Volume N°7, February 2017

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

"Best of" Paris Session 2016

Using the flow-based method for network investments

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

 Volume N°7, February 2017

Journal edited by CIGRE

President Robert STEPHEN

Technical Council Chairman Mark WALDRON
Secretary General Philippe ADAM / philippe.adam@cigre.org

Editorial Committee
Chief Editor Konstantin O. PAPAILIOU
Editors Pouyan POURBEIK
Nikos HATZIARGYRIOU
Rob STEPHEN

Copyright registration N° 007, February 2017

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

Submission Procedure http://www.cigre.org/Menu-links/Publications/CIGRE-

Science-Engineering
 Dear readers,
You have in your hands the
7th issue of CIGRE Science &
Engineering which traditionally
includes the “Best of ” papers of
the Paris 2016 Session. Now let me confess to you,
that this “exercise” was born two years ago out
of necessity! = At the time, there were no paper
submissions -remember this was a brand new
Journal- and we were eager to kick-off. So the
idea was born to ask the Study Committee Chairs
to select the three best papers from the papers
approved for the Session and the editorial board
would select the best of the best and publish it in
CSE. And so we did!
What had started as a convenient solution
became a “hit” with the authors of the Session
papers eagerly hoping to “score” and the Journal
readers equally eagerly looking forward to seeing
the results of this unwritten competition. No
wonder; this selection process has resulted in some
outstanding papers, like the already stringently
selected papers which make it to the Session (firstly
by the respective National Committee, then by the
Study Committee Chair and lastly, but not least, by
the Technical Council Chairman) only one paper
per SC is selected for publication. If I have got my
math right, this comes to an acceptance rate of 1%!
For proof, just take a look for yourself at the
contents of this issue. You will find some remarkable
papers on ‘Evaluation of the thermal performance
of transformers’, ‘Operating experience with and
future challenges for coordinating power plant and
power system protection’, ‘Solid-gas insulation in
HVDC gas-insulated system’ and many others.
Cigre Science & Engineering • N°6 October 2016
3
Not only that, In this issue we have included
for your benefit a paper we believe deserves to
override our internal rule, that only “Best of ”
papers should be included. I am talking about
the paper on ‘Using the flow-based method for
network investments’ evaluation – a new concept
by Dr. Susana de Graaff, Chair SC C2, and her
co-authors on a subject we have judged to be so
innovative and eagerly anticipated by the scientific
community, that we did not want to wait until our
next issue in June 2017 in order to publish it. I am
sure you will understand and value this decision.
Before closing let me make a personal statement:
I believe any technical body, like CIGRE, can
consider itself privileged, when their top officers
are so technically “fit” that they qualify as authors
in some of the “Best of ” papers. I am talking about
the contributions of our President Dr. Rob Stephen
(‘Calculation accuracy of high-temperature sag
for ACSR in existing lines’) and our Chairman of
the Technical Council Mark Waldron (‘Reduction
of greenhouse gases in GIS pilot project in UK’).
Bravo!
Konstantin O. Papailiou
Chief Editor
konstantin@papailiou.ch
Contents of this issue

page

Editorial

Improved generator performance with

a nanocomposite high voltage insulation system for stator windings – A status report 5
Evaluation of the thermal performance of transformer windings by numerical
investigations and measurements 15
Arresters with advanced cooling performance for protection of valves
in HVDC converters 23
Development of the riser cable system for offshore floating wind power project 32
Calculation accuracy of high-temperature sag for ACSR in existing lines 39
Current situation and recent challenges in asset management of aging
T&D substation facilities in Japan 49
Brazilian experience regarding interactions between series capacitors and svcs –
Main challenges of Tucuruí-Macapá-Manaus interconnection project 55
Operating experience with and future challenges for coordinating power plant
and power system protection 63
Managing regional security of supply: a case study from Scotland 81
Operational security challenges and tools for a synchronous power system
with high penetration of non-conventional sources 91
Reduction of greenhouse gases in GIS pilot project in UK 102
Amplification of harmonic background distortion in wind power plants
with long high voltage connections 109
Capacity mechanisms: Results from a world wide survey 117
An adaptive protection infrastructure for modern distribution grids with
distributed generation 125
Solid-gas insulation in HVDC gas-insulated system: Measurement, modeling
and experimental validation for reliable operation 133
Classification of customers based on temporal load profile patterns 143
Using the flow-based method for network investments’ evaluation – a new concept 149

