It where i = 1,2, ..., nj.

Cigre Science & Engineering • N°14 June 2019

75
2.3 Wavelet Entropy
The concept of wavelet entropy was derived from Shannon
entropy and presented in [19]. Suppose
wavelet coefficients obtained through a J-scale wavelet
transform, in which j represents the decomposition scale
and j = 1, 2, ..., J, i denotes the ith element in Cj,i and
i = 1, 2, ..., nj, nj is the length of wavelet coefficients at scale
j. The energy of wavelet coefficients at the decomposition
scale j can be calculated by
The distribution of energy probability for wavelet
coefficients at scale j can be derived by
3. Partial Discharge Signals
Two theoretical PD pulses, i.e., damped exponential PD
are the pulse (DEP) and damped oscillating PD pulse (DOP), are
simulated using their mathematical frames derived based
on two different PD detecting circuits [4], [10]. In this
paper, DEP and DOP are given by the formula in [10]:
(8)
(9)
(4) where s1(t) and s2(t) are the DEP and DOP respectively.
The values of A, α1, α2, fd, Wd and φ used in these two
equations are listed in Table 1.
Table 1. Values of paramet

(5)

with Σi pi = 1. Wavelet entropy WE(j) is defined as follows

[19]:

(6)
ers used in (8) and (9) [10].
Similar to Shannon entropy, wavelet entropy is applied
to measure the degree of disorder of wavelet coefficients
The simulated sampling frequency fs is set to 60MHz.
or signify the randomness of wavelet coefficients. It is
Figure 4 shows these two simulated PD signals both in
important to note that wavelet entropy is not an information
time and frequency domains. Generally, DOP signal
cost function (ICF), since it requires the energy of wavelet
shown in Figure 4 (c) and (d) is closer to a real high-
coefficients to be normalized as shown in (5), and is thus
frequency PD signal detected from electrical power
not additive [13], [14]. Substituting (5) into (6) yields:
equipment in practice [10]
To develop the new scheme for practical use, real PD
signals were generated through an artificial defect of a
7mm × 7mm breach in the outer conductor created in a
1.5m 11 kV ethylene propylene rubber-insulated (EPR)
cable sample [20]. PD signals were collected using a high
frequency current transformer (HFCT). The specifications
of the HFCT are listed in Table 2. Details regarding the
experiment setup are depicted in Figure 5 [20].
Table 2. Specifications of the HFCT.
Parameters HFCT
Sensitivity 5 V/A
-3 dB bandwidth 90 kHz – 20 MHz
(7) Internal diameter 50 mm
External diameter 110 mm
In (7), l is an ICF based on the definition in [13]. As such,
Load resistance 50 Ω
wavelet entropy is a monotonic-increasing function of
l, which means minimizing l over wavelet coefficients Output conductor BNC
minimizes wavelet entropy. Manufacturer IPEC

Cigre Science & Engineering • N°14 June 2019

76
 Figure 4. (a) and (b): DEP signals simulated in time and frequency domain respectively; (c) and (d): DOP
signals simulated in time and frequency domain respectively.

Stress Artificial
Voltage cone defect
Zn probe
Cable sample
Ck Earth Strap
HFCT
Tr

Zm LeCroy 104Xi, 1GHz

+25
AC dB
Wide-band
amplifier
-30
dB
Wide-band
attenuator

Figure 5. PD testing of a defective 11 kV EPR cable. HFCT was used to collect PD pulses (Ck and Zm
represent the coupling capacitor and measuring impedance respectively).

Experiments were performed at various voltage levels. in this paper. One PD pulse, named , with 2048 sample
The PD pulses measured at 9kV are used as the real PD points was selected and depicted in Figure 6.
signals to demonstrate the new wavelet selection scheme

Figure 6. Real PD pulse, s3, detected from a defective EPR cable under 9kV AC voltage

Cigre Science & Engineering • N°14 June 2019

77
4. Wavelet Selection Schemes
(WSS)
4.1 Correlation-based Wavelet Selection Scheme
In signal processing, correlation is a measure of association
between two signals, and most commonly used is the
linear correlation coefficient. For two signals, xi, and yi,
i = 1, 2, ..., N, the normalized correlation coefficient γ is
given by [9]
(10)
where is the mean of xi and is the mean of yi. The
value of γ is in the range of -1 to 1. It takes on a valueclose
to 1 indicating and are positively correlated, and a value
close to -1 denoting they are negatively correlated. A
value of γ near zero means xi and yi are uncorrelated.
For CBWSS [9], correlation is used as a measure of the
similarity between a pure PD signal and a wavelet, and
this similarity is referred to as their shapes. The more
similar their shapes, the higher the correlation coefficient
is. The wavelet that has the highest correlation coefficient
with the shape of a PD signal is selected to maximize the
wavelet coefficients through wavelet analysis.
The general process for the choice of appropriate wavelet
using CBWSS is described as follows:
a. Analyze the detected PD signal to generate a ‘typical’
PD pulse.
b. Set up a wavelet library, consisting of the wavelets that
have similar characteristics to the PD pulse.
c. Normalize the PD pulse and each wavelet retrieved
from the wavelet library.
d. Calculate the correlation coefficient, , between the PD
pulse and each wavelet.
e. Select the wavelet that has the maximum correlation
coefficient with the PD pulse, it will be applied for the
following wavelet-based denoising.
As indicated in Section 1 the CBWSS approach is limited
by noise and is scale-independent. Also, a heuristic method
was introduced in [10] to obtain better correlation results.
Resampling both the PD signal and wavelet function in
time domain is applied to align their peaks as well as their
first zero-crossing points after the peaks. This method is
Cigre Science & Engineering • N°14 June 2019
78
also adopted in this paper for comparisons of denoising
results between EBWSS and WEBWSS methods.
4.2 Energy-based Wavelet Selection Scheme
EBWSS was proposed by Li [10], in which the wavelet
that can maximize the energy ratio of the approximation
coefficients is selected as the best wavelet for PD denoising.
For a one-dimensional wavelet decomposition, the energy
ratio of approximation aj at scale j is defines as follows [10]:
(11)
where i = 1, 2, ..., ni, ni , is the length of approximation
coefficients or detail coefficients at scale j, and dj is the
detail coefficients at scale j.
The idea of wavelet energy was introduced in EBWSS.
For an orthogonal wavelet, energy preservation is one of
the desirable properties of DWT [21]. The equation for
energy preservation is given by
(12)
where α and d are the approximation and detail
coefficients of the DWT of a signal X. This property is
also applied to PD signals using DWT decomposition. A
PD signal can be decomposed into J scales with signals,
i.e., s1, s2, ..., sJ, sJ+1. Among these signals, s1, s2, ..., sJ, are
detail coefficients from scale 1 to scale J, while sJ+1 is
the approximation coefficients at scale J The energy of a
decomposed signal sk is given by
(13)
where k = 1, 2, ..., J + 1, i = 1, 2, ..., ni, and ni is the length
of si. Then, s can be represented by a normalized energy
vector (e1 , e2 ,..., eJ, eJ+1), where ek is defined as
(14)
It can be seen that the concept of energy ratio in EBWSS
can be interpreted as a normalized energy vector. Figure 7
shows the DEP, DOP, and white Gaussian noise (WGN)
used in [10] to explain the criterion of EBWSS for wavelet
selection. Figure 7 (a) and (b) show the DEP signal and
its normalized energy vector respectively. Equally, Figure
 Figure 7. Representation of (a): DEP, (c): DOP, and (e): WGN by normalized vectors (b),
(d) and (f) respectively.

7 (c) and (d) show the DOP signal and its normalized
energy vector. Figure 7 (e) and (f) show WGN and its
normalized energy vector. Based on Figure 7 (b), (d) and
(e), the approximations of the DEP and DOP signals cover
the most energy of total coefficients while the details of
WGN preserve the most energy of total coefficients [10].
The general process for the choice of an appropriate
wavelet using EBWSS is presented as follows:
a. Given a wavelet library {ψi:i= 1, 2, ..., N}, select a
wavelet from {ψi}, and perform a one-scale DWT
decomposition of a noisy PD signal. Obtain its
approximation coefficients a1(i) and detail coefficients
d1(i).
b. Calculate the energy ratio of approximation coefficients
based on (11). If is the maximum of
, ψp, select as the optimal wavelet
for the first scale.
c. Apply ψp to obtain the approximation coefficients a1(p)
and d1(p).
d. a1(p) is used as the input signal for next-scale DWT
decomposition, and select the optimal wavelet based
on the strategy used in steps a, b, and c.
e. Iterate the steps above until the predefined
decomposition scale J reaches. The optimal wavelet
for each decomposition scale will be selected.
Results in [10] show that EBWSS outperforms CBWSS
for the best wavelet selection. However, EBWSS is not
as robust as expected. It selects the wavelet that can
maximize the energy ratio of approximation coefficients.
It is not strictly true for DOP signals, particularly when
the decomposition scale increases. It can be seen from
the normalized vector of DOP in Figure 7 (d), the energy
of PD signal with a 6-scale decomposition is preserved
on the detail coefficients rather than approximation
coefficients. When more scales are required, for example,
7 scales, the EBWSS is still trying to select the appropriate
wavelet by maximizing the energy ratio of approximation
coefficients.
The limitation of EBWSS can be interpreted in the
frequency domain. Based on Parseval’s theorem, the time
and frequency domains are equivalent representations

G4(w) G3(w) G2(w) G1(w)

H6(w) G6(w) G5(w)

Figure 8. Spectrum of (a): DEP, (b): DOP, and (c): WGN.

Cigre Science & Engineering • N°14 June 2019

79
of the signal, and thus, they must have the same energy
Start
[22]. As mentioned in Section 2.1, the filter pairs of
DWT iteratively halve the frequency bands of a signal Load PD signal, s(n), set Up a
with the increase of decomposition scales. The spectrum wavelet library, Ψi, i = 1 ,2,…,N,
and the decomposition scales, J
of DEP, DOP, and white Gaussian noise are illustrated
in Figure 8 (a), (b), and (c) respectively. With a 6-scale DWT of s(n) using Ψi for one-scale
decomposition, i = 1, j = 1
decomposition, the filter pairs iteratively separate these
signals into disjoint frequency bands, G1(w), G2(w), ..., Obtain the approx. cfs, aj(i)
G6(w) and H6(w) From the spectral curve of DOP, it is
clear that the magnitudes of frequency in G5(w) and G6(w) i=i+1
Calculate the energy of aj(i) and its
energy probability (equation (5))
are larger than those at other frequency bands. It is in
agreement with the normalized vector of DOP shown in Calculate the wavelet entropy, and
store it into a entropy vector WE(i)
Figure 8 (b). With further decomposition, the energy of (equation (6))

the signal will be preserved in detail coefficients rather

No
than approximation coefficients. i=N? Set s(n) = aj(p) for optimal wavelet
selection at next scale, j = j+1

4.3 Wavelet Entropy-based Wavelet Selection Scheme Yes

[~,p] = min(WE(i)), p [1,N]

As mentioned in Section 2.3, wavelet entropy is not
an ICF, but in (7) is monotonically increased with the Ψp is selected as the optimal
wavelet entropy. As such, the best wavelet also can be wavelet for scale j and stored into
an wavelet vector ow(j)
selected when the value of wavelet entropy is minimum.
Obtain the approx. cfs aj(p)and
In [13], it was shown that wavelet entropy value is detail cfs dj(p) for scale j using Ψp
inversely proportional to the energy concentrated in the
number of wavelet coefficients. It is also known that white Store approx. & detail cfs to the
vectors a and d respectively
noise, the noise source for PD corruption in this paper,
has high degree of randomness or disorder, and thus, the
No
entropy value can describe the random characters of noise j=J?

[23]. Based on this, a smaller value of wavelet entropy Yes

indicates that the wavelet used for WT decomposition End

can preserve more energy of the original signal in fewer

number of coefficients and contain less white noise in the
wavelet coefficients and, consequently, the wavelet used
is closer to the best wavelet as expected. A new criterion
for the best wavelet selection is therefore proposed, i.e.,
a wavelet that can have minimum wavelet entropy of the
approximation coefficients at each decomposition scale
through WT decomposition will be selected for denoising
of PD detection. The new method has several promising
advantages: it is scale-dependent, automated, and data-
driven.
The general process for the proposed novel wavelet
selection scheme is illustrated in the flow chart in Figure 9.
Given a wavelet library {ψi:i= 1, 2, ..., N}, one wavelet of
which is selected for a one-level DWT decomposition of a
noisy PD signal s(n) each time. Next the wavelet entropy
of the generated approximations is calculated based on (6)
Figure 9. Flow chart of the general process of WEBWSS.
and (7). The wavelet ψp( 1 ≤ p ≤ N) that minimize the
wavelet entropy of approximations will be selected ψp
as the best wavelet. The selected is then applied for the
DWT decomposition of s(n) for the first scale, obtaining
approximation coefficients a1(p) and detail coefficients
d1(p). Finally, a1(p) is used as the input signal for next scale
DWT decomposition, using the strategy presented above.
When the predefined decomposition scale J reaches, the
best wavelet for each scale will be successfully selected.
5. Results and Analysis
Generally, parameters, e.g., magnitude error (ME),
mean square error (MSE), signal to noise ratio (SNR),
and cross correlation (XCORR) are adopted to evaluate

Cigre Science & Engineering • N°14 June 2019

80
 Figure 10. (a) and (b): s1 and its noisy signal NS1 with SNR = -10, (c) and (d): s2 and its noisy signal NS2
with SNR = -10, (e) and (f): s3 and its noisy signal NS3 with SNR = -10.

the performance of a proposed denoising method or
algorithm. ME, MSE, and XCORR are used in this
paper to compare the denoising results of different
wavelet selection schemes. XCORR is calculated based
on (11), and ME, MSE are calculated by the equations
as follows,
(15)
(16)
where m and m' are the magnitudes of s(i) and s'(i)
respectively. s(i) represents the original signal and s'(i)
denotes the denoised signal. N is the length of signals.
Better denoised results can be obtained with lower ME,
MSE, and higher XCORR.
In this paper, PD signals are corrupted by white noise,
and then, various wavelet selection schemes are used
to remove the noise and evaluated by the parameters
mentioned above. As mentioned in Section 4.2,
EBWSS is not strictly true when the decomposition
scale increases over 6. To highlight the limitations of
EBWSS, the decomposition scale is set to 7 for the
results analysis. Two simulated PD signals, s1 and s2,
and real PD signal, s3, as well as their noisy signals NS1,
NS2 and NS3 with SNR = -10 are depicted in Figure 10.
It is important to note that the original real PD signal
shown in Figure 6 is corrupted by ambient noise during
experiment. To mitigate the effect of this noise on the
denoising results, it has been pre-processed using the
method introduced in [10]. The smoothed real PD
signal is depicted in Figure 10 (e).
Denoised S1, S2, and S3 using CBWSS, EBWSS, and
WEBWSS are shown in Figure 11 (a), (b), and (c)
respectively. The related parameters used to evaluate

Figure 11. (a): Denoised s1 using CBWSS, EBWSS and WEBWSS, (b): Denoised s2 using
CBWSS, EBWSS and WEBWSS, (c): Denoised s3 using CBWSS, EBWSS and WEBWSS.

Cigre Science & Engineering • N°14 June 2019

81
their performance on PD detection are listed in Table 3.
It can be seen from Figure 11 and Table 3 that WEBWSS
has better performance than the others of wavelet-based
denoising of PD detection. Simultaneously, it also
verifies the conclusion presented in [10] that EBWSS
outperforms CBWSS in PD denoising.
Significant advancement of the proposed scheme in
the denoising of single PD signals cannot be directly
seen from Figure 11. However, two columns in Table
3, Improvement 1 and Improvement 2, have presented
the improvements by WEBWSS. Improvement 1 is the
improved ratio (%) of the use of WEBWSS to CBWSS
and Improvement 2 is the improved ratio (%) of the
use of WEBWSS to EBWSS. From magnitude error
(ME), mean square error (MSE), and cross-correlation
coefficients (XCORR), significant improvements can be
seen from these figures. The underlying meaning of the
improvement of ME is PDs with small magnitude may
be picked up by the use of WEBWSS as compared to
the other two schemes. This enhanced capability of PD
detection has been verified through the application of
WEBWSS in on-site PD data, which will be delineated
in Figure 15. Also, the improvement of MSE and
XCORR indicates that less distortion of the denoised
signals can be achieved through WEBWSS. It is good
for the accuracy of PD location.
In the attempt to fully evaluate the performance of the
new wavelet selection scheme, PD signals buried by
various noise levels are investigated. SNR are set to
-15, -10, -5, 0, and 5 for this investigation, representing
different noise levels. The parameters used to evaluate
the performance of three different schemes are plotted
in Figure 12, Figure 13, and Figure 14 for S1, S2, and S3
respectively. From the trends of ME, MSE, and XCORR
for all denoised signals with the increase of SNR (the
higher SNR, the lower noise level is), the new wavelet
selection scheme is better than the existing schemes for
wavelet-based denoising of PD detection under various
noise levels, and the performance is particularly good
when the SNR is low.
A PD signal is collected from one power substation in
the UK with a sample rate of 100MS/s. The PD sensor is
the same type of HFCT used in laboratory in Section 3.
Figure 15 delineates the original on-site PD signal and its
denoised versions by CBWSS, EBWSS, and WEBWSS,
respectively. It can be seen that not only the PD pulses
with magnitudes higher than the noise level has been
extracted, but those ones with small magnitudes buried
in the noise has been successfully extracted through the
application of WEBWSS in Figure 15 (d). However,
the number of the PD pulses with small magnitude that
have been extracted by CBWSS and EBWSS, as shown
in Figure 15 (b) and (c), is less than that by WEBWSS.
The difference of the number of small-magnitude
PD extraction among these three schemes has been
highlighted in red line in Figure 15. The denoising result
of on-site PD signal provides further support that the
proposed wavelet selection scheme is more advantageous
than the existing EBWSS and CBWSS for denoising of
PD detection of electrical apparatus in practice.

Table 3. Parameters used to evaluate the performance of wavelet selection schemes.

CBWSS EBWSS WEBWSS Improvement 1 (%) Improvement 2 (%)

ME 0.3701 0.3457 0.3071 17 11.2

s1 MSE 0.0055 0.0028 0.0024 56.4 14.3

XCORR 0.8689 0.9358 0.9495 8.5 1.4
ME 0.5686 0.5673 0.5238 7.9 7.7

s2 MSE 0.0054 0.0044 0.0041 24.1 6.8

XCORR 0.8757 0.9182 0.9414 7.0 2.5
ME 0.5937 0.5457 0.4997 15.8 8.4

s3 MSE 0.00035 0.00037 0.00027 22.9 27

XCORR 0.8501 0.9369 0.9502 10.5 1.4

Cigre Science & Engineering • N°14 June 2019

82
 Figure 12. ME, MSE and XCORR between s1 and denoised s1.

Figure 13. ME, MSE and XCORR between s2 and denoised s2.

Figure 14. ME, MSE and XCORR between s3 and denoised s3.