Cigre Science & Engineering • N°6 October 2016

4
Improved generator performance with
a nanocomposite high voltage insulation
system for stator windings – A status report
Thomas Hildinger
Voith Hydro, Brazil
Peter Gröppel, Manuel Weil, Jürgen R. Weidner*
Siemens AG, Germany
Abstract
The flexible peak load operation of conventional power
plants raised from fast growth of renewable energy
request for improved generator design solutions. As the
stator winding is a key component of hydro and turbine
generators, a fundamental research program had been
started with spherical SiO2 nanoparticles to improve
the electrical properties of stator winding insulation.
Partial discharge and electrical treeing tests showed best
voltage endurance when using a well dispersed amount
of 20% wt. of nanoparticles in epoxy resin matrix. A new
retardation mechanism of nanoparticles had been found,
resulting in longer electrical life.
The new nanocomposite insulation system was tested at
tube-like specimens prepared by the same manufacturing
process as the original stator winding. Compared to the
standard system, the new nanocomposite shows an electrical
lifetime curve with flatter sloop resulting in longer lifetime
by one decade at operational electrical stress.
First voltage endurance tests with original stator bars
(hydro generator) result in significant longer lifetime
compared to the actual insulation system. Additional
tests at different voltage level have to be performed to
statistically ensure the better lifetime performance.
1. Introduction
The demand for new highly flexible power stations in
combination with rising amount of renewable energy
and in parallel the request for upgrading of existing units
with life time extension incite the generator engineering
community to come up with improved generator design
*Juergen.R.Weidner@siemens.com
KEYWORDS
Electrical Lifetime, Hydro- and Turbo-Generators, Silica-Nanoparticles, Stator Winding Insulation System, Treeing
Retardation
Cigre Science & Engineering • N°7 February 2017
5
solutions. On the other hand, the deregulated and
liberalized energy market leads to cost pressure and high
operational reliability on the power plant assets of the
utilities. This motivates the manufacturer to develop
a new generator portfolio with higher efficiency, high
reliability at all kind of operation regime, maximum
availability, adequate maintenance intervals and much
lower operational lifetime costs.
As the high voltage stator winding is the key component
for each hydro and turbine generator a fundamental
research program had been started years ago to utilize
nanotechnology to improve the electrical and mechanical
properties of the high voltage main insulation system.
It is well known that the different thermal, mechanical
and ambient ageing stresses during flexible long term
operation weaken the compound insulation structure and
reduce the overall electrical strength of the insulation
system. The electrical treeing mechanism is the main
propagation process that finally leads to a breakdown of
a high voltage winding insulation [1].
The significantly higher electrical field at small edges or
protrusions of copper strands, which form all together
the inner conductor of the winding, could generate a
first tree initiation. Because of unavoidable local field
enhancement or small delamination voids, the electrical
tree propagates through the epoxy-mica insulation system
along the resin path of the mica tape insulation system as
shown in the schematic microscopic pictures of Figure 1.
The inorganic mica layer acts as a strong barrier with high
electrical strength. Therefore the small treeing channel
of about 10 µm cross section follows the resin interface
between mica layer and carrier material [2].
 Figure 1: Propagation of small treeing channels in epoxy-mica insulation system [2]

Voltage endurance tests at stator winding bars with
vacuum impregnated epoxy-mica insulation system
showed that the dielectric strength increases with mica
content up to an optimum [3]. The analysis of many
break down channels demonstrated clearly that the
treeing path at the resin to mica interface should be
as long and narrow as possible to get a high electrical
break down strength in the insulation sleeve [1, 4]. Or
with other words, to get a long electrical life time of the
insulation system.
In addition to these well-understood design measures,
a significant retardation and lengthening of electrical
tree propagation through the main wall insulation
could be reached by using specially treated inorganic
SiO2 nanoparticles within the resin matrix as shown in
Figure 2. This paper presents the positive influence of
nanoparticles on the performance of high voltage main
insulation for generator stator windings stressed by
operation at highly flexible grid demands. It is an actual
status report of the ongoing R&D project, which is still a
step away from commercialisation of the new improved
nano insulation system.
2. Preparation of nanoparticle
loaded epoxy resin
The nanocomposite processing technology has a
tremendous influence on the electrical and mechanical
performance of the insulation system to be developed
into an industrial product [5]. Therefore high attention
was given to find the adequate processing method for the
epoxy nanocomposite. The nanoparticles were prepared
as alkaline aqueous colloidal silica solution by so-
called water-glass process starting from sodium silicate
chemistry. The Silica-sols were optimized for organic
surface treatment and for maximum stability in organic
Epoxy type systems. Utilizing the correct particle and
surface treatment chemistry is very important for long
term stability with regards to changes in viscosity and
optimized average particle size.

Figure 2: Delay mechanism of tree propagation throughout the main wall insulation by using small
inorganic SiO2 nanoparticles within the epoxy resin

Cigre Science & Engineering • N°7 February 2017

6
Maintaining a well dispersed colloidal silica/resin system
helps to maximize cured system performance with regards
to physical properties including: reducing coefficient of
thermal expansion, increasing E-modulus and tensile
strength and improving resistance to electrical erosion
caused by high electrical fields and partial discharge activity.
Methylethylketon MEK Silica-sols were incorporated
into Epoxy/Anhydride resin systems followed by MEK
solvent removal by Thin Film Evaporation. The Thin
Film Evaporation process is useful to quickly produce
resin-silica composite systems with very low residual
solvent content while minimizing the heat exposure to
the system, which improves shelf life of the uncured
resin system. This is very important for using this
nano-silica/epoxy system in a typical Vacuum Pressure
Impregnation (VPI) process of taped mica insulation
systems for generator windings. Properly prepared
liquid resin nanocomposite systems can remain stable to
viscosity changes for many months up to years, which is
a precondition for the VPI process of electrical insulation
systems at stator windings for large rotating machines.
Nanocomposites used in this R&D project were
prepared by mixing silica nanoparticles with particle Figure 3: Typical test specimen with needle-plane electrode configuration (top)
diameters between 5 nm and 20 nm dispersed in an and lifetime tL of non-filled (oEP) and 26wt.% nano-SiO2 filled (n26)
epoxy resin stressed at 16 kV (bottom)
organic solvent with a bisphenol-diglycidylether
resin. The epoxy-nanocomposite was mixed with a
methylhexahydrophthalic anhydride as hardener for the The comparison of treeing structure for samples with
VPI production of original stator winding bars with the pure epoxy resin and with 26 wt.% nano SiO2 filled
new nanocomposite insulation system. epoxy show an increase of number and dispersion of
tiny tree channels at samples with nano-particles, but
3. Investigation on treeing a significant decrease of tree size and long direct path
to ground electrode. To get more information about
process retardation the different growth structure of treeing channels
The electrical treeing behaviour of epoxy resin needle-plane specimens had been cut and the internal
specimens with SiO2 nanoparticles were investigated wall surface of the small channels of 6 – 15 µm cross-
by several researchers [6 - 7]. Typical needle-plane test section had been examined by using a scanning electron
configurations were used to obtain an extreme high microscopy (SEM). Figure 4 presents an example of
electrical field enhancement to start the treeing process internal wall surfaces in the sub-µm range taken from
pure and nanoparticle filled epoxy.
(Figure 3 top). The lifetime tL until electrical breakdown
was recorded and plotted in a Weibull distribution At pure epoxy specimen the channel has a smooth
probability diagram (Figure 3 bottom). It can be seen cylindrical shape with a fleecy wall structure like ash
that the electrical lifetime tL of the set of specimens with deposits in a chimney (see Figure 4 left). This could
nanoparticles (n26) is more than one decade longer than be deposits of sublimated resin and carbon which were
for the set of specimens with pure epoxy resin (oEP). produced by partial discharge activity when expanding