Figure 15. The denoising results of on-site PD signal: (a) On-site detected PD signal; (b), (c), (d) are the denoised PD
signal by CBWSS, EBWSS, and WEBWSS, respectively.

Cigre Science & Engineering • N°14 June 2019

83
6. Conclusion
A novel wavelet selection scheme based on the concept of
information cost function and Shannon entropy has been
proposed in this paper and the results showed that the
technique improved the effectiveness of wavelet-based
denoising of PD detection, especially for situations when
the signal to noise ratio is unity or less. The technique
is scale-dependent, automated, and data-driven and these
properties enable it to be a promising technique in the
context of PD detection. Based on the denoised results
obtained from both simulated and real PD signals, it shows
better performance than the existing wavelet selection
schemes, particularly when the SNR is low. Accordingly,
it has the potential to extend the range of PD detection in
cables through the application of this technique.
7. References
[1] H. Borsi, “Digital location of partial discharges in HV cables,” IEEE
Trans. Electr. Insul., vol. 27, no. 1, pp. 28–36, 1992.
[2] M. S. Mashikian, F. Palmieri, R. Bansal, and R. B. Northrop,
“Location of partial discharges in shielded cables in the presence of
high noise,” IEEE Trans. Electr. Insul., vol. 27, no. 1, pp. 37–43, 1992.
[3] I. Shim, J. J. Soraghan, and W. H. Siew, “Detection of PD Utilizing
Digital Signal Processing Method. Part 3 : Open-Loop Noise,” Electr.
Insul. Mag. IEEE, vol. 17, pp. 6–13, 2001.
[4] X. Ma, C. Zhou, and I. J. Kemp, “Interpretation of wavelet analysis
and its application in partial discharge detection,” IEEE Trans.
Dielectr. Electr. Insul., vol. 9, no. 3, pp. 446–457, 2002.
[5] L. Satish and B. Nazneen, “Wavelet-based Denoising of Partial
Discharge Signals Buried in Excessive Noise and Interference,” IEEE
Trans. Dielectr. Electr. Insul., vol. 10, no. 2, pp. 354–367, 2003.
[6] H. Zhang, T. R. Blackburn, B. T. Phung, and D. Sen, “A novel wavelet
transform technique for on-line partial discharge measurements
part 1: WT de-noising algorithm,” IEEE Trans. Dielectr. Electr.
Insul., vol. 14, no. 1, pp. 3–14, 2007.
[7] P. Ray, A. K. Maitra, and A. Basuray, “Extract Partial Discharge
Signal using Wavelet for On-line Measurement,” Int. Conf.
Commun. Signal Process., pp. 888–892, 2013.
[8] Y. F. Sang, D. Wang, J. C. Wu, Q. P. Zhu, and L. Wang, “The relation
between periods’ identification and noises in hydrologic series data,”
J. Hydrol., vol. 368, pp. 165–177, 2009.
[9] X. Ma, C. Zhou, and I. J. Kemp, “Automated wavelet selection and
thresholding for PD detection,” IEEE Electr. Insul. Mag., vol. 18, no.
2, pp. 37–47, 2002.
[10] J. Li, T. Jiang, S. Grzybowski, and C. Cheng, “Scale dependent
wavelet selection for de-noising of partial discharge detection,” IEEE
Trans. Dielectr. Electr. Insul., vol. 17, no. 6, pp. 1705–1714, 2010.
[11] C. E. Shannon, “A Mathematical Theory of Communication,” Mob.
Comput. Commun. Rev., vol. 5, no. 1, 1948.
[12] R. R. Coifman and M. V. Wickerhauser, “Entropy-based algorithms
for best basis selection,” IEEE Trans. Inf. Theory, vol. 38, no. 2, pp.
713–718, 1992.
Cigre Science & Engineering • N°14 June 2019
84
[13] M. V. Wickerhauser, Adapted Wavelet Analysis from Theory to
Software, vol. 38, no. 1. IEEE Press, 1996.
[14] H. Šikić and M. V. Wickerhauser, “Information cost functions,”
Appl. Comput. Harmon. Anal., vol. 11, no. 2, pp. 147–166, 2001.
[15] S. Alessio, Digital Signal Processing and Spectral Analysis for
Scientists: Concepts and Applications. Springer, 2015.
[16] A. O. Boudraa, J. C. Cexus, and Z. Saidi, “EMD-based Signal Noise
Reduction,” Int. J. Signal Process, pp. 33–37, 2004.
[17] Y. Kopsinis and S. McLaughlin, “Development of EMD-Based
Denoising Methods Inspired by Wavelet Thresholding,” Signal
Process. IEEE Trans., vol. 57, no. 4, pp. 1351–1362, 2009.
[18] D. L. Donoho and I. M. Johnstone, “Ideal spatial adaptation by
wavelet shrinkage,” Biometrika, vol. 81, no. 3, pp. 425–455, 1994.
[19] Y. Zhuang and J. S. Baras, “Existence and Construction of Optimal
Wavelet Basis for Signal Representation,” IEEE Trans. Signal
Process., 1994.
[20] X. Hu, W. H. Siew, M. D. Judd, and X. Peng, “Transfer function
characterization for HFCTs used in partial discharge detection,”
IEEE Trans. Dielectr. Electr. Insul., vol. 24, no. 2, pp. 1088–1096,
2017.
[21] M. Misiti, Y. Misiti, G. Oppenheim, and J.-M. Poggi, Wavelet
ToolboxTM User’s Guide. 2015.
[22] S. W. Smith, The Scientist and Engineer’s Guide to Digital Signal
Processing, Second Edi. California Technical Publishing, 1999.
[23] Z. He, X. Chen, and G. Luo, “Wavelet Entropy Measure Definition
and Its Application for Transmission Line Fault Detection and
Identification; (Part I: Definition and Methodology),” 2006 Int.
Conf. Power Syst. Technol., pp. 1–6, 2006.
8. Biographies
Jiajia Liu was born in Jiangsu, China, in 1985. He received
the BSc degree in electrical engineering from Shanghai
Maritime University in 2007, the MSc degrees in electrical
engineering from Shanghai Maritime University and the
University of Southampton in 2009 and 2014 respectively.
He is currently pursuing his PhD degree at the University
of Strathclyde. His major research interest includes partial
discharge monitoring, and transformer thermal modeling.
W. H. Siew is a Professor in the Department of Electronic
& Electrical Engineering, University of Strathclyde,
Glasgow, U.K. He is a Triple Alumnus of the University of
Strathclyde with the B.Sc. (Hons) degree in electronic and
electrical engineering; the Ph.D. degree in electronic and
electrical engineering; and the M.B.A. degree. His areas
of research interest include large systems electromagnetic
compatibility, cable diagnostics, lightning protection,
and wireless sensing systems. He was the Convener of
the CIGRE WG C4.208 and of CIGRE WG C4.30. He
is a member of the Advisory Group for CIGRE SC C4,
which addresses System Technical Performance of Large
Electric Systems. He is a Chartered Engineer and is an
MIEEE and an MIEE.
John J. Soraghan (S’83–M’84–SM’96) received the
B.Eng. (Hons.) and M.Eng.Sc. degrees from University
College Dublin, Dublin, Ireland, in 1978 and 1983,
respectively, and the Ph.D. degree from the University
of Southampton, Southampton, U.K., in 1989, all in
electronic engineering. His doctoral research focused on
synthetic aperture radar processing on the distributed array
processor. After graduating, he worked with the Electricity
Supply Board in Ireland and with Westinghouse Electric
Corporation in the United States. In 1986, he joined the
Department of Electronic and Electrical Engineering,
University of Strathclyde, Glasgow, U.K., as a Lecturer and
became a Senior Lecturer in 1990, a Reader in 2000, and a
Professor in signal processing in September 2003, within
the Institute for Communications and Signal Processing
(ICSP). In December 2005, he became the Head of the
ICSP. He currently holds the Texas Instruments Chair
in Signal Processing with the University of Strathclyde.
He was a Manager of the Scottish Transputer Centre
from 1988 to 1991 and a Manager of the DTI Parallel
Signal Processing Centre from 1991 to 1995. His main
research interests include signal processing theories,
algorithms, and architectures with applications to remote
sensing, telecommunications, biomedicine, and condition
monitoring. Prof. Soraghan is a member of the Institution
of Engineering and Technology.
Cigre Science & Engineering • N°14 June 2019
85
Xiao Hu graduated from Xi’an Jiaotong University with a
BSc degree and a MSc degree in electrical engineering in
2006 and 2009, respectively. He received his PhD degree
from the University of Strathclyde in 2014 for research
on FDTD modeling of partial discharge in high voltage
cables. He is currently a lecturer in Guizhou University,
China and his research includes partial discharge
measurement and modeling.
Xiaosheng Peng received the B.Sc. And M.Sc. degrees
from Huazhong University of Science and Technology,
China in 2006 and 2009, respectively, and the Ph.D.
degree in electrical engineering at Glasgow Caledonian
University in 2012 funded by EPSRC. He has worked
as a Post-Doctoral Researcher in Glasgow Caledonian
University funded by EDF Energy. He is currently a
lecturer in school of Electrical and Electronic Engineering
of Huazhong University of Science and Technology. His
research interests are partial discharge signal processing
and condition monitoring of power plant. He is a member
of IET.
Euan A. Morris was born in Glasgow, Scotland, in 1991.
He received the MEng. degree in electrical and electronic
engineering with business studies from the University of
Strathclyde, Glasgow in 2014 and is currently undertaking
the Ph.D. degree at the EPSRC Centre for Doctoral
Training in Future Power Networks and Smart Grids, run
by the University of Strathclyde, Glasgow and Imperial
College London, London.
Real-time co-simulation model using
electromagnetic transient and dynamic
phasor simulations
H. KONARA*, U. D. ANNAKKAGE**
University of Manitoba, Canada
C. KARAWITA***
TransGrid Solutions Inc., Canada
This paper presents a co-simulation model to enable
simulation of large power systems in real-time. One
approach for co-simulation is to model a small part of the
network in detail using an EMT simulation and model the
rest of the network using a Transient Stability (TS) model.
To capture the high-frequency behavior of the system,
Frequency Dependent Network Equivalents (FDNE) are
used at interfacing busses. This paper proposes a different
approach where the TS model is replaced with a Dynamic
Phasor (DP) model. Since the DP model covers a wider
bandwidth, an FDNE is not needed. The numerical
instability of a real-time EMT-DP interface due to time-
step delay is overcome by using a data extrapolation
technique. The proposed co-simulation platform was
tested in real-time with detailed synchronous machine
models in both the EMT and DP parts of the network.
1. Introduction
The Electromagnetic Transient (EMT) simulation is
the most accurate tool available for simulating power
systems. EMT simulations can capture dynamics in a
broad frequency bandwidth and therefore, can accurately
model fast dynamics such as switching, dynamics of
HVDC systems, etc. However, EMT simulations can be
computationally demanding since it uses a small time-
step (50μs or less). Due to the computational burden of
EMT models, a simulation can run for an unaccceptable
duration making EMT time simulation unsuitable for a
large power system. Real-time simulation tools use EMT
modeling and updates its values each time-step in real
time; i.e. 50μs in the simulation is 50μs in the real world.
Using real time simulation for a large power system
*konarhkk@myumanitoba.ca
**udaya@ieee.org
***ckarawita@tgs.biz
KEYWORDS
Co-simulation, dynamic phasors, electromagnetic transient, real-time, transient stability.
Cigre Science & Engineering • N°14 June 2019
86
would require a lot of parallel processing hardware which
is expensive. On the other hand, Transient Stability (TS)
models use a larger time-step (half a cycle) and simpler
models compared to EMT simulations delivering faster
simulations. As a result, TS models are used to simulate
very large power systems. Because of the larger time-step
and simpler models used, TS programs are only accurate
enough to model slow electromechanical transients. The
need of a simulation model that is capable of simulating
a large power system while accurately capturing high
frequencies is overcome by co-simulation models (also
known as hybrid simulations) [1].
Interfacing transient stability-type simulation models to
EMT simulation for analyzing large power systems has
been reported in [1]-[4]. Typically, a small part of the
network containing power electronic devices is modeled
in detail at the switching level using an EMT model
and the rest of the system (external system) is modeled
using a TS model. Since EMT is a high-frequency model
and TS is a low-frequency model, interfacing them into
one simulation platform is not possible without using
Frequency Dependent Network Equivalents (FDNE) at the
boundary buses [5]. In this approach, TSA represents the
slow electromechanical behavior of the external network
and FDNE represents the high-frequency behavior of the
external system. FDNE has a fitted admittance. A rational
function is derived from the frequency response of the
system using curve-fitting techniques. A lumped parameter
model is then derived from the rational transfer function
using a recursive convolution method. FDNE improves
the accuracy of the EMT-TS simulations by representing
the high-frequency behavior of the TS network. However,
FDNE reduces the computational efficiency in the
interface dramatically [4]. The passivity of the FDNE
is a necessary requirement for the numerical stability of
the co-simulation [6], [7]. Many fitting methods tend to
produce non-passive models [8]. If the derived rational
function does not preserve passivity then it is made
passive through approximations. Most of the methods
that enforce passivity are not computationally efficient.
Inaccuracies to the simulation may be introduced due
to the approximations that are made in order to preserve
passivity [9]. FDNE also reduces the accuracy of the
simulation by introducing errors when there is a change in
topology in the TS part of the network [2]. Since the state
variables of the FDNE are not properly initialized for the
post-fault condition, spurious transients are generated.
The tuning of the FDNE depends on several factors
including the initial pole assessment, number of iterations,
and frequency band under consideration. The tuning of the
FDNE directly influences the accuracy of the simulation
[10]. Developing automated methods to derive a system
equivalent model is difficult [11]. The user must have a
specialized knowledge to derive the FDNE model for a
particular system, which is not very convenient.
The concept of Dynamic Phasors (DP) is to use the
phasor quantities of power system voltages and currents
by assuming a nearly periodical behavior of the system
[12]-[14]. Similar to TS simulations, the DP also uses
positive sequence phasor quantities to model the power
system. However, in DP, the phasors are modeled as state
variables and hence use differential equations, whereas,
in a TS model, algebraic equations are used to model the
phasors. This way of modeling enables the DP simulation
to capture network dynamics in a wide frequency range.
The constant admittance network representation used in
TS simulations are only valid at the system frequency.
Some applications of dynamic phasors for power system
simulations are analysis of electromagnetic transients
[15], [16], subsynchronous resonance studies [14], HVDC
interaction studies [17], [18], modelling synchronous
machines [19], [20].
In this paper, the TS representation of the external network
used in a typical EMT-TS co-simulation is replaced by a
DP representation. Since a DP model can capture high-
frequency dynamics, the interface between the two models
does not require an FDNE; hence, the objective of this
Cigre Science & Engineering • N°14 June 2019
87
paper is to show that a numerically stable co-simulation
can be achieved without using an FDNE at the boundary
buses. In this paper, the DP part of the network is modeled
as a balanced network and thus more efficient than an
EMT simulation.
Numerical instability is one of the main problems in
real-time co-simulation. One of the main reasons for
this is the time-step delay between the two interfacing
simulation models when exchanging data. In the proposed
method, the numerical instability due to time-step delay is
addressed successfully by data extrapolation.
The synchronous machine in DP as well as in EMT
parts of the network are modeled using detailed machine
models that include stator flux dynamics.
IEEE 68 bus system [21] is used to validate the EMT-DP
co-simulation interface. An LCC-HVDC line is added to
bus 54 to analyze the impact on the AC system from the
HVDC in-feed. A comparison of the results of the real-
time co-simulation model is carried out against a full real-
time EMT simulation.
2. Summary of recent
contributions
DP-EMT co-simulation models are presented in [22],
[23]. In [23], the dynamic phasors are considered for
three phasors. Since the phasors are complex variables
the computational burden is higher than that of EMT
simulation where the variables are real. Therefore, to
gain any computational advantage, a higher time-step
must be used in the DP side. The interface between the
EMT and DP parts of the network takes advantage of the
delay across the traveling wave transmission line [23].
The two networks are decoupled via the transmission
line if the travel time of the transmission line is at least
one simulation time step. The minimum transmission line
length required to decouple the networks is proportional
to the simulation time-step by the following relationship:
s = d/Δt, where the propogation across the line is at the
speed of light (s = 3*108 m/s), Δt is the simulation time-
step, and d is the transmission line length. As an example,
if the time-step of the simulation is 1ms, there should be a
transmission line of length 300km. When choosing a large
time-step the interface transmission line of an artificially
long length will cause inaccuracies to the simulation.
In [23], the simulation study considers only transmission
lines and voltage sources without using synchronous
machines in the simulation. The simulations in [23] are
carried out in a non-real-time environment. Data transfer
between the EMT and DP is done using signal processing
(FFT) techniques in [23]. In order to compute the FFT
samples over a time window are collected and numerical
integration is peformed to obtain FFT components making
the simulation inefficient for real-time applications.
In reference [24], dynamic phasors are used to analyze
the interaction between an HVDC system and AC grids
using EMT-TS co-simulation. Dynamic phasors are used
only for a buffer zone between the EMT and TS model.
It has been shown that an EMT-TS co-simulation can
be carried out without using an FDNE at the interface
if DP is used to replace the TS. The three phases of the
network are independently modeled in the DP buffer zone
which consists only of transmission lines. This is a very
promising outcome that deserves further investigation.
The goal of the research presented in this paper is to
further investigate the approach in [24] where an FDNE
is not needed at the interface between the detailed EMT
model and the external system. In addition, the feasibility
of using data extrapolation when data is transferred from
EMT to the DP model is investigated. The success of the
proposed approach is evaluated in terms of numerical
stability and accuracy. The following points outline the
unique features of this paper:
1. Compared to [23], only positive sequence dynamic
phasors are used in the simulation which results in an
efficiency gain compared to an EMT simulation even
with the same integration time step in both sides.
2. It will be shown that a stable and an accurate
simulation can be obtained by using a data
extrapolation technique to eliminate numerical
effects due to time-step delay in transferring data
between the two simulation models.
3. The proposed approach is validated in a real-time
environment using the Real Time Digital Simulator
(RTDS).
4. The DP part of the network contains full order
synchronous machine models.
5. Three phase instantaneous voltages of EMT voltages
are converted to phasor quantities using an alpha-
beta conversion. This way of conversion does not add
Cigre Science & Engineering • N°14 June 2019
88
a time delay to the simulation compared to methods
like PLL, FFT, consecutive curve fitting techniques
etc. Hence, it is suitable for a real-time simulation.
3. Numerical stability of EMT-DP
interface
Co-simulation is a challenging task due to the time-step
delay between the simulation models. If this time-step
delay is not properly addressed, it could lead to numerical
instability. The numerical instability due to time-step delay
is discussed in [25], [26] with some possible solutions.
The instability occurs when there is an inductor at one
side of the interface and the other side injects a current
into that model.
Two possible solutions to reduce the negative effect of
the time-step delay in EMT co-simulation models are
discussed in [27]. One solution uses the natural time
delay of propagation of signals in a transmission line.
The other method uses extrapolation of data to predict
the values for the next time-step. This paper will explore
the latter technique where a forward time-step shifting of
data is used. Timeshift can be done by predicting future
data using extrapolation. Linear extrapolation is used in
this paper to perform a forward time-step shift. It should
be noted that a smaller data interval will yield a more
accurate result. Iterative approaches are used in [28] to
avoid time-step delays in EMT-TS co-simulation. This
approach has a high computational demand which makes
it not attractive for a real-time simulation.
4. Modeling the network
components in dynamic phasors
The fundamental component dynamic phasor of the signal
u(t) is the complex envelope u+(t) of the signal u(t) shifted
from the fundamental frequency of ω0.
(1)
In traditional transient stability type studies, the derivatives
are substituted by the constant term jω0. This rule is
modified in dynamic phasors to model the dynamics [29].
The derivative of the signal u(t) in dynamic phasors is
given by:
(2)
 Figure 1: Block diagram of a single port of the proposed EMT-DP co-simulation model
Hence, dynamic phasors add the derivative of the phasor
signal to the derivative modeled in transient stability
studies. Using this property, differential equations for a
capacitor current and an inductor voltage can be written
as (3) and (4).
(3)
4)
Where signals Ic(t), Vc(t), VL(t) and IL(t) are the dynamic
phasor voltages and currents. The transmission lines in the
DP side of the network are modelled as PI section models
containing lumped capacitors and inductors.
4.1. Synchronous machine model for transient analysis
Different full order synchronous machine models are
used in EMT simulation platforms. In PSCAD-EMTDC
program, the synchronous machine is modeled as a Norton
equivalent circuit [30]. The numerical issues due to the
time-step delay between the machine and the network are
addressed by using a compensated resistor at the interface
bus.
A stability criterion for interfacing a synchronous
machine to a DP simulation with a time-step delay is
presented in [31]. It has been pointed out that depending
on the input impedance of the network that it is connected
to, the device modeled either as a Norton equivalent
or Thevenin equivalent (not both) will be stable at the
steady state. In DP nodal approach, the synchronous
machine state equations are solved separately from the
network and therefore an inherent time-step delay exists
in the simulation. Typically, the network impedance is
Cigre Science & Engineering • N°14 June 2019
89
higher than the stator impedance and therefore Norton
equivalent-type synchronous machine presented in [30] is
not stable. Hence, in this work, Thevenin equivalent type
[32]-[36] synchronous machine model is used to represent
the machines in the DP side of the network.
5. Implementation of the
proposed system in a Real-Time
platform
The steps of implementing the DP model in an EMT
simulation are summarized in this section. The RTDS real-
time simulator is used as the real-time EMT simulation
platform. The DP part of the network is interfaced to
the EMT model as a current source in parallel with an
impedance. The EMT solution is represented in the DP
model as a known boundary voltage at each time-step.
The time-step delay of reading the boundary voltages
from EMT model at the DP side is avoided by linear
extrapolation of voltages. Figure 1 shows the block
diagram of the complete EMT-DP co-simulation.
5.1. Data conversion between the two simulation
models
The solution of an EMT model is derived using time domain
instantaneous values of three phase voltages and currents
and the DP model uses phasor quantities throughout its
simulation process. Hence, there is a need to convert abc
instantaneous quantites to phasor quantites and vice versa.
The conversion between three phase instantaneous values
to phasors is done in TSA-EMT co-simulation by using
a Fast Fourier Transformation (FFT) [37], consecutive
curve fitting technique [38] or by balancing the energy
in two sides [39]. The real time simulation requires a
computationally efficient convertion method. In our work,
the phasor information is extracted by using the alpha-
beta transformation assuming a constant center frequency
and balanced conditions. Consider the time varying three
phase signals va, vb and vc, where va= Vm sin (θ) and vb, vc
are 120° shifted in phase from va.
The alpha beta voltages (vα, vβ) of the three phase voltage
can be calculated using the abc to α β transformation
matrix (Tα β).
(5)
(6)
The dynamic phasors can be derived from (7) and (8),
where Vm(t) and θ(t) are the magnitude and the phase angle
of the phasors. The terms ω0 refers to the center frequency
and t is the simulation time.
(7)
(8)
The solution of DP simulation can be converted back to a
three-phase signal as
,
by assuming balance conditions.
6. Test System and Validation of
the proposed platform
The proposed interfacing approach is validated using
the IEEE 68 bus system. A portion of IEEE 68 bus case
(shown in blue color in Figure 2.a) containing the buses
11, 30, 32, 33, 34, 38 are modeled using DP and the rest
of the system is modeled as an EMT model. Buses 30,
34 and 38 are the interface buses connected to the EMT
simulation. Therefore, this is a three port interface. The
proposed method is, however, valid for any number of
ports. The generator 1 in the EMT side is replaced by the
CIGRE HVDC benchmark model [40]. Figure 2.b shows
Cigre Science & Engineering • N°14 June 2019
the relevant part of the RSCAD simulation case with the
connection of the user-written DP component to the rest
of the 68 bus system.
(a)
,
(b)
Figure 2: (a) The DP part of the IEEE 68 bus system, (b) Snapshot of RSCAD
draft containing the DP model
The objective of the proposed co-simulation model is to
study the dynamics in the network modeled using EMT.
The entire IEEE 68 bus system is also modeled using
EMT simulation in RTDS for comparison against the co-
simulation model. Faults are applied at different locations
of the 68 bus system to analyze the accuracy of the co-
simulation model. This simulation is carried out using
a time-step of 50μs in both EMT and DP simulations.
A three phase fault (fault resistance is 0.1Ω) is applied
at t = 5s for a period of five cycles at HVDC converter
bus (bus 54). The voltage at the bus 54 and the current
through the DC line is shown in Figure 3. The speed of the
generator 8 which is the closest generator to the HVDC
in-feed is shown in Figure 4. The voltage of bus 30 which
90
is one of the interface buses of EMT-DP shown in Figure
5. As it can be seen from the figures the co-simulation
results closely matches with the EMT simulation results.
Figure 3: Comparison of AC voltage and DC current for a three phase
to ground fault at the converter bus
Figure 4: Comparison generator 8 speed for a three phase to ground fault
at the converter bus
Figure 5: Comparison of bus 30 voltage for a three phase to ground fault
at the converter bus
Cigre Science & Engineering • N°14 June 2019
91
The three-phase fault is then applied at the interface bus
38. The electric torque and speed of generator 11 obtained
with co-simulation are compared against EMT solution in
Figure 6. This confirms that the results of co-simulation
faithfully reproduce the results of the EMT simulation.
Figure 6: Comparison of speed and torque of generator 11 for a three phase to
ground fault at the interface bus 38
Figure 7 shows a comparison of the voltages of phase ‘a’ at
the EMT-DP interface buses. The co-simulation voltages
closely follow the shape of the EMT simulation results
except for minor differences in voltage spikes during the
transient. The phase ‘a’ line current from bus 36 to 34
and its magnitude are shown in Figure 8 and 9. Figure
8 shows the current during the transient and the results
matches with the EMT simulation results. Figure 9 shows
the current magnitude few seconds after the fault, and it
can be seen that the co-simulation and EMT simulation
results settle down at the same value.
 The co-simulation model was also validated by applying
faults at different locations of the IEEE 68 bus system.
The results for a three-phase fault at bus 46 is shown in
Figures 10 to 12. As seen from the figures, co-simulation
results agree with the complete EMT simulation results.

Figure 10: Comparison of speed of generator 11 for a three phase

to ground fault at bus 46

Figure 7: Interface bus voltages for a three phase to ground fault

at the interface bus 38
Figure 11: Phase A current from bus 36 to 34 for a three phase to ground
fault at bus 46

Figure 8: Current from bus 36 to 34 for a three phase to ground fault

at the interface
Figure 12: Phase A voltage at bus 46 for a three phase to ground fault at bus 46

The robustness of the proposed interface was validated

Figure 9: Current magnitude from bus 36 to 34 for a three phase to ground
fault at the interface
by moving the dynamic phasor part of the network to
different locations in the 68 bus case.
It can be shown that when the system is unstable the co-
simulation also simulates it as unstable. This scenario
was created by removing a Power System Stabilizer. The
magnitude of the current in the line between buses 36 and
34 is shown in Figure 13. The results between EMT and

Cigre Science & Engineering • N°14 June 2019

92
DP-EMT agree up until the first 13 seconds at which point will be 4n + 2n + 2(n - 1) flops giving a total flops of
the results begin to diverge. The difference between the 8n2 - 2n for n number of rows. For an example, consider
EMT and DP-EMT for unstable scenarios is not a concern a network of 50 busses. The EMT requires 44,850
as their initial transients are accurately captured. flops where as DP requires only 19,900 flops. The
computations at the interface nodes do not depend on
the size of the network and it depends only on number of
interface nodes.
The idea of using a buffer zone has been reported for
dynamic system equivalencing in EMT simulations [41],
[42]. The same idea can be extended for EMT-TS co-
simulation with DP as the surface layer of the interface.
For large networks, we recommend the use of DP as a
Figure 13: Line current of an unstable case for a three phase to ground surface layer that connects the TS simulation to the EMT
fault at the interface
model. This EMT-DP-TS co-simulation model will have a
7. Discussion computational burden similar to an EMT-TS co-simulation
model while avoiding the need for an FDNE.
7.1. Computational Efficiency and Recommendations
7.2. Limitations of the proposed model
An efficiency comparison with a DP simulation with
an EMT simulation is carried out in this section. In the The DP part of the co-simulation model is modeled using
EMT simulation, currents and voltages are computed as only positive sequence voltages and currents. Phase ‘’A”
three-phase quantities, whereas the proposed DP model to ground fault (fault resistance is 0.1Ω) is applied at t = 5s
uses single-phase quantities which therefore reduces the for a period of five cycles at the interface bus 38. Figure 14
computational burden. However, the phasors in the DP shows the phase “A” and “B” voltages at bus 38. The phase
model are complex signals unlike in the EMT simulation “A” line current from bus 36 to 34 is shown in Figure 15.
which only considers real values. Assume that in EMT As it can be seen from the figures the co-simulation results
simulation the inverse of the admittance matrix is computed closely follow EMT results. However, the accuracy of the
only once at the beginning (in some implementations model under an unbalnce fault is case dependent. Therefore
it is computed in every time step). Then the computing we are currently developing a method to handle unbalnaces
burden in real time simulation is predominantly for the in a DP simulation.
multiplication of a square matrix by a column vector. The
Floating Point Operations (flops) required to calculate the
network solution (V = Y-1I) for EMT and DP simulations
for a system with number of buses can be calculated as
follows:
For the EMT system, the size of the admittance matrix
is 3n×3n and the size of the current vector is 3n×1.
Matrix multiplication requires 9n2 multiplications and
(3n - 1)*3n additions giving 18n2 - 3n total flops. For
the DP network solution, the admittance matrix size
is n×n and the size of the current vector is n×1. Each
n sized row and vector multiplication results in n
number of complex number multiplications and n - 1
number of complex number additions. Multiplication
of two complex numbers requires four mutiplications
of real numbers and two additions of real numbers.
The addition of two complex numbers requires two
real number additions. Therefore, for one row there Figure 14: Voltages for a phase “a” to ground fault at bus 38

Cigre Science & Engineering • N°14 June 2019

93

Figure 15: Phase A current from bus 36 to 34 for a phase “a”
to ground fault at bus 38
8. Conclusions
A real-time EMT-DP co-simulation approach was
presented in this paper using the IEEE 68 bus test
system to demonstrate the performance of the proposed
algorithms. It was shown that unlike an EMT-TS co-
simulation, the proposed EMT-DP co-simulation does
not require an FDNE model at the interface bus. The
numerical stability of the simulation in the presence of
time-step delay was achieved using data extrapolation.
The synchronous generators in the DP side of the network
were modeled using detailed machine models allowing to
capture frequencies higher than electromechanical modes.
The efficiency of the co-simulation model was analyzed
in terms of floating point operations and it was shown that
EMT-DP model results in a considerable computational
gain compared to EMT. It is proposed to extend the EMT-
DP co-simulation presented in this paper to EMT-DP-TS
co-simulation.
Acknowledgment
The authors would like to thank the Natural Sciences
and Engineering Research Council (NSERC) and the
University of Manitoba for funding this research and
RTDS Technologies Inc. for the software support.
References
[1] V. Jalili-Marandi and V. Dinavahi and K. Strunz and J. A. Martinez
and A. Ramirez. Interfacing Techniques for Transient Stability and
Electromagnetic Transient Programs IEEE Task Force on Interfacing
Techniques for Simulation Tools. IEEE Transactions on Power
Delivery, 24(4):2385-2395, 2009.
[2] Y. Zhang and A. M. Gole and W. Wu and B. Zhang and H. Sun.
Development and Analysis of Applicability of a Hybrid Transient
Simulation Platform Combining TSA and EMT Elements. IEEE
Transactions on Power Systems, 28(1):357-366, 2013.
[3] Hongtian Su and Ka Wing Chan and L. A. Snider and Tak Shing Chung.
A parallel implementation of electromagnetic electromechanical
hybrid simulation protocol. 2004 IEEE International Conference on
Cigre Science & Engineering • N°14 June 2019
94
Electric Utility Deregulation, Restructuring and Power Technologies.
Proceedings, pages 151-155 Vol.1, 2004.
[4] D. Shu and X. Xie and Q. Jiang and Q. Huang and C. Zhang. A Novel
Interfacing Technique for Distributed Hybrid Simulations Combining
EMT and Transient Stability Models. IEEE Transactions on Power
Delivery, 33(1):130-140, 2018.
[5] U. D. Annakkage et al., “Dynamic System Equivalents: A Survey of
Available Techniques,” in IEEE Transactions on Power Delivery, vol.
27, no. 1, pp. 411-420, Jan. 2012.
[6] Y. Hu and W. Wu and A. M. Gole and B. Zhang. A Guaranteed
and Efficient Method to Enforce Passivity of Frequency-Dependent
Network Equivalents. IEEE Transactions on Power Systems,
32(3):2455-2463, 2017.
[7] Y. Hu and W. Wu and B. Zhang. A Semidefinite Programming
Model for Passivity Enforcement of Frequency-Dependent Network
Equivalents. IEEE Transactions on Power Delivery, 31(1):397-399,
2016.
[8] J. Morales-Rodriguez and J. Mahseredjian and K. Sheshyekani and
A. Ramirez and E. Medina and I. Kocar. Pole-Selective Residue
Perturbation Technique for Passivity Enforcement of FDNEs. IEEE
Transactions on Power Delivery, :1-1, 2018.
[9] A. Thakallapelli and S. Kamalasadan. Optimization based real-time
frequency dependent reduced order modeling of power grid. 2017
IEEE Power Energy Society General Meeting, pages 1-5, 2017.
[10] Y. Hu and W. Wu and B. Zhang. A Fast Method to Identify the Order
of Frequency-Dependent Network Equivalents. IEEE Transactions
on Power Systems, 31(1):54-62, 2016.
[11] Yuefeng Liang. An Improved Wide-Band System Equivalent
Technique for Real Time Digital Simulators. PhD dissertation,
University of Manitoba, Canada, 2011.
[12] S. Henschel. Analysis of electromagnetic and electromechanical
power system transients with dynamic phasors. PhD dissertation,
University of British Columbia, 1999.
[13] V. Venkatasubramanian and H. Schattler and J. Zaborszky. Fast
time-varying phasor analysis in the balanced three-phase large
electric power system. IEEE Transactions on Automatic Control,
40(11):1975-1982, 1995.
[14] P. Zhang and J. R. Marti and H. W. Dommel. Shifted-Frequency
Analysis for EMTP Simulation of Power-System Dynamics. IEEE
Transactions on Circuits and Systems I: Regular Papers, 57(9):2564-
2574, 2010.
[15] A. M. Stankovic and T. Aydin. Analysis of asymmetrical faults in
power systems using dynamic phasors. IEEE Transactions on Power
Systems, 15(3):1062-1068, 2000.
[16] T. Demiray and G. Andersson and L. Busarello. Evaluation study
for the simulation of power system transients using dynamic phasor
models. 2008 IEEE/PES Transmission and Distribution Conference
and Exposition: Latin America, pages 1-6, 2008.
[17] Matthew Alexander Kulasza. Generalized Dynamic Phasor-
Based Simulation for Power Systems . Master’s thesis, University of
Manitoba, Canada, 2014.
[18] Chandana Karawita. HVDC Interaction Studies Using Small Signal
Stability Assessment. PhD dissertation, University of Manitoba,
Canada, 2009.
[19] P. Zhang and J. R. Marti and H. W. Dommel. Synchronous Machine
Modeling Based on Shifted Frequency Analysis. IEEE Transactions
on Power Systems, 22(3):1139-1147, 2007.
[20] A. M. Stankovic and S. R. Sanders and T. Aydin. Dynamic phasors in
modeling and analysis of unbalanced polyphase AC machines. IEEE
Transactions on Energy Conversion, 17(1):107-113, 2002.
[21] B.C. Pal and B. Chaudhuri. Robust Control in Power Systems of New
York, Springer-Verlag. 2005.
[22] H. Konara and U. D. Annakkage. Evaluation of two positive sequence
filters for an interface between electromagnetic transient and phasor
solutions. 2015 IEEE 10th International Conference on Industrial
and Information Systems (ICIIS), pages 7-12, 2015.
[23] K. Mudunkotuwa and S. Filizadeh and U. Annakkage. Development
of a hybrid simulator by interfacing dynamic phasors with
electromagnetic transient simulation. IET Generation, Transmission
Distribution, 11(12):2991-3001, 2017.
[24] D. Shu and X. Xie and V. Dinavahi and C. Zhang and X. Ye and
Q. Jiang. Dynamic Phasor Based Interface Model for EMT and
Transient Stability Hybrid Simulations. IEEE Transactions on Power
Systems, :1-1, 2017.
[25] Harshani Konara. Interfacing dynamic phasor based system
equivalents to an electromagnetic transient simulation . Master’s
thesis, University of Manitoba, Canada, 2015.
[26] W. Ren, M. Steurer and T. L. Baldwin, “Improve the Stability and the
Accuracy of Power Hardware-in-the-Loop Simulation by Selecting
Appropriate Interface Algorithms,” in IEEE Transactions on Industry
Applications, vol. 44, no. 4, pp. 1286-1294, July-aug. 2008.
[27] Oh, Seaseung and Chae, Suyong. A Co-Simulation Framework for
Power System Analysis.
[28] F. Plumier, P. Aristidou, C. Geuzaine and T. Van Cutsem, “Co-
Simulation of Electromagnetic Transients and Phasor Models:
A Relaxation Approach,” in  IEEE Transactions on Power
Delivery, vol. 31, no. 5, pp. 2360-2369, Oct. 2016. doi: 10.1109/
TPWRD.2016.2537927
[29] V. Venkatasubramanian. Tools for dynamic analysis of the general
large power system using time-varying phasors. International Journal
of Electrical Power Energy Systems, 16(6):365 - 376, 1994.
[30] A. M. Gole and R. W. Menzies and H. M. Turanli and D. A.
Woodford. Improved Interfacing of Electrical Machine Models to
Electromagnetic Transients Programs. IEEE Power Engineering
Review, PER-4(9):56-57, 1984.
[31] Harshani Konara and U.D. Annakkage and C. Karawita. Stability
Criterion for Interfacing a Transient Stability model to a Dynamic
Phasor model. IEEE PES TD Conference and Exposition, 2018.
[32] Vladimir Brandwajn. Synchronous generator models for the analysis
of electromagnetic transients. PhD dissertation, University of British
Columbia, Canada, 1977.
[33] H. K. Lauw and W. S. Meyer. Universal Machine Modeling for the
Representation of Rotating Electric Machinery in an Electromagnetic
Transients Program. IEEE Transactions on Power Apparatus and
Systems, PAS-101(6):1342-1351, 1982.
[34] P. Subramaniam and O. P. Malik, “Digital simulation of a
synchronous generator in direct-phase quantities,” in  Electrical
Engineers, Proceedings of the Institution of, vol. 118, no. 1, pp. 153-
160, January 1971.
[35] M. Rafian and M. A. Laughton, “Determination of synchronous-
machine phase-co-ordinate parameters,” in  Electrical Engineers,
Proceedings of the Institution of, vol. 123, no. 8, pp. 818-824, August
1976.
[36] L. Wang and J. Jatskevich, “A Voltage-Behind-Reactance Synchronous
Machine Model for the EMTP-Type Solution,”  2007 IEEE Power
Engineering Society General Meeting, Tampa, FL, 2007, pp. 1-1.
Cigre Science & Engineering • N°14 June 2019
95
[37] Xuegong Wang and Wilson, P. and Woodford, D. Interfacing transient
stability program to EMTDC program. Power System Technology,
2002. Proceedings. PowerCon 2002. International Conference on,
pages 1264-1269 vol.2, 2002.
[38] H. Su, K. W. Chan, L. A. Snider and J-C. Soumagne. Advancements
on the Integration of Electromagnetic Transients Simulator and
Transient Stability Simulator. International Conference on Power
Systems Transients Montreal, 2005.
[39] Xi Lin. System Equivalent for Real Time Digital Simulator. PhD
dissertation, University of Manitoba, Canada, 2010.
[40] M. Szechtman, T. Wess and C. V. Thio, “A benchmark model for
HVDC system studies,”  International Conference on AC and DC
Power Transmission, London, UK, 1991, pp. 374-378.
[41] X. Nie, Y. Chen and V. Dinavahi, “Real-Time Transient Simulation
Based on a Robust Two-Layer Network Equivalent,” in  IEEE
Transactions on Power Systems, vol. 22, no. 4, pp. 1771-1781, Nov.
2007.
[42] M. Abdel-Rahman, A. Semlyen and M. Reza Iravani, “Two-
layer network equivalent for electromagnetic transients,” in  IEEE
Transactions on Power Delivery, vol. 18, no. 4, pp. 1328-1335, Oct.
2003.
Biographies
Harshani Konara received the B.Sc. (Eng.) degree from
the University of Peradeniya, Peradeniya, Sri Lanka, in
2012 and M.Sc. degree from the University of Manitoba,
Winnipeg, MB, Canada, in 2016. She is currently working
toward the Ph.D. degree at the University of Manitoba,
Winnipeg, MB, Canada. Her research interests include
power system dynamic modelling and simulation.
Udaya D. Annakkage received the B.Sc. (Eng.) degree
from the University of Moratuwa, Moratuwa, Sri
Lanka, in 1982 and the M.Sc. and Ph.D. degrees from
the University of Manchester Institute of Science and
Technology (UMIST), Manchester, U.K., in 1984 and
1987, respectively. He is presently a Professor at the
University of Manitoba, Winnipeg, Manitoba, Canada.
His research interests include power system stability and
control, security assessment and control, operation of
restructured power systems, and power system simulation.
Chandana Karawita received the B.Sc. (Eng) degree
from the University of Moratuwa, Moratuwa, Sri Lanka,
in 2002 and the M.Sc. and Ph.D. degrees from the
University of Manitoba, Winnipeg, MB, Canada, in 2006
and 2009, respectively. He joined TransGrid Solutions,
Canada, in 2007 and is now serving as a senior electrical
engineer. He has specialized in power systems planning
studies and was instrumental in developing a software
package to analyse sub-synchronous interactions. He is
still involved in academic research as an adjunct professor
at the University of Manitoba.
IDtools: An automated tool for modal
identification from time-domain
simulation results for establishing
system operating limits
S. D. RAO*, J. J. SANCHEZ-GASCA, GE Energy Consulting
B. THOMAS, PowerGEM LLC
USA