Cigre Science & Engineering • N°7 February 2017

7
 Figure 4: SEM photos of treeing channels showing wall surface of pure epoxy
(left) and SiO2 nanocomposite n26 (right)

the tree channel by erosion of epoxy resin. This carbon
coating could explain the black coloration of the broad
main treeing channel. Beneath this fleecy coating,
the treeing wall has a smooth surface without any
macroscopic structures.
The channel surface at the nanocomposite material is
covered with nanoparticles that are half duged in the
wall surface (see Figure 4 right). The SiO2 particles act
as an inorganic shield like mica does and avoid direct
erosion of polymeric epoxy by PD attack in the channel.
Furthermore there are µm-size agglomeration structures
at the inner wall side, formed after permanent PD activity.
The sequence of PD erosion at epoxy nanocompesites is
demonstrated step-by-step in Figure 5 which compares
the SEM pictures (right) with sketches of nanoparticle in
an epoxy matrix (left).
More details on investigation and interpretation of
treeing phenomena are given in [7] and [8].

Figure 5: Sequence of PD erosion mechanism at epoxy nanocomposite

insulation surface with retardation effect due to nanoparticle sintering

Cigre Science & Engineering • N°7 February 2017

8
 Figure 6: Six winding tubes with grounded OCP (black) and grading system (grey ECP) prepared for high voltage endurance testing
4. Performance of epoxy-mica
stator winding insulation with
nanoparticles
Very first results with nanoparticle loaded epoxy-
mica insulation system at original winding coils were
achieved by a voltage endurance test at small stator bars
with thin insulation for motors of 6.6 kV rated voltage
only. Because of the large scatter in lifetime results
further investigations had to be carried out at better
defined tube-like test samples. A more uniform taping
condition and a quasi-homogeneous electrical field
distribution allows a better analysis of electrical strength
and lifetime characteristics of the new nanocomposite
insulation system.
4.1. Fundamental lifetime investigation on tube
samples with taped insulation
4.1.1 Test specimen
For fundamental lifetime investigation test specimen
with a round profile (tube) had been fabricated and
voltage endurance tested. All test tubes were taped with
the same standard fine mica tape which is also used for
original stator coils of rated voltage Un = 11 kV. The
tube specimens had been vacuum pressure impregnated
(VPI) with epoxy resin and cured according to the
(a) VEI-Factor: 2.0 @ 3-times Un
Figure 7: Weibull distribution plot of voltage endurance test results of MICALASTIC system and new nanocomposite system at constant voltage level of (a) 3.0Un,
Cigre Science & Engineering • N°7 February 2017
original manufacturing process. The table in Figure
6 summarizes the design features of the tube-like
insulation test specimens, which were produced in
different batches with pure epoxy resin and with SiO2
nanoparticle loaded epoxy resin. After curing process
an outer corona protection layer (black OCP) had been
applied as ground electrode. An end corona protection
system (grey ECP) avoided surface discharges during
high voltage endurance testing at acceleration levels
(Figure 6).
4.1.2 High voltage endurance tests
To compare the electrical strength and lifetime
characteristics of the established reference system with
the new nanocomposite insulation system a set of voltage
endurance tests with three different stress levels of 2-,
2.5- and 3-times of rated voltage had been performed at
room temperature.
The electrical lifetime tL distribution of the different
collectives of stressed winding tubes is given as Weibull
Probability Plot in Figure 7. At all three tested voltage
levels the new SiO2 nanocomposite insulation improves
the lifetime compared to the established MICALASTIC
insulation.
The Voltage Endurance Improvement (VEI) Factor is
defined as the ratio of the 63% mean value of Weibull
distribution between MICALASTIC test collective and
(b) VEI-Factor: 5.1 @ 2.5-times Un (c) VEI-Factor: 14.6 @ 2-times Un
(b) 2.5Un and (c) 2.0Un
9
nanocomposite specimens. It is shown in Figure 7 (a)
– (c) that the VEI-Factor depends on the applied test
voltage level. The VEI-Factor increases from a value of
2.0 at 3.0Un up to a value of 14.6 at 2.0Un. The lower the
applied voltage the bigger is the VEI-Factor.
4.1.3 Electrical lifetime curves
In Figure 8 the mean values (Weibull 63%) of the three
different voltage endurance test are plotted in a semi-
logarithm scaled diagram of electrical field strength
ED versus lifetime tL. The two lifetime curves clearly
demonstrate that the improved erosion stability of new
insulation with SiO2 nanoparticles leads to significantly
increased lifetime.
Figure 8: Comparison of electrical life time curve ED = f (tL) of tubes with
established MICALASTIC and new nanocomposite insulation system; rated
winding field strength Er shown as a reference
The different slope of the two lifetime curves results
in an estimated crossover point if an extrapolation of
the measured values is made to higher electrical field
strength. At this high field strength both systems would
perform similar. The extrapolation to lower field strength
in the area of usual operation field strength Er of the
stator winding leads to significant longer lifetimes up to
a calculated VAI-Factor above 100.
The different slope of lifetime curve for the new
nanocomposite insulation compared to established
insulation systems with a crossover point at high electric
fields and much longer lifetime at lower field strength
is in good coincidence with other research work done
on nanocomposite varnish for motors [9]. In [8] the
Cigre Science & Engineering • N°7 February 2017
10
lifetime and electrical aging mechanism for epoxy
nanocomposites had been investigated with needle-
plane specimens. The voltage endurance characteristic of
these tests is given Figure 9 and it clearly demonstrates
the different lifetime slops with a crossover point at very
high electrical fields.
Figure 9: Voltage endurance characteristic of neat epoxy resin and SiO2
nano-composite at treeing test with crossover point (taken from [8])
At technical important low electrical field strength the
degradation process of partial discharge activity together
with the step-by-step treeing growth is predominant.
Therefore it can be concluded that the aging of the new
nanocomposites follows a different, retarded degradation
mechanism compared to neat epoxy resin [8].
Basic R&D work performed at the Nanodielectrics
Laboratory in Bangalore, India [10] came up with
voltage endurance tests at platelet samples, which were
stressed at 60°C with a homogeneous electrical field
of 6 kV/mm, 7 kV/mm and 8 kV/mm, also verified the
different slope of lifetime curves of neat epoxy resin and
5 %wt. nanocomposite material. The nanocomposite
samples have a more flat slope compared to neat epoxy
resin with about nine times longer lifetime at design
stress of motors.
With the information from other researchers and based
on the own test results it is unambiguously proved that
the Voltage Endurance Improvement (VEI) Factor at the
design field strength of stator winding insulation would be
higher than a factor of ten. Nevertheless this needs to be
proved on original generator stator bars in the next steps.
 Figure 10: Voltage endurance test at 1.6 m long stator winding bars with the original cross
section of an indirect cooled hydro generator of Un = 13.8 kV