Abstract disturbance measured data, thus making the computation

There is an increasing expectation to ensure that
electromechanical oscillations in power systems have a
positive damping ratio above a certain threshold while
establishing the system operating limits (SOLs). The
work reported in this paper aims at achieving complete
automation of this assessment using modal identification.
The Eigensystem Realization Algorithm (ERA) is applied
to compute the modes present in a given signal(s), and the
corresponding damping ratios are computed. Traditionally
the application of modal identification methods to power
systems relies on user interaction to estimate the number of
oscillatory modes to be computed. In this paper, a metric
is computed for each mode to determine its contribution
to the signal, used to eliminate insignificant or spurious
modes. This screening threshold has been observed to
work reliably in approximately 98% of the cases.
1. Introduction
Ensuring that the power system oscillations are well-
damped is one of the important considerations of power
system planners and operators while establishing the
system operating limits (SOLs) and interconnection
reliability operating limits (IROLs). One way to ascertain
whether a system is well-damped is to linearize the
underlying power system dynamic models and perform
a small signal stability analysis to identify the dominant
modes of the system and compute their damping ratios.
However, the computational expense of this approach
can be prohibitive for large systems. An alternative
approach is to use time-domain signals obtained from
either the results of transient simulations or from post-
of system damping independent of the system size. This
paper focuses on estimating the damping from the time-
domain signals obtained as a result of transient stability
studies. Since these studies are routinely performed for
ascertaining acceptable system dynamic performance
and typically involve the execution of a large number of
time domain simulations; techniques that facilitate this
process are needed today. IDtools, a modal identification
tool was developed with this need in mind and is aimed at
expediting the assessment of system damping.
Different methods have been introduced for computing
system damping from time-domain signals following
a system disturbance. A simple approach consists of
computing the reduction in the magnitude of oscillations
present in the signal over a given time period. A 53%
reduction in the magnitude of oscillation amplitude
over four periods of oscillation is typically considered
acceptable, which translates to a 3% damping ratio. This
approach yields reliable results if the signal has only one
oscillatory mode. However, power system oscillations
often include multiple oscillatory modes, such as inter-
area modes (typically in the range of 0.1-0.9 Hz) and local
modes (typically in the range of 1 to 3 Hz). For accurate
analysis of such signals with multiple modes, more
sophisticated modal identification methods have been
developed, e.g., Prony, matrix-pencil, ERA, etc. [1], [7],
[9], [10]. The Eigensystem Realization Algorithm (ERA)
[2] – [6] is the method used in the work reported here.
The method is based on the singular value decomposition
(SVD) of a matrix easily built from transient stability
simulation results. The method is numerically stable and

* shrutidwarkanath.rao@ge.com

KEYWORDS
Damping ratio, Eigensystem Realization Algorithm (ERA), electromechanical oscillations, modal identification, Time-
domain simulation

Cigre Science & Engineering • N°14 June 2019

96
has been used in a variety of applications. A summary
of the ERA is provided in the appendix. While several
publications have explored different aspects of the ERA
algorithm, of which [2] – [6] are a few, all such tools require
a higher level of user involvement to identify the true
modes. It will be shown that for nonlinear systems such
as power systems, the demarcation between significant and
insignificant/spurious modes, using either the magnitudes
of the singular values or FFT results, is not always clear.
This is because such linearization techniques can only
approximate the underlying nonlinear system with a
linear system. This paper advances the work done in [3],
by computing a metric for each mode whose magnitude
indicates the contribution of the given mode to the signal.
The metric is used to detect the true and significant modes
by comparing its value against a threshold. Results using
Western Electricity Coordinating Council (WECC) cases
representing the western interconnection in United States
are provided that validate the proposed approach and
illustrate the application of the program; guidelines for the
use of such a modal identification tool are provided as well.
The application of the program is straightforward; a
user only needs to specify the file with the time-domain
simulation results and the list of signals that are to be
processed. It then creates a summary report with a list of
signals that failed to meet the minimum damping ratio
requirement and a report with the details of modes present
in each individual signal.
While in principle the ERA can be used to process a
large number of signals obtained from transient stability
simulations, the size of the matrix required by the
algorithm would become too large and, perhaps more
importantly, the task of correlating modes with individual
signals would become too burdensome. Hence in the work
reported in this paper, each signal was processed separately
to identify the modes in it. Since the computations for
different signals are independent of each other, parallel
processing can be used to increase the execution speed.
2. Pre-processing of signals in the
program
Before performing the modal identification, the initial
value of the signal is removed, then the signal is normalized
by dividing each data point by the magnitude of the
largest data point. For example, if a signal has an initial
Cigre Science & Engineering • N°14 June 2019
97
value of 100, and a maximum absolute value of 105.7,
100 (initial value) will be subtracted from all data points
in the signal and then all data points will be divided by
5.7. If the data point with the largest absolute value (after
removing initial value) is smaller than 10-16, normalizing is
not done. These operations improve the condition number
of the Hankel matrix while still retaining the necessary
information about the signal shape and hence its modes.
Typically, the signals from time-domain simulations
are sampled every quarter cycle (0.004s). However,
identification of modes in the range 0.1 to 3Hz, does not
require data with such granularity. Hence, internally the
program further samples the signal. The required sampling
rate for modal identification is automatically determined,
to avoid picking too many data points.
Additionally, checks are performed internally to ensure
that the tool does not process signals with very small
variations. Such signals are not suitable for modal
identification as they are often either linear (and hence
have no oscillatory modes) or are very noisy. The noise
or the “chatteriness” of such signals is often due to the
convergence tolerance used in the simulations, since
all digits beyond a desired tolerance have no physical
meaning.
3. Impact of inclusion of spurious
modes
For truly linear systems, the magnitude of the singular
values obtained by ERA or the results of an FFT of a signal
are often good indicators of the contribution of a mode
to a signal. As an example, the FFT outputs of two linear
systems are shown in Figure 1, where clear peaks are seen.
The plot on the left is the FFT of a system with a single
mode at 2.0 Hz, while the plot on the right is for a system
with two modes at 1.0 Hz and 1.5 Hz.
However, since power systems are inherently nonlinear, the
FFT results or singular value magnitudes of the linearized
approximations are not always clear indicators of the true
modes which can be seen from Figure 2, where no clear
peaks are displayed. Hence, in a case like this, it is difficult
to differentiate between the true dominant modes of the
system and spurious/insignificant modes. In some such
cases, the spurious modes could be poorly damped, while
the dominant modes are well-damped. In that case, a signal
 Figure 1. FFT outputs for purely linear systems

Figure 2. Sample FFT results for power systems

could be incorrectly flagged as being poorly damped. Such
signals where FFT results, singular values etc. are not clear
indicators of the true modes of the system, often do not
exhibit a clear oscillatory behaviour. This will be discussed
in detail in latter sections of the paper.
A good measure of the validity of the identified modes
is whether a signal constructed using the identified
modes approximates the signals from which the modes
are derived. Examples of such comparison are shown
in Figure 5, where the blue curves represent the impulse
response of the linear system while the black curves are
the true signals used for modal identification. Based on
the excellent match between the two, it can be concluded
that the modes identified for these two signals are very
accurate.
Spurious modes do not contribute much to the signal
shape. An example of this is shown in Figure 3, where the
plot on the left has the signal reconstructed from a system
with seven eigenvalues while the signal on the right has the
same signal reconstructed from five eigenvalues. It can be
seen that the spurious mode i.e. the last pair of eigenvalues
does not contribute much to the shape. In some rare cases,
the inclusion of a spurious mode can make the match
between the actual signal and reconstructed signal worse.
This is shown in Figure 4 where the plot on the left has a
match with 13 modes, while the plot on the right has the

Figure 3 Example of lack of contribution from spurious modes

Cigre Science & Engineering • N°14 June 2019

98
 Figure 4 Example of detrimental effect of spurious modes

results with 17 modes. It is clearly seen that the inclusion (4)

of the additional modes worsens the match between the
actual signal and reconstructed signal.
4. Modal screening metric and
numerical results
In order to differentiate between dominant modes and
spurious/insignificant modes, a metric is computed using
the residues for each mode. The left and right eigen vectors
of the state matrix obtained as a by-product of the eigen
analysis done in ERA, are used for this. Assuming the right
eigenvector matrix of Aδ is given by V and left eigen vector
matrix is given by W, two matrices are computed as:
(1)
(2)
Note that for a single input, single output system, B is a
column vector while C is a row vector, hence Bjord and
Cjord are also vectors. For the ith mode, the magnitude of
the residue is given by
(3)
The weight of each mode with eigen value λi is computed
from the residues as:
The modes are ranked based on the weight for each mode.
The relative weight of each mode is computed as:
(5)
Modes with significance less than 0.001 are ignored.
This threshold was determined based on the results for
several signals of different types for two WECC cases for
different disturbance simulations.
In most cases, up to 7 eigenvalues (i.e. up to 3
oscillatory modes) suffice to replicate the behavior of
a signal. This is illustrated in Figure 5 which shows
six examples of a good match between actual signals
and signals constructed using identified modes. In this
figure, the actual signal (after removing initial value
and normalizing) is plotted in black while the signal
generated from the modes is plotted in blue. For most
signals, the match was observed to be fairly accurate
as shown below. However, in a few cases, the shape of
the signal constructed from the modes did not match the
true signal too well. However, such signals typically had
very small variations and hence the difference between
the actual quantity and the reconstructed quantity was
fairly low despite the poor match in shape.

Figure 5 Examples of cases where the signals constructed using the modes match the true signal accurately

Cigre Science & Engineering • N°14 June 2019

99
In order to validate the threshold, two WECC cases were
used to run 10 simulations, using different disturbances
at different locations in the system. For each case, the
signals to be assessed were chosen by picking three to
10 signals from each area that had the worst spread. The
statistics are provided in Table 1. A total of 968 signals
were inspected of different types (voltage magnitude,
angle, active and reactive power, field voltage,, speed
etc.), of which 70 signals were poorly identified despite
using the above-described metric. The poorly identified
signals either contained spurious modes selected, or did
not provide an accurate match between the true signal and
the reconstructed signal. Of these 70 signals, 58 signals
fell in one of the following categories:
1. Variations in the signal are too small, with the relative
spread ((Max_value – Min_value)*100/Initial_value)
being less than 2% of the initial value.
2. Signal is too noisy/non-linear to be used for modal
identification.
3. The magnitude of the original signal is too small.
For signals that are small in magnitude, the probability
of falsely flagging a signal as poorly damped due to a
spurious mode was found to be higher. From a system
stability perspective, the output of larger generating units
and branches with heavier flows are typically more critical
and are expected to be assessed using this tool, along with
Table 1. Summary of the performance of the screening metric
p.u. terminal voltages, currents and speeds. Of the larger
signals looked at (output of the larger generating units
and branches with heavier flows, along with p.u. terminal
voltages, currents and speeds), less than 2% of the signals
that were flagged as being poorly damped were due to
spurious modes.
The quality of the match can be determined by looking at the
RMS error between the actual signal and the reconstructed
signal. For a few cases, the signal shape is not captured
very accurately by modal identification as shown in Figure
6. However, note that in such cases, it was observed that
the signal variations were typically so small, that the
difference between the actual electromechanical signal and
the re-constructed signal was still quite low. Note that in
the example shown in Figure 6, the relative spread of the
signals over the time-frame chosen is less than 2%. In such
cases, more than 7 eigen values are needed to accurately
identify the system. For the example shown in Figure 6, a
good match can be obtained with 17 eigen values as shown
in Figure 7. It has been observed that if the time frame for
modal identification is chosen to be smaller, the accuracy
of the match improves. Moreover, the signals with higher
errors are observed to be signals that are too nonlinear; or
are signals that are very small in magnitude and are not
expected to be critical from a system stability perspective
(for example reactive power/flow of <5 MVAr).

WECC Disturbance Time- Total number A: Number of B: Number of Comments regarding poorly identified
case frame of of signals signals where signals where cases (both categories A and B)
study studied spurious mode chosen number
was selected modes are
insufficient for
accurate match
Case1 3ph bus fault 8-10s 220 0 1 Spread less than 0.1 MVAr
Case1 Step voltage 8-10s 220 2 2 Very flat/small
reference of a large
nuclear generator
Case1 3ph bus fault 5-10s 66 7 2 Very flat/small
Case2 Trip large 6-15s 66 13 11 11/13 signals with spurious mode:
generator signal very small. 8/11 with poor match
are signals that are very small
Case2 Trip large generator 10-15s 66 8 0 Very flat/small
Case2 Trip load 6-10s 66 9 5 All too “chattery”/nonlinear with low-
magnitude variations
Case2 3ph linear fault 6-10s 66 5 0 2 signals have significance metric close
to threshold, all signals less than 10
MW/MVArs
Case2 LLG bus fault 6-10s 66 2 0
Case2 LG line fault 6-10s 66 1 1
Case2 3ph bus fault 10-13s 66 1 0 Signal almost linear, no oscillations

Cigre Science & Engineering • N°14 June 2019

100
 Figure 6 Example of case where the signal shape is not captured too accurately by modal identification

Figure 7 Example where higher number of modes are needed for accurate modal identification

5. Recommendations for use of flagged as being poorly damped due to spurious or

automated modal identification
tools
Modal identification tools are meant to compute lightly
damped electromechanical modes from one or several
signals within a given time frame. Examples of such
signals are shown in Figures 5 and 7. These signals clearly
show an oscillatory behaviour in the frequency range
typically associated with electromechanical oscillations
~0.1-3 Hz.
Such algorithms should not be used to process signals
where there are no discernible electromechanical
oscillations in the time frame selected for modal
identification. Based on the authors’ experience, it is
recommended that this program not be used for signals
that have less than 2% relative spread within the time
frame used for modal identification as they may be
insignificant modes. In some cases, especially if the
time frame used for the modal identification is towards
the end of the simulation, the signals do not have any
oscillatory modes present (as the dominant oscillatory
modes may have already died down by then). For
example, see Figure 8 (the left image shows the signal
over the entire simulation duration while the right image
shows the zoomed in image focusing on the last 2
seconds of simulation); the plot of the signal in the last
two seconds of the simulation show what is essentially
a straight line from which it is not possible to extract
oscillatory modes. Note that the range of variations for
the signal (bus voltage angle) is less than 0.001 degrees
which is essentially constant from a practical standpoint.
The program assumes that the signals have at least one
oscillatory mode and hence, if such near-linear curves
are processed, it will inevitably compute some spurious
oscillatory modes that may be poorly damped.

Figure 8 Signal with very small variations

Cigre Science & Engineering • N°14 June 2019

101
 Figure 9 Signal with small variations and chattery behaviour

Figure 10 Example of a case where the signal exhibits abrupt changes

Other problematic situation occurs when the variations are
so small that the signal appears to be noisy, for example
see Figure 9 (The left image shows the signal over the
entire simulation duration while the zoomed in image
focusing on the last 2 seconds of simulation is shown on
the right, variations in the last 2 seconds are of the order of
0.0001 MVAr). The program should not be used to process
signals that are noisy and/or exhibit “chatter”, as the
signal constructed from the modes will not match the true
signal well due to the presence of sudden variations in the
signal (which are physically meaningless). Such signals
can be detected by metrics such as signal to noise ratio.
The program should not be used for signals that exhibit
discrete and/or abrupt changes such as those encircled in
Figure 10, since the signal constructed from the identified
modes cannot capture the abrupt changes in the signal.
This is an indication that the identified modes might be
in error and the RMS errors between the actual signal and
the reconstructed signal will consequently be higher.
6. Conclusions
This work is the latest effort toward merging modal
damping identification from time domain simulations
in a commercial product. Using the described screening
approach, the automated modal identification tool
described in the paper, IDtools has been integrated in a
transient stability simulation program. For each signal
being processed, it identifies up to three oscillatory
modes, out of a total of 7 eigenvalues (including purely
real eigenvalues as well as complex-conjugate pairs). If
appropriate signals with sufficient signal variations were
used, it appropriately identified approximately 98% of
nearly 900 signals inspected. With a C implementation, it
takes on average 0.12 s to process each signal, with a major
portion of the time spent in reading and pre-processing the
data. Since all signals are processed independent of each
other, the process can easily be parallelized for speed.
This is the first step towards a complete automation of
modal identification from time-domain signals. These
computations are strongly recommended to be done after
all automatic discrete adjustments in the system have
already been made following a fault clearance or other
major switching event, in order to ensure that the signals
do not have any sharp changes. The process of filtering
out unnecessary modes is a key step which will continue
to be refined. While the screening is not 100% accurate,
users can look at the predominant modes for several
signals in a given area since they are expected to have the

Cigre Science & Engineering • N°14 June 2019

102
same true modes. One future approach to further refine
the screening process, could be to check for the consistent
presence of a given mode as the linearized system size is
increased [3].
7. Appendix A: Summary of ERA
algorithm
The first step of ERA involves obtaining a discrete system
is such that the input signal is its free-state impulse
response (i.e. system output when all initial states are at
zero and an impulse input is applied to the system):
(6)
(7)
Given that for zero-state impulse response, X[0] = 0, one
can easily obtain that:
(8)
(9)
(10 )
ERA uses (10) to obtain the A,B,C matrices from the
sampled data points [3] – [6]. The first step in ERA
algorithm involves constructing two Hankel Matrices H0
and H1 from the data points in the signal.
( 11 )
( 12 )
The dimensions of the above matrices are no x ni where
no is number of outputs (i.e. number of modes to be
computed) and ni is the number of inputs (i.e. number of
signals being processed). The most important part of ERA
involves performing an SVD of H0.
( 13 )
It has been shown that the state matrix A, input matrix B
and output matrix C for this linear system are given by
(14) – (16) [4].
( 14 )
Cigre Science & Engineering • N°14 June 2019
103
( 15 )
( 16 )
The discrete system can then be converted to a continuous
system as:
( 17 )
( 18 )
( 19 )
Once the state matrix of the continuous system is
obtained, its eigen values λ, are the modes of the system.
The damping ratio of each mode can be computed as:
( 20 )
Since each signal is processed separately in the program,
the value of ni is 1. For each signal, a single-input, single-
output system is identified, with the input being an
impulse and the output being the original signal.
8. References
[1] IEEE Task Force on Identification of Electromechanical Modes,
“Identification of Electromechanical Modes in Power Systems”,
Special Publication TP462, June 2012.
[2] I. Kamwa, R. Grondin, J. Dickinson, S. Fortin, “A Minimal
Realization Approach to Reduced-Order Modelling and Modal
Analysis for Power System Response Signals”, IEEE Transactions on
Power Systems, Vol. 8, No. 3, Aug. 1993, pp. 1020-1029.
[3] J. J. Sanchez-Gasca, “Toward the automated computation of
electromechanical modes from transient simulations,” IEEE PES
General Meeting, Minneapolis, MN, 2010, pp. 1-7.
[4] J. Juang, R.S. Pappa, “An Eigensystem Realization Algorithm
for Modal Parameter Identification and Model Reduction,” J. of
Guidance and Control, Vol. 8, No. 5, Sept.- Oct. 1985, pp. 620-627.
[5] J. Juang, Applied System Identification, Prentice Hall, 1994.
[6] J. J. Sanchez-Gasca and J. H. Chow, «Computation of power system
low-order models from time domain simulations using a Hankel
matrix,» IEEE Transactions on Power Systems, vol. 12, no. 4, pp.
1461-1467, Nov 1997.
[7] J. F. Hauer, C. J. Demeure and L. L. Scharf, «Initial results in Prony
analysis of power system response signals,» in IEEE Transactions on
Power Systems, vol. 5, no. 1, pp. 80-89, Feb. 1990.
[8] L. L. Grant and M. L. Crow, «Comparison of Matrix Pencil and
Prony methods for power system modal analysis of noisy signals,»
2011 North American Power Symposium, Boston, MA, 2011, pp. 1-7.
[9] Hua Y., Sarkar T., “Generalized Pencil-of-function method for
extracting poles of an EM system from its transient response,” IEEE
Transactions on Antennas & Propagation, Vol. 37, No. 2, February
1989.
[10] T. K. Sarkar and O. Pereira, «Using the matrix pencil method to
estimate the parameters of a sum of complex exponentials,» in IEEE
Antennas and Propagation Magazine, vol. 37, no. 1, pp. 48-55, Feb.
1995.
9. Biographies
Shruti Rao is a Senior Engineer at GE’s Energy Consulting
Department in Schenectady, NY, where she is involved
in developing and improving GE-PSLF, conducting
dynamics studies, as well as modeling. She holds a PhD
in Electrical Engineering from Arizona State University
(ASU). Her research projects at ASU involved voltage
stability analysis and network equivalencing using the
holomorphic embedding method, as well as determining
the maximum load that a transformer can sustain using
thermal models.
Juan J. Sanchez-Gasca is a Technical Director at GE’s
Energy Consulting Department in Schenectady, NY,
where he has been involved in the study of power system
Cigre Science & Engineering • N°14 June 2019
104
dynamics and in the development of dynamic models for
transient stability analyses. He holds a PhD in Electrical
Engineering from the University of Wisconsin-Madison.
He is an IEEE Fellow and a past Chair of the IEEE PES
Power System Dynamics Performance Committee.
Brian Thomas received his Master's degree in Electrical
Engineering from Illinois Institute of Technology in
2006. Mr. Thomas is Lead Consultant at PowerGEM,
where he works on the development of market simulation
tool. Mr. Thomas was Technical Solutions Director and
Commercial Software Manager in Energy Consulting
at General Electric and had been involved with the
development of GE PSLF and performing various
consulting studies related to transmission planning and
model development.
A direct calculation of locational marginal
value of distributed energy resources
M. NDRIO, NAYEEM M. ABDULLAH*, M. MAIGHA, A. PAASO, Commonwealth Edison (ComEd), USA
A. KHODAEI, University of Denver, USA
E. NTAKOU, F. FARZAN, Quanta Technology, USA
Summary
There is a growing world-wide interest in the integration
of distributed energy resources (DERs) into electric power
grids due to their potential economic and environmental
benefits. Broadly defined, a DER represents any electric
power resource, such as solar or combined heat and power,
installed and operated in the distribution system at voltage
levels below the typical bulk power system levels. The
increased DER installations provide an opportunity for
distribution utilities to view such resources as non-wire
alternatives that enable network investment deferrals.
However, this brings considerable challenges in the
development of methodologies and evaluation algorithms
to determine the value of DERs to the distribution grid.
In this paper, we provide mathematical formulation for
the direct calculation of the value of DERs based on the
AC power flow equations. The mathematical framework
is incorporated into a generalized algorithm for the
assessment of the locational and temporal value of a
generic DER to the distribution grid. Preliminary study
results are provided to demonstrate the locational nature
of the value of DERs and the impact of the magnitude of
violations and network losses in the value. The proposed
methodology can be applied to any planning horizon and
enables the consideration of the DER value in terms of
managing voltage levels, circuit power factor, as well as
real and reactive power.
1. Introduction
The electric power industry is entering a period of
fundamental transition due to growing climate change
concerns, the integration of intermittent renewable
*Nayeem.MohammadAbdullah@ComEd.com
KEYWORDS
Distributed energy resources (DERs), locational marginal value (LMV), programming tools, power system
simulations
Cigre Science & Engineering • N°14 June 2019
105
generation and the emergence of distributed resources
that offer the potential for additional resilience. In this
changing environment, electricity customers have more
choices and control over their electricity service, such as
ownership of generation systems (solar PV, combined heat
and power), leveraging demand response opportunities,
as well as, energy efficiency measures. Collectively,
these load-altering measures are referred to as distributed
energy resources or DERs. For distribution grid operators,
DERs can potentially provide value to the grid in terms of
deferred capacity upgrades or reduced operating expenses.
DERs are mostly located near end-use customers and
can incorporate renewable generation of intermittent
nature that can provide significant environmental benefits
resulting from their operation and aggregation. DERs
can bring a wide range of values to the electricity grid
[1]. From the increase in usage of low-carbon energy,
reduction in the customer’s electricity payments to
improvement of power quality, reliability and resilience.
The focus of this paper is only on the value that DERs
bring to the distribution grid.
This value defined in terms of the avoided cost for
providing delivery capacity, which is primarily peak load
shaving or other loading reduction on parts of the system
that would otherwise require investment to increase
capacity.
DERs can provide voltage control and power quality
benefits in terms of avoided costs to maintain voltages
within ANSI or state specific standards. Furthermore,
DERs may conceivably in the future provide reliability and
resilience benefits via not only behind-the-meter backup
 Figure 1. Illustration of DER Overload Mitigation.
generation but also via local or community microgrids
that provide, in effect, a form of backup generation to
one or multiple customers. DERs can provide these
benefits via the supply of real power, reactive power, and
reserve capacity. Once DERs become a significant part of
the distribution system portfolio, some DERs may also
provide a reliability benefit in terms of local reserves; that
is, additional real and reactive power capacity available to
provide capacity, voltage, or reliability in case the DERs
primarily providing that service become available.
DERs pose opportunities as well as challenges.
Unmanaged DERs are often a source of power quality
problems, especially high voltage and flicker if power
electronically-coupled, and possibly circuit ampere
overloads in cases of extreme reverse flow [2], [3]. DERs
can, however, be a means to mitigate these problems, even
increasing hosting capacity for other DERs, including EV
charging by providing local real and reactive power to
mitigate the adverse effects of locally high amounts of PV
or other variable DG.
We first discuss the fundamental aspects of the value of
DER and the nature of the problem focusing on distribution
avoided cost. The key contribution of this work is to
provide a generic algorithm for direct calculation of the
value of DER that is technology agnostic and is based
on the AC power flow formulation. We present the
preliminary results of two test systems.
2. Value of DER Fundamentals
The first principle in determining the value of DER to the
distribution grid in terms of avoided costs is that all DER
value is locational [4]. That is because all avoided costs
for capacity and voltage are investment costs in response
to actual forecasted needs of the system, which are always
based on specific grid issues in specific locations. On a
given circuit, the ability of a DER to avoid capacity or
voltage costs by providing real or reactive power depends
very much upon the location relative to the grid constraint
Cigre Science & Engineering • N°14 June 2019
106
to be addressed. Generally, for radial distribution circuits
the DER must be downstream of the location of a loading
issue to help mitigate the problem. Figure 1 illustrates
the DER overload mitigation concept through a simple
example. The DER downstream of the congested line
provides more benefits in comparison to the DER
upstream of the congestion. Similarly, voltage problems
on circuits are typically local in nature. Low voltage at
the end of the circuit does not imply low voltage near or
at the station. Indeed, attempting to cure a low voltage at
one location by raising overall circuit voltages may cause
too high a voltage in another location on the same circuit.
Traditional mitigation to voltage issues are to deploy local
apparatus that affects voltage locally, such as capacitors or
voltage regulators.
If we examine the simple case of a radial feeder shown
above where the first mile of a circuit needs to be reinforced
and the argument that DER only affects loadings on the
upstream portions of the circuit, then this introduces the
concept of how to place a locational value on the DER. If
the DER midway on the overloaded portion of the circuit
deferred the need to upgrade capacity on the first half
of the circuit – conceivably it could at maximum avoid
half of the cost. The DER located at the downstream end
of the affected portion of the circuit could, by contrast,
avoid all the cost. So conceptually from a distribution grid
perspective, the second DER offers twice the value of the
first. This concept can be generalized. For any DER, its
locational value is affected by the sensitivity of the circuit
condition requiring capacity or voltage investments to the
real and reactive power from that DER.
Another concept is important when we evaluate the value
to the distribution grid of specific DER technologies.
Circuit capacity and voltage problems typically do not
happen to the same extent across every hour of the year,
and in fact may only occur for a limited number of hours
a year. An effective and efficient DER technology would
essentially provide real and reactive power when and
where they are needed. As such, the DER value is both
locationally and temporally granular. Therefore, there are
three elements to valuing DER: First is the avoided cost of
required circuit upgrades and how that cost is apportioned
among affected parts of the circuit; and second is how
effective real and reactive power input from the DER is
at a given location in deferring that cost. Third is how
different DER technologies align with the temporal value
of DER and how much of the generic DER value a given
technology can realize.
An accurate and efficient methodology for valuing DERs
would compensate such resources for services they
provide to the distribution grid by addressing different
characteristics /capabilities of the DER technologies
as well as taking into consideration their locational and
temporal value. Moreover, such methodology would be
fair and equitable to limit impact on non-participating
customers and avoids under and over-compensation.
The double compensation of DERs from other existing
markets would be avoided. The methodology would
support the value provided by the DERs to the grid based
on when and where they are functional.
3. Direct Calculation of the Loca-
tional Marginal Value of DERs
This study extends the work provided in [5] and presents
a methodology to directly compute the locational
marginal value (LMV) of a generic DER. The concept of
marginal cost of capacity (MCC) was first introduced in
[5] and is defined as the marginal cost of grid upgrades
required to mitigate overloads and voltage violations in
the network. To calculate the MCC, the branch ampere
overloads are obtained for each hour of the specified
study period (8760 for a year) from the results of a series
of unconstrained distribution load flows. The MCC
values are used to establish the penalty functions for the
ampere overloads in each branch of the study system.
The LMV evaluation presented in [5]-[6] requires the
solution of an optimal power flow problem with the
objective to minimize the total cost of supplying real and
reactive power at the substation and the total penalty costs
of overloads or voltage violations. The dual variables
(Lagrange multipliers) of the real and reactive power
equality constraints at each node are the LMVs since they
represent the marginal impact on the total MCC of an
Cigre Science & Engineering • N°14 June 2019
107
increment of real and reactive power at that node. In this
work, however, the mathematical definition of Lagrange
multiplier is used as follows:
(1)
where MCCT is the total penalty function summed across
all branches and MCC(j) is the penalty for branch j. A
similar equation can be written for the reactive power
LMV. At this step in the formulation, the form of the
MCC(j) is very general. The MCC may be formulated
to explicitly represent the branch current overload, nodal
voltage, branch real/reactive power, or any electrical
variable in the network. In this paper, the LMV is derived
for MCCs representing branch current overloads. In fact,
(2)
where Imag(j) is the current magnitude flowing on branch
j, Ilim denotes the ampacity limit of the branch and MCCpu
is the total cost of the grid upgrade required to mitigate
the branch overload divided by the total number of hours
in the study period. Using the AC power flow equations
and the Jacobian matrix terms as well as the phasor
expressions for branch current, nodal voltages and branch
admittance, we can rewrite the LMVp(i) as follows:
(3)
IOT is a vector of binary values where it is 1 for a violated
branch and 0 otherwise.
Figure 3. The case study system (load in p.u.).
4. Generalized Algorithm for the
Direct LMV Computation
The generalized procedure for the calculation of the LMV
for generic DERs is shown in Figure 2. This procedure
is applied to each hour of the planning horizon. The first
 Figure 2. Computational procedure of the LMV calculation for every (hourly) simulation.
step in the process is to implement a “reduction” in the
specified distribution network considered in the study.
The reason for this pre-step is that the proposed method
is designed for balanced, single-phase networks and as
such, elimination of two-phase or three-phase elements
is required for the appropriate power flow solution. The
“network reduction” step also aims to simplify the network
and remove redundant elements and loops to expedite the
computation procedure. Furthermore, the removal of very
small element impedances is required to avoid numerical
issues in the computation of the admittance matrix and
the Jacobian elements. The “network reduction” step is
particularly useful when dealing with large-scale realistic
distribution feeder topologies.
Step 2 of the process is to specify the real and reactive
nodal demand in the network to solve the AC power flow.
This is accomplished by first computing the Jacobian
matrix around an initial state of the system. The Newton-
Raphson method is implemented for the AC power flow
solution at each iteration and the desired solution is derived
when the mismatch error is below a certain threshold. The
solution of the power flow are the state variable, i.e., the
nodal voltage magnitudes and angles are used. In Step 5,
the power flow solution is compared to current and voltage
limits to identify the location, number and the magnitude of
violations in the system (voltage violations or overloads) is
determined. Since the MCC can be readily obtained from
the power flow solution as indicated by equation (2), the
LMV for real and reactive power can be computed in Step
6, based on equation (3). The process can be repeated for
each hour in the planning horizon.
5. Test Cases Results
In this section, results from two networks are presented.
The first set is from a simple three-bus network shown
in Figure 3. The power flow results and the branch flow
overloads are provided in Tables 1 and 2. The tolerance
for the solution is equal to 1E-5. In Table 1 we impose
branch current limits such that there is an overload only
Cigre Science & Engineering • N°14 June 2019
108
on branch 1-2.As it can be observed from the reported
results, the LMV for real power is much higher than that
for reactive power. This is due to the load power factor.
The branch overload can be relieved when real and
reactive power is injected downstream the location of the
violation, as shown in Figure 1. Moreover, the LMV for
both real and reactive power seems to be increasing as
the electric distance from the violation increases. This
is intuitively correct, as DER further away would also
reduce additional losses. In Table 2, the simulation results
for modified current flow limits such that only branch
2-3 experiences violation have been presented. A few
interesting observations can be made here. First, the LMV
for real and reactive power has dropped significantly
at node 2 because the current limit for branch 1-2 has
more than doubled compared to Table 1; and the amount
of violation has dropped to zero, implying that the per
unit MCC value has decreased accordingly. The LMV
is not exactly zero because it accounts for the losses in
the network. As expected, the LMV for real and reactive
power are significantly higher for DER located at node 3
where it can contribute to alleviate the overload.
The second set of results is based on the analysis of
an IEEE 33-bus network. This case is set up such that
thermal overloads are exist on branches 1-5, as shown in
Figure 4. We also set the MCCT to be $12,000 annually
to represent the annualized cost of upgrades. The resulting
real and reactive LMVs are as shown in Figure 5 for the
peak hour. It is observed that both real and reactive LMV
increase as we move down the circuit down the circuit
such that DER at the nodes will affect more and more
overloaded branches. For all the nodes downstream of the
overloads, the LMV increase very slightly and only due
to losses. Lower LMVs in nodes 19-25 are due to limited
contribution of these nodes to relieving congestion in
upstream branches 1 or 2. It is also important to point that
reactive LMV shows the value of reactive injections at
each node in relieving overloads by improving the power
factor.
 Table I. Direct LMV calculation results for overload on branch 1-2.