4.2 Voltage endurance tests at original hydro
generator stator winding bars
4.2.1 Test specimen
Forty stator winding bars with the original cross
section of an indirect cooled hydro generator of rated
voltage Un = 13.8 kV and 1.6 m straight bar length were
manufactured with standard fine mica glass fabric tapes
and original hot curing VPI process called MICALASTIC
insulation system. For voltage endurance comparison
tests different sets of bars with pure epoxy resin and with
new nanoparticle loaded epoxy resin were produced.
The photo in Figure 10 gives an impression of the high
voltage test cell with a set of these bars connected to
high potential of 2.5 times rated voltage for an electrical
lifetime (voltage endurance) test. Due to the high AC
test voltage of 2.5Un = 34.5 kV an accelerated cooling of
the grading system (grey varnish) with fans is needed to
avoid overheating and electrical flash over.
The dissipation factor measurement, which had been
used as a quality measure for VPI production process,
resulted in the same low tip-up values for the bars
with standard MICALASTIC insulation and with new
nanoparticle loaded insulation system. This clearly
demonstrates the good impregnation of the thicker
insulation of original 13.8 kV stator bars. All tested
bars had a maximum dissipation factor tip-up below
1.5·10-3/0.2UN. At the new international standard IEC
60034-27-3:2015 the tip-up limit is set to a much
higher value of 5·10-3/0.2UN.

Figure 11: Weibull probability plot of voltage endurance tests at RT with standard insulation system MICALASTIC and new
nanocomposite insulation system at 3-times and 2.5-times rated voltage