Table II. Direct LMV calculation results for overload on branch 2-3.

Figure 4. IEEE 33-Bus Circuit with Overloaded Sections

Figure 5. Overload Only, LMVP and LMVQ

6. Conclusion nature of such value. A general mathematical formulation

In this paper a methodology for the direct calculation of
the locational marginal value of real and reactive power
by DERs has been presented. The focus of the paper is
to calculate the value of DER to the distribuiton grid
taking into consideration the locational and temporal
of the direct LMV calculation for both real and reactive
power based on the AC power flow equations has been
presented. The methodology was implemented as a
generalized algorithm providing a framework for the
consistent evaluation of the LMV of any DER regardless
of its type or technology.

Cigre Science & Engineering • N°14 June 2019

109
Results were provided for two test systems where the
impact of the location and the magnitude of violation of
the DER were clearly illustrated in the LMV. Future work
will include simulation results of the methodology for
large-scale realistic distribution feeders and investigate
the impact of the different DER technologies on the LMV.
7. References
[1] R. Tabors, G. Parker, P. Cantonella and M. C. Caramanis. (Apr.
2016) “White Paper on Developing Competitive Electricity Markets
and Pricing Structures”. [Online]. Available: https://www.bu.edu/
pcms/caramanis/NYPSC%20TCR%20WhitepaperApril2016.pdf.
[2] R. J. Bravo, R. Salas, T. Bialek and C. Sun, “Distributed energy
resource challenges for utilities”, IEEE 42nd Photovoltaic Specialist
Conference (PVSC), Jun. 2015.
Cigre Science & Engineering • N°14 June 2019
110
[3] P. D. F. Ferreira, P. M. S. Carvalho, L. A. F. M. Ferreira and M.
D. Ilic, “Distributed Energy Resources Integration Challenges in
Low-Voltage Networks: Voltage Control Limitations and Risk of
Cascading”, IEEE Transactions on Sustainable Energy, Jan. 2013,
p. 82-88.
[4] J. Smith, B. Rogers, J. Taylor, J. Roark, B. Neenan, T. Mimnagh
and E. Takayesu, “Time and Location: What Matters Most When
Valuing Distributed Energy Resources”, IEEE Power & Energy
Magazine, March 2017, p. 29-39.
[5] P. Adrianesis, M. C. Caramanis, R. Masiello, R. Tabors, S,
Bahramirad, “Locational Marginal Value of Distributed Energy
Resources as Non-Wires Alternatives”, Submited for per review in
IEEE Transactions on Power Systems.
[6] E. Ntakou and M. C. Caramanis “Distribution Network
Spatiotemporal Marginal Cost of Reactive
[7] D. P. Bertsekas, “Nonlinear Programming”, 3rd Edition, Athena
Scientific, 2016.
Effect of Earth surface on lightning
electromagnetic pulse propagation
A. YAMANAKA*1, N. NAGAOKA1, Y. BABA1, and M. SAITO2
Doshisha University
1
2
Central Research Institute of Electric Power Industry
Japan
Summary
Lightning strikes to electric power facilities with tall
structures often disturb the safety and stable operation of
these systems. Immediate identification of the lightning
struck facility is important to perform speedy and appropriate
recovering works. Accurate locating of strike points
provided by lightning location systems (LLSs) contributes
to preparing speedy and suitable countermeasures against
those accidents.
Lightning electromagnetic waves are delayed and deformed
by the earth surface conditions such as soil resistivity. It is
desirable to correct the propagation delay due to the earth
surface conditions for improving the accuracy of the LLS.
Effects of ground surface conditions, especially the
horizontally stratified ground case, on propagation
characteristic of a lightning electromagnetic pulse (LEMP)
are analyzed by the finite-difference time-domain (FDTD)
method in this paper. In order to simulate the ground
condition in a rainy day, soil resistivity within 1 m below the
ground surface is set a value lower than that of the second
layer. Since the 1 m thick ground surface layer with soil
resistivity lower than some tens of Ωm acts as a shielding
layer, LEMPs cannot penetrate into the second layer. Thus
the propagation speed is little reduced by the surface with
low resistivity. The over-correction for propagation delay
based only on a high soil resistivity assuming a sunny day
can result in the decrease of the location accuracy since
precipitation reduces the soil resistivity of the surface layer.
1. Introduction
Lightning strikes to electric power facilities with tall
structures, such as transmission and distribution systems
* cyjd1302@mail4.doshisha.ac.jp
KEYWORDS
Lightning, LEMP propagation, stratified ground, lightning location system, FDTD method
Cigre Science & Engineering • N°14 June 2019
111
and wind power generation plants, are one of the serious
threats to the safety and stable operation of them. Hence,
effective and reasonable countermeasures against the
lightning strikes to each facility are required. In addition,
it is necessary to identify the lightning struck facility
immediately to perform appropriate recovering works.
Lightning location systems (LLSs) are operated by, for
example, electric utility companies, and they have become
indispensable tools for increasing the reliability of the
power system. Accurate locating of strike points provided
by LLSs contributes to preparing speedy and suitable
countermeasures against lightning accidents.
The delay and deformation of the lightning electromagnetic
waves are affected by the earth surface conditions such as
soil resistivity. The propagation delay should be corrected
according to the conditions [1], [2]. In order to correct the
propagation appropriately, the effect of ground surface
conditions on propagation characteristic of lightning
electromagnetic pulse (LEMP) waveforms need to be
clarified.
This paper investigates the effect of ground surface
conditions, especially horizontally stratified flat ground [3].
The effect of the stratified ground, whose soil resistivity of
the first layer is higher than that of the second layer, has
been analyzed (e.g. in [3]) since that effect enlarges the
magnitude of the induced voltages on distribution lines
located at close or intermediate distance from the lightning
strike point. The characteristic of the opposite condition
(with a lower resistivity in the first layer), however, has
not yet been clarified. The surface layer can have lower
resistivity in rainy days, and this surface condition might
decrease the accuracy of propagation delay correction
based on a conventional method. These characteristics are
analyzed by means of the finite-difference time-domain
(FDTD) method and by using an approximated theoretical
formula.
2. Calculation methods for
analyzing effects of ground
surface conditions on LEMP
propagation characteristics
2.1 Analytical Formulation
When a lightning electromagnetic field radiated from a
lightning channel propagates over the perfectly conducting
flat ground, vertical electric field ezp appears at the
observation point (D, z) as shown in Figure 1 [4].
(3)
Since the wave front of the LEMP is mainly determined by
the lightning current at the bottom of the lightning channel,
the attenuation function shown in (3) can be written as
F(0, jω) [5]. Finally, the vertical electric field, observed
at sufficiently far from the lightning struck point, can be
written as follows.
(4)
The attenuation function for the channel base current is
given below.

(5)

The normalized surface impedance Δ is separately given for

a uniform and horizontally stratified ground cases. For the
(1) uniform ground case, the normalized surface impedance
Δuni and associated coefficients are given as follows [6]

where, ε0, c0, and vch are vacuum permittivity, light speed,
and lightning-current propagation speed along a lightning
channel, respectively.
The first, second, and third terms in the above equation
are electro static, induction, and radiation components,
respectively. The LEMP waveform on the flat ground
beyond some tens of kilometers, which is used for LLS,
can be regarded as the radiation component, and is written
in frequency domain as follows.
(6)
where ρuni and εruni are resistivity and relative permittivity of
the uniform ground.
For the horizontally stratified ground case, the normalized
surface impedance Δstr and associated coefficients are given
as follows [3], [7].

(2)

(7)
where ω is angular frequency and j is imaginary unit

When the LEMP propagates over imperfectly conducting

ground, an attenuation function F(z’, jω) [5] is introduced.

Cigre Science & Engineering • N°14 June 2019

112
where ρi and εri are resistivity and relative permittivity of
ith layer of a ground, and h1 is the depth of the first layer of
the ground.
Note that the approximated formulations mentioned above
neglect a curvature of the earth surface. Hence these
formulations are valid within 300 km from the lightning
struck point (in this paper the LEMP propagation within
50 km is discussed).
Figure 1. Geometry used in deriving approximate theoretical formula.
2.2 FDTD Model
Figure 2 shows the FDTD model in the 2-D cylindrical
coordinate system for calculating LEMP propagation over
the uniform or horizontally stratified ground [8]. An r-z
plane of the FDTD calculation space is divided uniformly
into 1×1 m square cells. The soil resistivity of ground is
set from 0 Ωm to 1 kΩm in the uniform ground case. In
the horizontally stratified ground case, soil resistivity
of ground within 1 m below the surface (1 cell) is set to
a value ranging from 1 Ωm to 500 Ωm, and that for the
second layer is set to 1 kΩm. The relative permittivity of
soil is set to 10 in all cases.
(a)
Figure 3. (a) Vertical electric field waveforms and (b) their time derivatives calculated with the FDTD method and theoretical expression in the
case of uniform ground.
Cigre Science & Engineering • N°14 June 2019
113
In order to reduce the computation time and memory
requirement, the FDTD analysis space is divided into
two, and a moving-window technique is employed [9].
The boundary of the FDTD analysis space is treated with
Liao’s second order absorbing boundary condition to
avoid undesirable field reflections. The lightning current
is expressed with a sum of Heidler’s function and double
exponential function. The rise time, 50% duration, and
maximum peak value of the channel base current are set to
1.2 μs, 30 μs, and 11 kA, respectively, to express a typical
subsequent return stroke current.
(8)
where I01 = 11 kA, I02 = 7.5 kA, η = 0.717, τ1 = 0.75 μs, τ2
 s, τ3 = 100 μs, and τ4 = 6.0 μs, respectively.
= 5.0 μ
The 7-km high straight vertical lightning channel is
expressed with the lossless transmission line model.
The propagation speed of the lightning current along the
lightning channel is set to the half of the light speed. Vertical
electric field waveforms at horizontal distance of 20 and 44
km from the lightning struck point are calculated.
Figure 2. FDTD analysis model of LEMP propagation over horizontally
stratified ground. FDTD analysis space is divided into two regions to employ a
moving-window technique.
(b)
3. Results and discussions Since the rise time of the employed lightning current is
1.2 μs, the maximum frequency for the investigation of
3.1 Uniform Ground Case
the vertical electric field 44 km away from the lightning
For the evaluation of the vertical electric field in the time struck point is considered to be around 0.5 MHz. The
domain by analytical expressions in the frequency domain penetration depth shown in Figure 4 indicates that in case
(4)–(6), modified Fourier transform is employed [10]. the soil resistivity of the first layer is set to 1 Ωm, the LEMP
Figure 3 (a) shows the vertical electric field at 44 km from propagation characteristic is mainly determined by this
the lightning struck point on the uniform ground with layer. On the other hand, in cases the soil resistivity of the
various soil resistivity. The results computed by the FDTD first layer is greater than 10 Ωm, the LEMP propagation
method agree well with those derived by the approximated characteristic is determined by the combination of the first
theoretical formulation. and the second layers.
The arrival of LEMP is determined at which the time
derivative of vertical electric field reaches to 80% of its peak
in this paper [11]. Figure 3 (b) shows the time derivatives of
the vertical electric field waveforms. The arrival of LEMP
in the case of 1-kΩm soil resistivity is delayed 0.95 μs
(corresponding to about 300 m) comparing with that in the
case of the perfectly conducting ground. The result shows
the importance of the propagation delay correction in LLSs
for improving the location accuracy [11].
3.2 Horizontally Stratified Ground Case
Figure 4 shows the penetration depth of the ground given Figure 4. Penetration depth of ground depending on resistivity and frequency.
by the following formula. Figure 5 shows the electric field magnitude in the uniform
ground with soil resistivity being 1, 10, 100, 1000 Ωm.
(9) The magnitude maps are calculated with the cell size of
10 × 10 m owing to the limitation of computational time by

(a) (b)

(c) (d)
Figure 5. Electric field magnitude in the uniform ground for soil resistivity of (a) 1 Ωm, (b) 10 Ωm, (c) 100 Ωm, and (d) 1000 Ωm.

Cigre Science & Engineering • N°14 June 2019

114
 (a) (b)

Figure 6. Vertical electric fields propagated 44 km over horizontally stratified ground. The solid lines show results calculated by the FDTD
method, and dashed lines show those by the approximated theoretical expressions.

(a) (b)

Figure 7. Time derivatives of vertical electric fields propagated 44 km over horizontally stratified ground. The solid lines show calculation result
by the FDTD method, and dashed lines show those by the approximated theoretical formula.

the FDTD method without the moving-window technique.
The electric field penetrates deeper with the increase of the
soil resistivity.
The ground with soil resistivity being 1 or 10 Ωm acts as
a shield against a high frequency component such as wave
front of LEMP waveforms. The characteristic indicates that
the LEMP propagation characteristics over the horizontally
stratified ground with surface resistivity of 10 Ωm or less
are determined by the surface. The LEMP propagation
characteristics over the stratified ground with surface having
higher than 100 Ωm of soil resistivity are determined by the
parameters of the both first and second layers.
Figure 6 and Figure 7 show examples of vertical electric
field waveforms and their time derivatives calculated by
the FDTD method and theoretical expression. As shown in
Figure 6 (a) when the soil resistivity of the first layer is set
to 500 Ωm, the electric field waveform is slightly distorted
and the peak value is slightly decreased comparing
with those in the case of a uniform ground with the soil
resistivity of 1 kΩm. However, the difference in the arrival
time is relatively small in terms of electric field derivative
waveform as shown in Figure 7 (a). When the soil resistivity
of the first layer is set to 10 Ωm, a clear difference appears
in comparison with the case of a uniform ground having the
soil resistivity of 1 kΩm. This is because the penetration
depth of the ground is about 0.5 m to 2 m at the frequency
around 0.5 MHz, and the electromagnetic field does not
sufficiently penetrate into the second layer having a soil
resistivity of 1 kΩm.
Figure 8 shows the delay of LEMP arrivals in cases of
uniform and horizontally stratified ground. When the soil
resistivity of the first layer decreases from a relatively high
value to some tens Ωm owing to rain or some other reasons,

(a) Analytical formula (b) FDTD method

Figure 8. Propagation delay of lightning electromagnetic field propagating over horizontally stratified ground
observed 44 km away from the lightning strike point. The soil resistivity of the second layer is 1 kΩm.

Cigre Science & Engineering • N°14 June 2019

115
 (a) Ratio of Ez max
Figure 9. Maximum value and rise time of the lightning electromagnetic field propagating 44 km over
a horizontally stratified ground.
the propagation delays are reduced from 1.0–0.9 μs to
around 0.4 μs. They correspond to from 300–270 m to 120
m. That means the constant propagation correction based
on the soil resistivity may decrease the location accuracy of
the area in rainy days, which has high soil resistivity such as
mountainous areas in Japan. As shown in Figure 9, though
the difference in the electric field amplitude is less than
10%, the rise time deviates 200%. Note that the ratios are
taken by setting the results derived in perfectly conducting
ground case as the reference.
4. Conclusion
This paper has studied effects of ground surface conditions
on propagation characteristics of lightning electromagnetic
pulses, especially the horizontally stratified ground case.
In the analysis, soil resistivity within 1 m below a ground
surface is set to a value smaller than that of the second
layer, assuming the decrease of soil resistivity in an area
due to precipitation. Since the ground surface with soil
resistivity lower than some tens Ωm acts as a shielding
layer, LEMPs are not capable of penetrating into the
second layer. The phenomena decrease the accuracy of the
propagation delay correction in rainy days at areas having
high soil resistivity. The paper has shown feasibility of
propagation delay corrections depending on weather
conditions for the advance of LLSs. The results obtained
in this paper might enable an estimation of rise-time of
lightning current using LLSs.
5. Bibliographies
[1] M. Matsui, K. Michishita, S. Kurihara, and N. Honjo,
“Discussions on location accuracy improved by propagation
delay corrections for the Japanese Lightning Detection Network”
(Proc. 33rd International Conference on Lightning Protection
(ICLP) 2016, Estorill, Portugal, 2016)
[2] N. Honma, K. L. Cummins, M. J. Murphy, A. E. Pifer, and T.
Rogers, “Improved lightning locations in Tohoku region of
Japan using propagation and waveform onset corrections” (IEEJ
Transactions on Power and Energy, vol. 133, no. 2, pp. 195–202,
2013)
[3] A. Shoory, A. Mimouni, F. Rachidi, V. Cooray, R. Moini, and S.
H. Sadeghi, “Validity of simplified approaches for the evaluation
of lightning electromagnetic fields above a horizontally stratified
Cigre Science & Engineering • N°14 June 2019
116
(b) Ratio of 10%–90% rise time
ground” (IEEE Transactions on Electromagnetic Compatibility,
vol. 52, no. 3, pp. 657–663, 2010)
[4] M. A. Uman, D. K. Mclain, and E. P. Krider, “The electromagnetic
radiation from a finite antenna” (American Journal of Physics,
vol. 43, pp. 33–38, 1975)
[5] V. Cooray, “Effects of propagation on the return stroke radiation
fields” (Radio Science, vol.22, no. 2, pp. 757–768, 1987)
[6] V. Cooray, M. Fernando, T. Sörensen, and T. Götschl,
“Propagation of lightning generated transient electromagnetic
fields over finitely conducting ground” (Journal of Atmospheric
and Solar-Terrestrial Physics, vol. 62, pp. 583–600, 2000)
[7] J. R. Wait, “The ancient and modern history of EM ground-wave
propagation” (IEEE Trans. Antennas and Propagation, vol. 40,
no. 5, pp. 7–24, 1998)
[8] Y. Baba, and V. A. Rakov, “Electromagnetic computation
methods for lightning surge protection studies” (Wiley-IEEE
Press; 2016, 1st edition, chap. 3, pp. 43–72)
[9] F. Akleman, and L. Sevgi, “A novel Finite-Difference Time-
Domain wave propagator” (IEEE Trans. Antennas and
Propagation, vol. 48, no. 3, pp. 839–841, 2000)
[10] N. Nagaoka, “Transient analysis of cable systems by means
of a frequency-transform method”(Ph.D. thesis, Department
of Electrical Engineering, Faculty of Engineering, Doshisha
University, 1992, chap. 3, pp. 3,19–3,24)
[11] M. Saito, T. Miki, and M. Miki, “Study on development of
advanced LLS (1) –Factors of location errors of LLS–” (CRIEPI
Report, Electric Power Engineering Research Laboratory, no.
H14007, 2014) (in Japanese)
6. Biographies
Akifumi Yamanaka received the B.Sc. and M.Sc.
degrees in Electrical Engineering from Doshisha
University, Kyoto, Japan, in 2017 and 2019, respectively.
He is currently a Ph.D. Student at the same University.
He mainly studies lightning transients in power systems.
Mr. Yamanaka is a student member of IEEE, the Institute
of Electrical Engineers of Japan, and the Institute of
Electrical Installation Engineers of Japan.
Naoto Nagaoka received the B.Sc., M.Sc., and Ph.D.
degrees in Electrical Engineering from Doshisha
University, Kyoto, Japan, in 1980, 1982, and 1993,
respectively. In 1985, he joined Doshisha University,
where he has been a Professor since 1999. From April
2008 to March 2010, he was the Dean of the Student
Admission Center, Doshisha University. From April
2010 to March 2012, he was the Director of both Liaison
Office and the Center of Intellectual Properties, Doshisha
University. Dr. Nagaoka is a Member of IEEE, the
Institute of Electrical Installation Engineers of Japan, and
the Institute of Electrical Engineers of Japan.
Yoshihiro Baba received the B.Sc., M.Sc., and Ph.D.
degrees from the University of Tokyo in 1994, 1996
and 1999, respectively. In 1999, he joined Doshisha
University, Kyoto, Japan, where since 2012 he has been
a Professor. From April 2003 to August 2004, he was a
Visiting Scholar at the University of Florida. He has been
the Convener of CIGRE C4.37 Working Group since
2014. He received the Technical Achievement Award from
the IEEE EMC Society in 2014. He is the Chairperson of
Technical Program Committee of the 2015 Asia-Pacific
International Conference on Lightning (APL), Nagoya,
Japan. He has been the Vice Chairperson of the APL
Steering Committee since 2017. He had been an Editor of
Cigre Science & Engineering • N°14 June 2019
117
the IEEE Trans. Power Delivery from 2009 until 2018. He
has been an Editor of the IEEE PES Letters since 2009,
a Guest Associate Editor of the IEEE Trans. EMC since
2018, and an Associate Editor of Electric Engineering
(Springer Journal) since 2019. He is a Fellow of IEEE
and IET.
Mikihisa Saito received B.S. in instrumentation
engineering from Keio University in 1997, and Ph.D.
degree in electrical engineering from the University of
Tokyo in 2009. He joined Institute of Industrial Science,
the University of Tokyo in 1999, and was a research
associate from 2013 to 2014. In 2014, he joined the Central
Research Institute of Electric Power Industry (CRIEPI),
Kanagawa (formerly in Tokyo), Japan. He has been
engaged mainly in the study of lightning phenomena. Dr.
Saito is a member of IEEE and the Institute of Electrical
Engineers of Japan.
Material and emerging test techniques
general overview of AC and DC current
injection on high voltage potential for
HVDC long-term tests
M. HALLAS*, V. HINRICHSEN
Technische Universitaet Darmstadt, High Voltage Laboratories
Germany
Summary
HVDC transmission systems like cables or gas insulated
systems are of great interest for future grid expansions.
They are usually type tested in one-year long-term tests,
to ensure lifetimes up to 40 years and more. HVDC
equipment always has to be tested with voltage and current
at the same time to investigate temperature dependent
space charge effects inside the insulation material. Even
though the HVDC equipment in the grid is operated with
DC current and DC voltage, laboratory tests often have to
be performed with AC current heating due to laboratory
limitations. The following report shows the technical
limits of AC current injection and presents solutions
for DC current injection on high voltage potential. The
solutions are described and compared to each other.
Furthermore, the report shows the status of actual research
and prototyping for this new generator type.
1. Introduction
The design of HVDC equipment like cables or gas
insulated systems always has to take space-charge
distributions in the insulation material into account.
Because of accumulating charges on insulator surfaces
and the material, the electrical field distribution changes
depending on the temperature and the duration of the
voltage stress. High local electrical field stress can occur
due to growing space charge accumulation. These effects
depend on the electrical resistance of the insulating
material and thus also on the temperature. Therefore,
adequate testing of HVDC equipment can only be achieved
with thermal and electrical stress at the same time [1].
The heating of the test object is usually implemented by
* hallas@hst.tu-darmstadt.de
KEYWORDS
High voltage testing techniques, HVDC testing, current injection, prototype installation test, prequalification test
Cigre Science & Engineering • N°14 June 2019
118
current injection to the inner conductor to simulate actual
operating conditions. This can be performed by various
methods, which are described and compared in this report.
2. Long-term testing of HVDC
equipment
One year long-term tests are specified for HVDC
transmission systems like gas insulated systems or cable
systems. For HVDC cables, they are specified in IEC
62895 [2] as prequalification tests, whereas long-term
tests on gas-insulated HVDC systems are discussed in
Cigré JWG D1/B3.57 [1] as prototype installation tests.
The test procedures are different. During prequalification
tests of cable systems, the design limits of the conductor
temperature and the maximum temperature difference
across the insulation shall be reached, so that the
technical limits and the aging of the equipment is tested.
Therefore, the cable system is heated by a current, which
may differ from the nominal current rating [2]. Since
only the temperature and the temperature gradient are
of interest, long-term tests are
usually performed with AC
current injection to the inner
conductor according to Figure
1. Furthermore, IEC 62895
[2] requires a dummy loop
of several meters, without
any high voltage applied.
The dummy loop is used to
determine the temperature
at the inner conductor. The
advantage of AC current
Figure 1: AC current injection
 Figure 2: Heating of components during AC current testing with two-points and single-point earthing
(red = heated) [4]
injection is that the same current generator type can
easily be used for the dummy loop [7].
The prototype installation test for gas insulated systems
has a different purpose and, therefore, follows a different
test procedure. Gas insulated systems like switchgear or
lines form a more complex arrangement with several
contact surfaces. These are usually the most critical
components during the heating process. Therefore,
temperature rise tests on gas insulated systems are
performed with the nominal current. The temperature
rise at the contact surfaces and the overall assembly is
observed and has to be within the limits according to IEC
62271-203 [3]. The same procedure is specified for gas
insulated HVDC equipment [1]. Due to test equipment
limitations, usually AC current heating according to
Figure 1 is also used for gas insulated systems.
2.1. Technical limits of ac current testing
AC current heating has some technical limitations.
Especially larger HVDC test objects of large cross
sections have a higher reactive power demand due
to their high inductance and high nominal current of
the test object. Considering a typical current loop, the
overall inductance of the assembly will be in the range
of 1 μH/m [4]. With a testing current of 5000 A at 50
Hz, the required reactive power would be in the range
of 7,85 kVA/m and the active power in the range of only
0,3 kW/m. Assuming a 300 m testing loop, the reactive
power would be 2.4 MVA, while the active power would
be in the range of only 90 kW. The need of reactive
power to feed the test loop results in larger AC current
sources and power supplies. The high inductance during
AC current heating of HVDC equipment results from
the single point earthing of the outer sheath or enclosure,
respectively. Single point earthing is necessary to
avoid induced currents on the sheath/enclosure, which
would not occur during real service operation of the
DC equipment. Such sheath/enclosure currents would
change the temperature difference across the insulation
from ΔT to ΔT’. Figure 2 shows the differences between
two-points and single-point earthing [4].
2.2. Customer arguments against ac current heating
For special arrangements, AC current heating may also
Cigre Science & Engineering • N°14 June 2019
119
be stretched to its limits. Especially submarine cables
with iron armoring will have eddy currents and hysteresis
losses in the armoring, which influences the temperature
distribution. The armoring might be removed, but the
customer needs to accept the change in the test object.
Gas insulated systems with iron enclosures would have
the same limits, so that special agreements with the
customers of the HVDC equipment would have to be
made [7].
Gas insulated systems shall be tested with rated current.
This can be achieved with rated DC current, with a
representative AC current or the injection of the rated
DC current value as AC current. Testing with rated
DC current is only possible with the use of DC current
injection.
A representative AC current during the tests has to be
determined by pre-tests. The customer might assume a
different temperature distribution in case of AC current
heating instead of DC current heating, which might
distort the test results. The same or higher temperatures
for the representative AC current compared to the
rated DC current have to be ensured and proven to
the customer. To ensure the performance for higher
temperatures during testing the manufacturer of HVDC
equipment has to oversize the equipment. This results in
higher material investment for the design. The customer
might also assume induced currents in the overall
test assembly, which might change the temperature
distribution. For example, zero load and high load test
circuits in parallel will always result in induced currents
in the zero load test loop [7]. To solve all these issues,
special agreements for testing are necessary.
The injection of the rated DC current value as AC current
will result in much higher temperatures at the whole test
object. Therefore, the manufacturer has to oversize the
total design to pass the test.
3. DC current injection
To summarize, tests on HVDC equipment have to
ensure reliable operation during the total lifetime in the
HVDC application [1]. Usually power system operators
desire a test procedure as close as possible to practice,
which would become possible by DC current injection
(section 2.2). Furthermore, the technical limits of
AC current heating described in section 2.1 would be
solved. Even heating of AC cables by DC current could
be of interest for high voltage laboratories in order to
reduce the reactive power demand and to be able to
test longer loops of higher cross sections. On the other
hand, the dummy loop during prequalification tests of
cables mentioned in chapter 2 has to be considered for
DC current injection. In this case, a separate DC current
source or other technical solutions have to be considered
for the dummy loop [7].
Figure 4: Basic sketch for DC current injection on high voltage potential [4]
A basic sketch of a DC current injection IDC on high
voltage potential UDC is shown in Figure 4 [4]. The
test object can be represented by a resistor. The current
source must generate the high DC current on high voltage
potential. The power supply of the current source has to
fulfill two tasks. First, it has to transmit the power for
the current source. Secondly, it has to insulate the high
voltage of the test object against ground.
3.1. Technical implementations
To solve the task sketched in Figure 4, several technical
solutions are possible. Concepts to inject DC current
on high voltage potential are shown in Figure 5. All
concepts comprise a rectifier operating on high voltage
potential. For this rectifier cooling devices suitable for
high voltage applications are necessary. The capacitive
current injection (1) uses capacitors in order to insulate
the high DC voltage and to simultaneously transmit the
power. An AC/AC converter operated at several kHz
Cigre Science & Engineering • N°14 June 2019
120
feeds the capacitors, which results in a high frequency
current through the capacitors. The high frequency
current is transformed and rectified on high voltage
potential. The overall circuit operates in resonance. The
resonant frequency of the circuit can be met by adjusting
the AC/AC-converter [4].
The hydraulic current injection (2) uses a hydraulic
motor-generator-combination, while the hydraulic oil
pipes insulate the high voltage. The hydraulic engines
are combined with electrical motor/generator engines
to receive electrical power, which can be transformed
and rectified on high voltage potential. The hydraulic oil
is pumped through a closed loop from ground to high
voltage potential.
The inductive DC current injection (3) uses conventional
AC current transformers to generate a high AC current in a
shorter loop, which can directly be rectified on high voltage
potential in order to feed the larger testing loop [5].
The DC current injection by an isolating shaft and
electrical engines (4) uses a motor- generator-combination
to transmit the power mechanically. The isolating shaft
insulates the high DC voltage. The mechanical power is
converted to electrical power on high voltage potential,
then transformed and finally rectified.
The current injection by isolating transformers (5) in
series insulates the voltage and transmits the power
simultaneously. A resistive voltage divider ensures a
homogenous voltage distribution among the isolating
transformers. On high voltage potential, the current is
transformed and rectified [6].
The current injection with a fuel-power-system (6) uses
a fuel generator on high voltage potential to generate the
electrical power. Since the generator can be operated
only for a couple of hours with one tank filling, an
insulated fuel-feeding has to be designed.
3.2. Comparison of technical implementations
A generator for DC current injection has to fulfil
several tasks during long-term tests. Most importantly,
the following points have to be considered during the
testing, operation and design:
1. Testing
1.1 DC voltage long-term performance (DC-
perform.)
 1. Capacitive current injection [4] 4. Isolating shaft and electrical engines
2. Hydraulic current injection 5. Isolating transformers [6]
3. Inductive DC current injection [5] 6. Fuel-power-system
Figure 5: Concepts for DC current injection on high voltage potential

1.2
Polarity reversal test performance (polarity rev.) 3. Design and erection of the generator
1.3
DC current ripple 3.1 Overall costs of the generator
1.4
Electro-magnetic compatibility (EMC) 3.2 Use of commercial standard parts without
1.5.
Possibility of superimposed impulse voltage tests redesign (stand. parts)
(SIMP) 3.3 Weight of the overall assembly
1.6. Influence on partial discharge measurements (PD) 3.4 Upscaling / downscaling of the concept for
2. Laboratory Operation different voltage, current and power. All concepts
2.1 Handling and maintenance of the test generator of Figure 5 are compared in Table 1. Each point
2.2 Controlling of the generator is evaluated separately for each concept and is
2.3. Influences on parallel working areas marked with “+” if an advantage is estimated,
2.4. Space requirements “0” if neutral and “-” in case of a disadvantage.

1. Testing 2. Laboratory 3. Design

1. DC-perform.

1. maintenance
2. polarity rev.

2. stand. parts
2. controlling

3. Influences

4. upscaling
3. weight
5. SIMP
3. ripple

4. EMC

1. costs
4.space
6. PD

1. capacitive + 0 + - - - + + 0 + + + + +
2. hydraulic + + 0 + + + - 0 - 0 + + 0 0
3. inductive 0 0 + + 0 + + + + - - 0 + -
concept

4. mech. shaft 0 + 0 + + + - 0 - 0 0 - 0 0
5. transformer 0 0 + + + + + + + - - 0 - +
6. fuel-engine - - 0 - - 0 - 0 - + + + + 0

Table 1: Comparison of the concepts in Figure 5 (“+” = advantage, “0” = neutral, “-” = disadvantage)

Cigre Science & Engineering • N°14 June 2019

121
The capacitive current injection (1) uses the total distance compared to other concepts, since the rotation needs to
between high voltage and ground to insulate the DC be slowed down. For the design of the generator, many
voltage. Furthermore, the RC-characteristics result in commercial standard parts can be used, which results in a
a very homogeneous voltage distribution. For polarity very cost efficient design. Since hydraulic drives usually
reversal, the overall capacitance of the RC transmission only operate in lower power regions, the technical limits
path has to be regarded, such that polarity reversal is of hydraulic drives set a limit to technical ratings, such as
possible in the specified time. Since inductances are maximum test loop geometry and maximum current.
included in the transmission path, the current source has
The functionality of the inductive current injection is
to be switched off during superimposed voltage tests.
highly dependent on the system used in the transmission
Special overvoltage protection during superimposed
path. The concept will be most efficient, when the shorter
impulse voltage testing might be possible, but has to
transmission path loop and the test object consist of the
be developed. In this case still the capacitance of the
same material and type. Therefore, both would have
current source has to be taken into account, since the
the same current and voltage ratings. Considering an
high capacitance may influence the voltage shape. The
AC current trough the cable/gas insulated system in the
high frequency generator has the advantage of a very low
transmission path, the temperature at the transmission
current ripple and a very compact design, with low space
path will be higher than the same assembly at the test
requirements. Furthermore a very flexible controlling,
object stressed with DC current. This results in a poor
without maintenance requirements will be possible. On
insulation performance. Therefore, the risk of failure
the other hand, the high frequency may influence PD
inside the transmission path will be higher than the risk
measurement systems, which are operated in the kHz
of failure for the test object. In conclusion, the current
range. Also currents at the grounding wire due to the
and voltage rating of the cable/gas insulated system for
AC/AC converter have to be considered, which might
the transmission path has to be higher than the ratings of
influence other laboratory segments. Due to the extensive
the test object. If special designs for the transmission path
use of electronics the generator can be designed very cost
or two parallel systems are used, the costs increase. The
efficient. In addition, mainly standardised components can
necessary space for current loop and terminations will
be used. Therefore, a strictly modular concept is possible
also be higher than in other concepts [7]. Superimposed
that can easily be scaled to higher voltage ratings. On the
voltage tests will be possible, but the overall capacitance
other hand, electronics are sensitive to electro-magnetic
of the transmission path has to be considered.
interferences especially during flashovers, which has to
be considered during the generator design. The performance of the current injection with isolating
shaft (4) strongly depends on the performance of the shaft.
The hydraulic concept (2) has a robust insulation across
Since the shaft has to withstand permanent mechanical
the total insulating distance provided by insulating oil. It
and electrical stresses, comprehensive knowledge about
is not affected by impulse voltage stress. Therefore, the
the material and its long-term behavior under multi-
insulation will be able to easily manage the electrical stress.
physical stress is required. Since electrical drives are used
Furthermore, the impulse voltage shape is not influenced
in this concept, similar disadvantages as in concept 2 are
by the transmission path, since its total capacitance will
assumed. Compared to other concepts, costs are higher,
be very low. As long as the total assembly itself is free of
because special parts and assemblies and, therefore,
partial discharges, no influence during PD measurement
engineering effort is required. For higher voltage levels
is to be expected. Since electrical drives with 50/60 Hz
and higher rated power the isolating shaft has to be
will be used, the current ripple and the space requirements
extended, with consequently ever higher mechanical
will be higher compared to other concepts. In addition,
requirements.
the drives will require higher maintenance efforts and
produce more audible noise in the laboratory compared A major disadvantage of the current injection with
to non-rotating concepts. The generator control will be isolating transformers (5) is that only a small distance
less flexible compared to other concepts. For example inside the transformer can be used to insulate the voltage.
more time for switching off the generator will be required Therefore, the system can be very sensitive to DC effects