Cigre Science & Engineering • N°7 February 2017

11
 Figure 12: Comparison of electrical lifetime curve ED = f (tL) of existing MICALASTIC system and new
nanocomposite insulation system applied on original hydro generator stator winding bars for rated voltage
Un = 13.8 kV
4.2.2 Voltage endurance test
The results of the electrical endurance tests at
different voltage level are given as Weibull probability
distribution plots in Figure 11. The Voltage Endurance
Improvement (VEI) Factor verifies the above at tube
insulation specimens described nanoparticle effect of
higher lifetime improvement at lower test voltage. The
VEI-Factor increases from VEI = 2.8 at 3Un = 41.4 kV
up to VEI = 4.3 at 2.5Un = 34.5 kV. The 2Un tests with a
sufficient number of specimens are still ongoing because
of long lifetime at the lower test voltage.
4.2.3 Electrical lifetime curves
Nevertheless a first electrical lifetime curve ED = f
(tL) of original hydro generator stator winding bars
produced in VPI insulation technique with the existing
MICALASTIC system and with the new nano-composite
resin can be drawn in Figure 12. The insulation system
of the bars is designed for a rated line-to-line voltage of
Un = 13.8 kV.
To establish the different slope of lifetime curves of
original stator bars additional voltage endurance tests at
four different voltage levels (field strength ED) of 2Un,
2.5Un, 3Un and 4Un with sample collectives of minimum
7 test bars are needed. At all a production contingent of
about 70 original hydro generator stator bars has to be
manufactured and life time tested.
5. Next steps of development
After successful verification of enhanced electrical
lifetime with nanocomposites at hydro generator bars
by more than one decade, if extrapolated to operational
Cigre Science & Engineering • N°7 February 2017
12
field strength of original stator windings, it is mandatory
to qualify the new nanocomposite insulation system
according to international standard IEC 60034-18 series
[11].
Therefore the next steps of this development project will
be:
1. Transfer the qualification test program also to full-
size generator bars with a main insulation up to rated
voltage of Un = 22 kV and a straight bar length of 3
- 6 m depending on generator type. The qualification
procedure of the new candidate insulation system
compared to the well proved and established
insulation system has to be performed according
to IEC 60034-18 series [11] dealing with a set of
functional evaluation tests for insulation systems
of rotating electrical machines. This qualification
program includes the following expensive and time
consuming tests:
- Thermal aging and classification tests at three
different temperatures,
- Electrical aging – voltage endurance tests at a
minimum of three different levels resulting in
lifetime curves,
- Multifactor aging by applying thermal cycle and
electrical stress in parallel to the main insulation,
- Thermo-mechanical bending endurance in parallel
with electrical stress as a multifactor aging test.
2. If the new candidate insulation system with
nanocomposite confirms the improved performance,
then an indirect air cooled generator will be equipped
with a new designed stator winding using the much
thinner, better thermal conductive and more flexible
 Figure 13: Road map of R&D project on nanocomposite high voltage insulation system for stator windings of hydro
and turbine generators
insulation system with SiO2 nanoparticles. It is
expected that the new nanocomposite insulation will
increase the overall generator performance (MVA) of
about 10%.
3. To better qualify the stator winding for new volatile
grid demands generated by renewable energy plants a
basic research project has been started to use special
elastic nanoparticles to make the main insulation
system and the winding structure more flexible and
to improve dynamic damping characteristics.
The described long term R&D project dealing with the
basics in defining applicable nanocomposites usable for
high voltage insiulation of stator windings is driven by a
well structured technology development process. This step
by step revied R&D project leads to enhanced performance
of stator windings in hydro and turbine generators facing
the severe stresses of new flexible grid demands. Figure
13 shows the road map of the project, which started with
basic research on different kind of nanoparticles in 2006
and is now focused on functional testing of original stator
winding bars before the last step of prototype production
and fleet introduction could be fulfilled.
6. Conclusions
Based on the promising results of basic material
investigations and further on of insulation system testing
at generator stator bars the following conclusions can be
drawn:
(1) The nano-sized SiO2 particles act like a barrier to
the treeing process and hold up the treeing channel
propagation. Obstruction of the tree growth results in
a significant longer lifetime until breakdown of the
Cigre Science & Engineering • N°7 February 2017
13
insulation and in a much higher electrical strength of
the epoxy-mica insulation system for large rotating
machines.
(2) The voltage endurance tests carried out on tube-
like insulation system models and original hydro
generator stator bars show a great improvement in
the electrical lifetime of the new nanoparticle based
high voltage insulation system. The new insulation
system would allow a more efficient generator
stator winding design with increased performance in
terms of MVA by about 10%. The nanocomposite
insulation system will be more robust to peak load
operation at highly volatile grid demands with load
ramps up to 100 MW/min in the near future.
(3) The newly developed nanocomposite insulation
system offers a more effective design of stator
windings. It will result in a smaller thickness of the
insulation layer, leading to more space for copper
material and a better heat transfer. This improves
the specific power output and lifetime of new and
existing (after rewind) hydro and turbine generators.
7. References
[1] R. Vogelsang, “Time to breakdown of high voltage winding
insulation with respect to microscopic properties and manufacturing
qualities”, PhD dissertation, ETH Zurich, Hartung-Gorre Verlag
Konstanz, 2004.
[2] R. Vogelsang, K. Fröhlich, B. Fruth, “Detection of electrical tree
propagation in generator bar insulations by partial discharge
measurements”, 7th Intern. Conf. on Properties and Applications
of Dielectric Materials, ICPADM June 2003, Nagoya, Japan, pp.
281 – 285.
[3] A. Wichmann, “Reliability and Testing of High-Voltage Stator
Insulation for Large Rotating Machines”, IEEE Winter Meeting
1972, New York.
[4] T. Baumann, T. Hillmer, P. Klee, D. Stoll, “Optimized, Emission
free Production Process for Main Wall Insulation with Enhanced
Lifetime Properties”, EIC 2011, Annapolis, USA.
[5] T. Tanaka, G. C. Montanari, P. Mülhaupt, “Polymer Nanocomposites
as Dielectrics and Electrical Insulation-perspectives for Processing
Technologies, Material Characterization and Future Applications”,
IEEE Trans. on Dielectrics and Electrical Insulation, Vol.11, No.5,
Oct. 2004, pp. 763 – 784.
[6] T. Tanaka, A. Matsunawa, Y. Ohki, M. Kozako, M. Kohtoh, S.
Okabe, “Treeing Phenomena in Epoxy/Alumina Nano-composites
and Interpretation by a Multi-core Model”, IEEJ Trans. FM,
Vol.126, No.11, 2006, pp. 1128 – 1135.
[7] M. G. Danikas, T. Tanaka, “Nanocomposites – A Review of
Electrical Treeing and Breakdown”, IEEE Electr. Insul. Magazine
Vol.25, No.4 July/Aug. 2009, pp. 19 – 25.
Cigre Science & Engineering • N°7 February 2017
14
[8] T. Tanaka, “Similarity between Treeing Lifetime and PD Resistance
in Aging Mechanisms for Epoxy Nanocomposites”, Proc. 9th Intern.
Conf. on Properties and Applications of Dielectric Materials, July
19-23, 2009, Harbin, China, Paper H-3.
[9] F. Pohlmann, “Isoliersysteme für Drehstromniederspan-
nungsmotoren, die von Pulsfrequenzumrichtern angesteuert
werden”, PhD dissertation (in German), Fortschritt-Berichte VDI:
Reihe 21, Elektrotechnik 317, XIII, 130 Seiten Düsseldorf: VDI-
Verl. (2001).
[10] P. Preetha, M. J. Thomas, “Life Estimation of Electro-thermally
Stressed Epoxy Nanocomposites”, IEEE Trans. on Dielectrics and
Electrical Insulation, Vol.21, No.3, June 2014, pp. 1154 – 1160.
[11] Rotating electrical machines - Functional evaluation of insulation
systems, IEC 60034-18-Series
Evaluation of the thermal performance
of transformer windings by numerical
investigations and measurements
S. Tenbohlen1*, C. Breuer2, F. Devaux3,
R. Lebreton3, N. Schmidt1, T. Stirl2
1
University of Stuttgart, Germany
2
GE Grid Solutions, Germany
3
GE Grid Solutions, France
Summary
Thermal behaviour of an oil-cooled transformer winding
is presented by means of an experimental and a numerical
CFD winding model. The experimental model comprises
two passes of a zig-zag cooled disc type winding. A
2D and a 3D model, resembling both the experimental
winding geometry were created in order to investigate
the implications resulting from a 2D simplification of the
winding geometry. Because of the thermal dependency of
the oil properties, the laminar character of the oil flow is
being affected at high oil temperatures and oils speeds.
For that reason, the presented numerical investigations
also include the application of turbulence models. The
2D simplifications result partly in an overestimation
of the predicted conductor temperatures. With regard
to the 3D CFD results, a very good agreement with the
experimental findings throughout the entire spectrum of
investigated operating conditions can be observed. The
presented results give a deep insight into the oil flow and
temperature behavior of OD cooled windings enabling
the designer to optimize the cooling of power transformer
windings.
1. Introduction
Power transformers are key components applied in
transmission and distribution systems. Their power
rating and life-cycle characteristics are strongly
dependent on thermal aspects. Although today’s
transformers are highly efficient, their cooling systems
still have to discharge non-negligible amounts of heat
dissipated in their windings. The higher the evolving
*stefan.tenbohlen@ieh.uni-stuttgart.de
KEYWORDS
Hot Spot Temperature. Oil Flow Rate, Power Transformer, Thermal Modelling
Cigre Science & Engineering • N°7 February 2017
15
temperature levels at a given loading rate are settling,
the faster the insulation materials will age at that loading
rate. Moreover, a thermal failure results not from the
overall condition of the insulation but from the most
degenerated part of the insulation caused by the highest
local temperatures. Therefore, a profound knowledge
about the temperature distribution inside a transformer is
crucial for an appropriate assessment of the component’s
power rating. Consequently, an efficient thermal
design offers the opportunity to significantly cut costs
in the production process and to improve the thermal
performance [1].
For the investigation of the temperature distribution
within the windings of an oil cooled transformer,
different approaches can be pursued [2]. Applying
temperature sensors directly within the windings of real
power transformers proves to be a fairly limited approach
to analyse the temperature distribution in detail during
operation. Aside from monetary aspects connected to
the sensors and their respective assembly, additional
disadvantages result from the connected measurement
accuracy of suitable sensors and the limited control
of prevailing boundary conditions. Next to direct
measurements, empirical correlations are commonly
applied within the design process of power transformers.
While these approaches are an appropriate way to tackle
daily business in the design process, their potential to
improve the thermal design and therewith cut costs for
the manufacturers is rather limited.
For that reason, this contribution focuses on winding
models derived from a specific winding design with
characteristics that are representative for a disc winding.
 Figure 1: Basic design characteristics for disc type transformer windings [3]