Cigre Science & Engineering • N°14 June 2019

122
 Figure 6: (a) Capacitive current transmission [4] / (b) Inductive current transmission with isolating transformers
during long-term DC and polarity reversal stress. Thus,
a certain number of transformers have to be used in
series, which increases costs and the weight of the overall
assembly. If these disadvantages can be managed, the
concept will have a very stable operating behavior since
no rotating machines are used. The concept can easily be
scaled up by adding further isolating transformers. But
each transformer has to carry the rated power and thus
to be dimensioned with the required cross-section of the
copper windings. Overall material invest and weight and
thus the costs of the design will be higher compared to
other concepts.
The current injection with a fuel-power-system is a very
simple and cheap solution to fulfill the task. But the high
risk of fuel explosion in combination with electrical arcs
and the constant fuel feeding are crucial points for the
designer and the operator of such generator. The influence
on PD measurement is uncertain [7].
3.3. Prototype experience
Part of the concepts described in Figure 5 have already
been prototyped or are actually in discussion to
overcome the limitations of AC current injection. Known
publications also show the progress in research for this
new generator type. Figure 6 shows CAD prototypes of
different generator concepts.
A capacitive current injection generator is being built up
in the high voltage laboratories at Technische Universitaet
Darmstadt. First experiences with smaller prototypes
were positive and showed good performance. The
construction of a larger generator for 5000 A and 660 kV
voltage is currently in progress. It consists of a feeding
unit on ground potential, two capacitor columns similar to
voltage dividers and the rectifier on high voltage potential.
Due to the high frequency of approximately 50 kHz the
assemblies on high voltage potential can be designed very
compact.
The current injection with isolating transformers had
been investigated before, but was discarded during
Cigre Science & Engineering • N°14 June 2019
123
the development. The feeding unit could be designed
very simple. Only a standard regulating transformer
was required. Transformers with solid insulations were
considered and investigated during the design process.
Since the DC voltage has to be insulated between primary
and secondary winding, only a small insulating distance
is possible. Thus, insulation breakdowns occurred during
the prototyping process. This proves that insulation faults
are more likely compared to other concepts. Figure 6 (b)
shows a possible implementation of the overall assembly.
It turns out that this option is far less compact than the
capacitive solution (Figure 6 (a)).
4. Conclusion
DC current injection on high voltage potential offers
new technical possibilities for laboratory tests. Technical
limits of AC current testing can be overcome. Especially
the high reactive power consumption by AC heating of
HVDC equipment of high current ratings can be avoided.
Also all customer requirements for adequate testing
with the real technical ratings can be fulfilled. Even
DC current heating during long-term tests of AC cable
systems is of interest to achieve higher current ratings
and larger cross sections. Several technical solutions
can be developed to inject DC current on high voltage
potential. Each technical solution has its advantages
and disadvantages. Running prototyping projects for the
generator concepts are shown. Especially the capacitive
current injection seems very promising and is currently
followed.
5. Acknowledgment
The authors gratefully acknowledge support of this
work by the IWB-EFRE-Program of the State of Hessen
(Funding Code 20002558).
6. Bibliography
[1] Neumann, Claus ; Hallas, Martin ; Felk, Mahmoud ; Tenzer,
Michael ; Riechert, Uwe : “Some thoughts regarding prototype
installation tests of gas-insulated HVDC systems” In: CIGRÉ
Winnipeg 2017 Colloquium, September 30 – October 6, 2017
[2] IEC 62895:2017: “High voltage direct current (HVDC) power
transmission - Cables with extruded insulation and their
accessories for rated voltages up to 320 kV for land applications
- Test methods and requirements”
[3] IEC 62271-203:2011: “High-voltage switchgear and controlgear
- Part 203: Gas-insulated metal-enclosed switchgear for rated
voltages above 52 kV”
Cigre Science & Engineering • N°14 June 2019
124
[4] Hallas, Martin ; Wietoska, Thomas ; Hinrichsen, Volker :
“Generator for Current Injection on High DC Potential to Test
HVDC Equipment”. In: The 20th International Symposium on
High Voltage Engineering, Buenos Aires, Argentina, August 27 –
September 01, 2017
[5] W. Schufft, S. Schiering “Developmets in Power Engineering and
their Demands on High-Voltage Test Equipment” ELECO 1999.
[6] B. Yang, “Device for Testing Load of superconducting High-
Voltage DC Cable”, patent WO2014046467 A1.
[7] C. Frohne, “Exchange of Experience”, Nexans Deutschland
GmbH, Hannover, Germany, 2018
Application strategies for externally
gapped line arresters against lightning
outages on the 400 kV overhead
transmission lines in Hong Kong

K. VELITSIKAKIS1*, C. ENGELBRECHT1, S.K. LAU2**, W.K. CHEUNG2

DNV GL, The Netherlands1
CLP Power Hong Kong Limited, Hong Kong2

Summary are presented. It is noted that each line needed to be

Lightning related outages on overhead transmission
lines impact the reliability of the grids and, thus, form
a major concern to power supply utilities. In 2000s,
CLP Power Hong Kong Limited (CLP), who operates a
vertically-integrated electricity supply business in Hong
Kong, have experienced numerous outages on their
400 kV overhead lines due to high lightning activity.
As an initiative to improve power quality and supply
reliability, externally gapped line arresters (EGLAs)
have been introduced to CLP’s 400 kV overhead line
system. By increasing the line arrester coverage on the
400 kV towers, the lightning performance of the 400 kV
overhead line network has been greatly improved.
To evaluate the effectiveness of installing additional
line arresters on the 400 kV overhead transmission
lines, a detailed electromagnetic transient (EMT) study
was conducted with joint efforts from CLP and DNV
GL. In order to compare the performance of various
arrester schemes and to define cost-effective application
strategies, the study considered important parameters
that have a significant impact on the line performance.
In particular, focus was given on the impact of the tower
footing resistance on the backflashover rate and the
impact of the ground inclination angle on the shielding
failure flashover rate.
In this paper, the approach and the results of the analysis
studied separately from the rest in order to define cost-
effective strategies for the application of additional line
arresters. By doing this, the cumulative effect of the
tower footing resistance and the terrain characteristics at
the location of each tower structure are considered in the
final evaluation.
1. Introduction
In general, lightning activity during adverse weather
conditions could be the root cause for most outages
on overhead transmission lines. More specifically, a
lightning flashover across an insulator string could occur
a) due to shielding failures, i.e. when a lightning leader
bypasses the shield wire(s) and directly strikes one of the
phase conductors (Figure 1a), or b) due to backflashover
conditions, i.e. when a lightning leader strikes either
the shield wire or the tower structure (Figure 1b). The
total flashover failure rate of a transmission line due to
lightning depends on various parameters, such as the
ground flash density at the area of interest, the tower
configuration, the ground terrain conditions, etc. In
cases of poor lightning performance of a line, especially
in areas of high ground flash density and/or in areas of
poor grounding conditions, e.g. rocky terrain, line surge
arresters [1-5] form an effective option for improving
the line performance and, thus, ensuring a more reliable
operation of the transmission system.

* Konstantinos.velitsikakis@dnvgl.com
** siukilau@clp.com.hk

KEYWORDS
Line arresters, EGLA, Lightning performance, Shielding failure, Backflashover

Cigre Science & Engineering • N°14 June 2019

125
 Figure 1a. Example of a shielding failure
Hong Kong is located in the sub-tropical zone having
fairly high lightning activity. With the increasing use
of electronic controls for motor drives in industrial and
commercial sectors, power quality, in particular voltage
dips, is becoming a prominent issue. CLP is committed
to continuously improve the services to customers of
Hong Kong and, therefore, has initiated the installation
of line arresters on the 400 kV overhead transmission
lines since early 2000s. Following the increased line
arrester coverage on the 400 kV overhead transmission
lines, the lighting performance of the latter has been
significantly improved.
In 2017, CLP conducted a detailed study to evaluate the
effectiveness of installing additional line arresters on the
400 kV double-circuit overhead transmission lines. The
purpose of the study was a) to evaluate the effectiveness
of increasing the total number of installed arresters on
the lightning performance of the lines and b) to define
cost-effective application strategies.
2. Overview of the input data
CLP owns and makes use of a Lightning Location System
(LLS) to record lightning activity in Hong Kong. Based
on these recordings, the yearly average ground flash
Figure 2a. GFD map in Hong Kong
Cigre Science & Engineering • N°14 June 2019
126
Figure 1b. Example of a backflashover
density (GFD, number of flashes/km2/year) is estimated
for various locations in Hong Kong, as presented in
Figure 2a. In the study, the GFD was considered equal
to 5.2 flashes/km2/year, corresponding to the average
GFD, as recorded in 2016.
A typical configuration of a 400 kV suspension tower
in CLP overhead transmission lines is shown in Figure
3. In addition, for those phases of the lines protected by
means of EGLAs, there is a fixed external gap between
the arresters and the phase conductor (Figure 4).
The local terrain features for each tower structure were
deemed essential for studying the lightning performance
of an overhead transmission line against shielding failure
flashovers. In the study, the terrain features referred to
the ground elevation and the ground inclination angle,
φ°, at the location of each tower. An example is given in
Figure 5 that presents the terrain features along a 400 kV
line in Hong Kong consisting of 24 towers (tower No.1
refers to the substation gantry). The analysis of the
terrain features for all lines have shown that in average
54% of the tower structures are erected on a flat terrain,
i.e. φ<5°, 40% are erected on a moderately hilly terrain,
i.e. 5° ≤φ<15°, and 6% are erected on an extremely hilly
terrain, i.e. φ≥15°.
Figure 2b. Average recorded GFD in Hong Kong
 Fixed
External Gap

Figure 4. Typical EGLA configuration


 Figure 3. Typical suspension tower configuration

Figure 5. Terrain features of the tower structures of a 400 kV double-circuit line in Hong Kong

3. Backflashover analysis structure to structure. Therefore, the impact of the tower

At first, the backflashover analysis was performed
to study the performance of all possible line arrester
schemes, including the reference case, where no arrester
is applied on both circuits of a double-circuit tower
(Figure 6). Based on field measurement records, tower
footing resistance exhibits a strong variation from
footing resistance both on the critical lightning current
to cause a backflashover and on the backflashover rate
(BFR) was considered, by varying its value between
10 and 50 ohms. Figures 7a and 7b present the analysis
results; the backflashover rate is expressed in pu values
(1 pu refers to the reference case).

Figure 6. EGLA schemes (filled square in red indicates the presence of an arrester at that phase)

Cigre Science & Engineering • N°14 June 2019

127
 Figure 7a. Minimum critical lightning current leading to backflashover

Figure 7b. Backflashover rate analysis results expressed in pu

The analysis showed that the backflashover performance
is to a large extent influenced by the tower footing
resistance. This is because of its direct impact on the
minimum critical lightning current that leads to a
flashover of at least one phase. More specifically, for
tower structures in areas where low values of the tower
footing resistance can be achieved (e.g. Rtower≤10 ohms),
the minimum critical currents have extremely high
values, resulting in very low backflashover failure rates.
The latter is illustrated in Figure 7a, where the critical
lightning current is greater than 230 kA for each phase
angle of the power frequency voltage. On the other hand,
for tower structures in areas where higher tower footing
resistances occur (e.g. Rtower >10 ohms), lower critical
currents occur, resulting in higher backflashover rates.
For these rogue structures, the study results revealed
that the bottom phase of a circuit is more prone to
backflashovers, followed by the middle and the top
phases respectively. This can be explained as follows:
the lightning current results in a voltage built-up on the
metallic structure as well as on the shield wire. Each
phase conductor is electrically coupled to the shield wire
and the coupling factor is distance dependent. Therefore,
the top phase conductor, which is located to a closer
distance to the shield wire, will experience a higher
coupled voltage compared to the middle and bottom phase
respectively. Consequently, the voltage stress across the
line insulation is lower for the top phase and higher for
the bottom phase. Thus, by protecting the bottom phase
by means of a line arrester, it was calculated that the
backflashover rate would be significantly reduced.
4. Shielding failure flashover
analysis
According to reports published by Utilities in China
[6,7], it has been observed that the shielding failure
rates in their transmission lines are higher compared to
the estimated rates, as calculated by the conventional

Cigre Science & Engineering • N°14 June 2019

128
 Table 1. Maximum penetration lightning currents on a phase conductor

Flat terrain More exposed circuit Less exposed circuit

Im [kA] Im [kA] Im [kA]
φ=0ο φ=10ο φ=20ο φ=10ο φ=20ο
Top phase 13.0 20.0 34.5 11.0 10.5
Middle phase 10.5 17.5 32.0 9.5 9.0
Bottom phase 4.0 6.5 10.5 4.0 4.0
SFFOR [fl/0.37 km/year] 0.00008 0.00104 0.00659 0.00 0.00

electro-geometric model (EGM). Similar observations 5. Evaluation of the EGLA

have been made by the Utility in Japan, as published in
[8]. One common characteristic of these lines has to do
with the height (>50 m) and the vertical configuration of
the tower structures. Due to the above, for the analysis of
the shielding failure flashover rate, reference was made
to the modified electro-geometric model (EGM), as
presented in [9-10]. The maximum penetration lightning
current was calculated according to this model, taking
into consideration the impact of the ground inclination
angle as well. Table 1 presents a) the values of the
maximum penetration currents for each case and b) the
calculated shielding failure flashover rate (SFFOR) for a
tower structure. SFFOR is expressed in flashovers/0.365
km/year, considering a typical span of 365 meters for
a 400 kV transmission line. The values in red italic
indicate that the maximum penetration current exceeds
the critical flashover voltage (CFO) of the shortest air-
gap. It is evident that the ground inclination angle has a
significant impact on the exposure of a tower to direct
lightning strikes and, therefore, to the shielding failure
flashover performance of a line. More specifically, for
a tower structure erected on a hilly terrain, the circuit
facing away the hill is regarded as more exposed to
direct strikes compared to the circuit facing towards the
hill. This can be explained by the fact that the ground
plane falls further away from the phase conductors for
the more exposed circuit.
application strategies
To evaluate various strategies for applying additional
EGLAs on the 400 kV overhead transmission lines, the
total flashover rate (TFOR) was calculated as:
For the backflashover performance, the tower footing
resistance was regarded as a decisive factor. Similarly,
for the shielding failure flashover performance, the
terrain characteristics of each line route were analyzed
in terms of ground elevation and inclination angle. By
doing this, the percentage of the tower structures erected
on a flat or a hilly terrain was determined. In this case,
the total shielding failure flashover rate of a line was
calculated as follows:
Based on the above, the cumulative effectiveness of the
line arrester schemes on both the backflashover and the
shielding failure flashover performance was evaluated.
Table 2 presents the results of the inclination angle
analysis for three lines, where the percentage of exposed
towers to direct strikes varies. Figures 8-10 compare
the total flashover rate of each line for various arrester
schemes.

Table 2. OHL ground inclination analysis

Flat terrain Moderately hilly Extremely hilly

[% of total number of towers] [% of total number of towers] [% of total number of towers]

Line 1 50 50 0

Line 2 25 75 0

Line 3 50 35 15

Cigre Science & Engineering • N°14 June 2019

129
 Figure 8. TFR analysis for line 1

Figure 9. TFR analysis for line 2

Figure 10. TFR analysis for line 3

Cigre Science & Engineering • N°14 June 2019

130
6. Conclusions
The lightning performance study conducted for the CLP
400 kV overhead lines in Hong Kong revealed that each
line needs to be studied separately in order to evaluate the
effectiveness of installing additional line arresters. The
extent of natural shielding of the line by the geography and
the structure footings are regarded as two important factors,
when trying to define a cost- effective strategy for placing
additional line arresters. For instance, tower structures at
line sections characterized by high tower footing resistance
values should be protected against backflashovers compared
to towers with relatively low footing resistance values. In
such case, installation of line arresters on the bottom and/or
middle phase is more effective than protecting the top phase.
Also, towers on hilly terrains or on mountain tops should
also be protected against shielding failure flashovers. In this
case, the middle and/or the top phase should be targeted
for adding line arresters rather than protecting the bottom
phase. Consequently, such strategies become cost-efficient,
by improving the lightning performance of vulnerable
towers and, therefore, the performance of the line.
7. Bibliographies
[1] IEC 60099-5, Surge Arresters, Part 5: Selection and application
recommendations, 2013
[2] IEC 60099-8, Surge Arresters, Part 8: Metal-oxide surge arresters
with external series gap (EGLA) for overhead transmission and
distribution lines for A.C. systems above 1 kV, 2001 [3] CIGRE
Working Group C4.310, Technical Brochure No. 440, Use of Surge
Arresters for Lightning Protection of Transmission Lines, 2010
[4] E. Volpov, Application and dimensioning of line surge arresters to
improve lightning performance of the IECo transmission grid 170-
kV, CIGRE Science & Engineering, 2017
[5] I. Mohamed Rawi, M.Z.A. Ab-Kadir, Effectiveness of Line Surge
Arrester Application on the 132 kV Kuala Krai – Gua Musang Line,
CIGRE Session, 2014
[6] L. Xiaolan, C. Jiahong, G. Shanqiang, T. Xuefang, Statistics and
Analysis of Lightning Flashovers of Transmission Lines during 2000-
2007, IEEE, 2008
[7] X. Zhao, Y. Xie, H. He, J. He, X. Chen, H. Cai, Lightning Performance
Assessment of 500 kV Transmission Lines in Southern China, IEEE,
2013
[8] CIGRE Working Group C4.26, Technical Brochure, Evaluation of
Lightning Shielding Analysis Methods for EHV and UHV DC and
AC Transmission Lines, 2017
[9] S. Taniguchi, T. Tsuboi, S. Okabe, Y. Nagaraki, J. Takami, H. Ota,
Improved Method of Calculating Lightning Stroke Rate to Large-
sized Transmission Lines Based on Electric Geometry Model, IEEE
Transactions on Dielectrics and Electric Insulation, Vol. 17, 2010
[10] S. Taniguchi, T. Tsuboi, S. Okabe, Observation Results of Lightning
Shielding for Large-scale Transmission Lines, IEEE, 2008
Cigre Science & Engineering • N°14 June 2019
131
7. Biographies
Konstantinos Velitsikakis is a senior expert in the field
of Insulation Co-ordination and Electromagnetic Transient
(EMT) studies for transmission systems. His expertise is
related to insulation co-ordination and electrical aspects
for the design of transmission lines (HVAC and HVDC),
insulation co-ordination for HV substations and system
studies (switching, lightning and temporary overvoltages,
transient recovery voltages for circuit breakers, resonance
and ferro-resonance conditions, harmonics, etc.). He is an
active member of Cigré, acting as the secretary of the C4.46
Working Group on “Evaluation of Temporary Overvoltages
due to Low Frequency Resonance Conditions” and being
active member of the C4.48 “Overvoltage Withstand
Characteristics of Power System Equipment 35-1200 kV”
and C3.13 “Interactions between Electrical Infrastructure
and Wildlife”.
Since January 2018, he has taken up the role of team leader
within the department Transmission and Distribution
Technology of DNV GL Energy, Arnhem, The Netherlands.
Christiaan S. Engelbrecht is a specialist in Insulation
Coordination. He has more than 25 years of international
experience in the contamination design of insulators,
insulation co-ordination studies and overvoltage protection.
This includes also the lightning protection of transmission
line as well as substation and transmission line grounding
analyses. He is presently self-employed.
Siu Ki Lau obtained his BEng and MSc degrees in
Electrical Engineering from the Hong Kong Polytechnic
University, and received his MBA degree from Heriot-
Watt University. He has been taking up various positions in
generation maintenance, transmission resources planning,
power system network planning, special projects, equipment
development and asset utilisation for more than 30 years
in CLP Power. He led various improvement projects, new
development and smart grid projects and introduced new
type of equipment, new systems, and operating schemes
to improve the asset performance in the aspects of quality,
safety, cost and reliability throughout the life cycle of
the asset. He looks after the utilisation of overhead line
assets from LV to EHV in the power company. He set the
design standard and specification, devised the policy and
strategies to manage the overhead line asset replacement,
refurbishment and retirement decisions. He is a Chartered
Engineer and member of the HKIE and IET.