Starting from that design, both an experimental and
two numerical models are created, allowing a detailed
investigation of the temperature distribution inside the
modelled disc winding. While measurements, assuming
a well-controlled experimental setup with negligible
measuring errors, offer a dependable and direct insight
into the investigated physical phenomenon, they are
also afflicted with several substantial disadvantages.
Considering a strongly limited model flexibility
alongside substantial time requirements and high costs,
the development of alternative investigation concepts
appears increasingly advisable.
Consequently, two numerical winding models based on
computational fluid dynamics (CFD) were created. Due to
the considerable computational requirements connected
to CFD analysis, most CFD-based investigations of
transformer windings apply a reduction of the winding
geometry to a two-dimensional (2D) domain. While
this reduction appears to be a suitable strategy to
keep the connected computational efforts within the
operating range of a typical desktop workstation, this
contribution specifically investigates the implications of
this simplification. For that reason, a two dimensional
(2D) and a three dimensional (3D) winding model are
investigated alongside the experimental analysis.
2. Development of an
experimental model for a disc
winding
The design of disc type transformer windings shows a
strong symmetry in circumferential direction. To display
this design characteristic more clearly, Figure 1a)
offers a three dimensional view on a disc type winding
with four conductor turns per disc. For reasons of an
improved visibility concerning the cooling system inside
the winding, the outer cardboard cylinder is displayed in
a partly transparent color. The given enlargement shows
the horizontal cooling channels created by the opening
between two consecutive spacers and discs. The vertical

Figure 2: Design characteristics of the modelled winding geometry and the winding model [3]

Cigre Science & Engineering • N°7 February 2017

16
 Figure 3: Components of a conductor model for the experimental winding model [3]
cooling channels are bounded by the cardboard cylinders
and by the sticks aligning the spacers. Consequently,
the entire winding geometry can be subdivided into
symmetrical sections, as separately displayed in
Figure 1b). To take advantage of this geometrical
symmetry for the heat and mass transfer analysis, the
following premises have to be met.
• Homogeneous inlet conditions below the winding
• Homogeneous exit conditions above the winding
• Evenly distributed electrical losses within each
conductor turn
Although not every transformer design fulfills these
premises, they are assumed to be valid within this
contribution. Consequently, all windings models
investigated below represent one symmetrical section of
a disc winding.
As carried out in the introduction, a winding design
with characteristics being representative for an OD-
cooled disc winding is chosen as a common basis for all
investigated winding models. A top view on a symmetric
section of this chosen winding design, including all
relevant geometric details to assess this design, is
shown in Figure 2a). Since the exact reproduction of
this geometry would pose enormous challenges for
the connected manufacturing processes, the respective
design of the experimental model was modified. For
that purpose, the curvature of the conductors in
circumferential direction is neglected, as displayed in
Figure 2b).
Furthermore, the symmetry of the horizontal cooling
channels and spacers is forfeit. The reason for this
modification results from the material choice for the
smaller and rectangular spacer shown in Figure 2b).
To allow a future optical investigation of the oil flow
distribution with the experimental winding model,
these components were made of glass. Due to the far
more complex manufacturing processes connected to
that material, a rectangular shape was chosen as a cost
effective compromise. Nevertheless, both geometries
given in Figure 2a) and Figure 2b) share the identical
opening widths of the horizontal channels and a close
Cigre Science & Engineering • N°7 February 2017
17
to identical amount of conductor surface available for
heat exchange (marked green) and conductor surface
covered by spacers (marked red). Therefore, the thermal
and hydraulic impact of the design modifications is
considered negligible.
Finally, the paper wrapping typically applied around
conductors in transformer windings for purposes of an
electrical insulation is substituted by solid blocks of
plastic (Polyvinylidenefluoride) located only in between
the conductors of a disc. Prior experimental investigations
showed a great sensitivity of the evolving temperature
levels on the exact thickness of the applied paper
wrapping. Unfortunately, this property, especially with
regard to the comparably short experimental conductor
models, is difficult to keep within small tolerances. To
eliminate respective detrimental influences but still
allow a realistic thermal decoupling of the conductor
models in one disc, the paper is substituted, as illustrated
in detail in Figure 2c).
The design of the single conductor models, arranged
into discs in the further assembly process, is displayed
in Figure 3. To provide the heating power normally
dissipated in the winding turns of a power transformer
during operation, heating cartridges are injected into
through bore-holes. An accurate measurement of the
individual conductor temperatures is accomplished with
temperature sensors located inside a second drilling, as
indicated in Figure 3a).
The cooling system of OD-cooled disc windings is
usually subdivided into so-called passes in axial direction.
By periodical placement of washers inside the vertical
channels alternating between the inner and outer winding
diameter, two different types of passes result. Therefore,
the experimental winding model must consist of at least
two consecutive passes. In addition, representative
entering and exiting conditions are required before and
after the investigated passes. This ensures thermal and
hydraulic conditions that are representative for every
pass throughout the complete cooling system of the
modelled winding. Integrating these considerations,
Figure 4a) shows the chosen winding model layout in a
 Figure 4: Components of a conductor model for the experimental winding model [3]

schematic view. To its right, Figure 4b) provides a clear
reference for the pass, disc and conductor numbering the
presented post analysis relies upon. Finally, Figure 4c)
contains a photo of the experimental winding model
during operation. Although this picture gives no detailed
insight into the chosen mechanical design of the winding
model, comprising its sealing and insulation concept, it
already offers a first impression of the required technical
efforts.
Next to a representative winding design, also the
operating conditions of the modelled winding are
chosen accordingly with respect to the winding type and
dimensions. To control the relevant boundary conditions
precisely, a laboratory setup schematically illustrated in
Figure 5 was applied. It allows an accurate control over
the mass flow rate , the oil temperature at the model
inlet and the heating power of each conductor Plosses, i.
While the oil inlet temperature is set via a controlled flow
heater, the initially provided oil flow rate from a gear
pump is limited to a defined value by a controlled valve
processing data from a flowmeter. To minimize the heat
leakage from the winding model to its surroundings, the
entire model is submerged in an open experimental tank
filled with oil of the same temperature as provided by
the flow heater. Finally, the individual power supplied
to each heating cartridge is set by software controlled
power electronics. Due to the chosen laboratory setup,
a total measuring tolerance concerning the temperatures
of approximately 0.5 K is achieved. Consequently,
temperature gradients can be determined with an
accuracy of around 1 K. The occurring deviations
concerning the mass flow measurement and heating
power control are below 1 %.
3. Development of CFD models
for a disc winding

Figure 5: Piping and instrumentation diagram of the laboratory setup designed to create strictly controlled bound-
ary conditions inside the winding model [3]

Cigre Science & Engineering • N°7 February 2017

18
The development of numerical winding models suitable
for CFD analysis first of all comprises the definition
and creation of respective modelling domains [2]. In
case of a 3D winding model, the common design of disc
windings allows a significant reduction of the winding
geometry in circumferential direction, as carried out for
the experimental winding model described above. In
addition, the symmetric alignment of the spacers inside
a symmetrical winding section, as shown in Figure 1b),
offers potential for further reduction by application of
symmetric boundary conditions, as indicated in Figure 6.
However, since a key purpose of the experimental
winding model is to allow the assessment of the pursued
numerical modelling approaches, the created 3D winding
model corresponds exactly to the domain illustrated in
Figure 2b) and Figure 4a).
The chosen modelling domain of the 2D winding
model, allowing a significant reduction of the connected
computational efforts, requires further explanation.
Since the CFD software applied for this investigation
(ANSYS CFX) does not feature a dedicated 2D solver,
even the analyzed 2D geometry must consist of 3D finite
volumes. However, since only one layer of 3D cells
is necessary, the detrimental consequences resulting
from this requirement are negligible. On the contrary,
modelling the winding geometry with a 2D domain
can even take advantage of this restriction by creating
a 2D modelling domain featuring a non-uniform cell
depth in third dimension. Therefore, the widening of
the horizontal cooling channels from inner to outer
winding diameter, as displayed in Figure 6, can also be
acknowledged in the 2D winding model. The chosen 2D
cell depths represent a defined fraction of the existing
opening widths of the horizontal cooling channels.
Details about this modelling approach for a 2D winding
model are given in [4].
After the described choice of the respective modelling
domains, the creation of the CFD winding models
comprises the discretization process including the
determination of an appropriate meshing density. Next
to the global mesh size, special attention has to be paid
towards the discretization of the boundary layers. This
procedure is also described in [4] and results in a mesh
size of approx. 7·106 fluid cells for the 2D domain and
approx. 200·106 fluid cells for the 3D domain. Due to
Cigre Science & Engineering • N°7 February 2017
19
the connected computational efforts, only the analysis
of the 2D domain should be considered for state of the
art desktop workstations. For a more efficient analysis
following the described approach, high performance
computing centers become advisable. Both numerical
models were defined by an identical solver setup, applying
the SST turbulence model [5] with its transitioning onset
model at operating conditions, where a laminar software
configuration did not lead to converging results.
Figure 6: Modelling domains of the numerical winding models created for
CFD-analysis [3]
4. Results of the experimental and
numerical winding model analysis
This contribution presents the collected numerical
results and measurements for four different operating
points. The respective operating conditions are defined
by the temperature of the oil entering the model, the
oil flow rate passing the winding model and the losses
supplied to each conductor model. However, since the
latter two properties refer strictly to a single symmetrical
section, they are not easy to put into perspective of
common transformer design rules. Since those rules
usually refer to entire windings, the properties given
in this contribution, to reference investigated operating
conditions, are transferred accordingly. For example,
instead of specifying the heating power per conductor
model representing a section of 8 ° in circumferential
direction of on entire winding turn, the corresponding
losses per complete winding turn are referenced.
Analogous considerations are applied for the oil flow
rate. For that reason, the actually applied heating power
per conductor model and the oil flow rates passing the
winding models are 8 °/360 ° = 1/45 times the referenced
values pointed out in the figure captions.
As elaborated above, a set of four representative
operating conditions was chosen for this investigation,
resulting from two oil temperatures at the model inlet
and two oil flow rates. The inlet temperature is set to
ϑin = 40 °C or ϑin = 80 °C while the oil flow rate is set to
oil
= 9 kg/s or oil = 18 kg/s. Taking the minimum
opening width within the investigated cooling
system, maximum velocities of around 1 m/s can
be estimated at 18 kg/s. The losses per winding turn
are kept at Plosses, I = 360 W.
One central objective of this contribution is the
assessment of the pursued numerical modelling
approaches for transformer windings. Since 3D winding
models require significantly higher computational efforts
in comparison to 2D approaches, it is of keen interest
if those additional efforts are justified. To start the
respective comparison between the 2D and 3D winding
model, Figure 7 shows the numerical results for the oil
Figure 7: Comparison of the oil flow velocities within the 2D and 3D CFD winding model at
Cigre Science & Engineering • N°7 February 2017
20
flow distribution inside the winding models qualitatively.
Although no quantitative comparison of the oil amount
passing the different horizontal channels is carried out,
the chosen visualization already gives a clear insight
into the respective differences resulting from the chosen
modelling approach. While the 2D winding model by
definition suggests a homogeneous oil flow distribution
in the circumferential direction, the 3D results clearly
contradict this assessment. Especially the visible oil
flow in the vertical channels of the 3D winding model
displays a strong heterogeneous character. However,
because the oil flow distribution is only of secondary
interest in comparison to the temperature distribution,
Figure 8 gives the results of the respective temperatures
within the investigated winding geometry.
The displayed results in Figure 8 indicate notable
differences in the respective results of the compared
numerical winding models regarding the evolving
temperature distribution. While the 2D calculations
suggest hot spot temperatures around 95 °C at the
referenced operating conditions, the 3D calculations
predict lower maximum conductor temperatures of
approximately 90 °C. In addition, also the location of the
hot spot is dependent on the taken modelling approach.
According to the 2D calculations, the first disc after an oil
washer appears to be most critical at the last conductor
in oil flow direction, while the 3D results identify the
second disc past the oil washer as the disc holding the
hot spot at the last conductor in oil flow direction not
facing the vertical channel.
To better quantify the deviations connected to the
numerical modelling setup, Figure 9 presents the results
for the temperature gradients between oil and conductors
at all investigated operating conditions in both passes
of the winding models. Next to the CFD results, the
shown data also contains the measurements, allowing a
proper assessment of the respective modelling approach.
To calculate the local temperature gradient at a specific
conductor, the oil temperature at the inlet of the pass holding
that conductor is applied. In case of the measurements,
these oil temperatures are not measured but determined
thermodynamically, taking the mass flow rate and the Figure 9: Comparison of the measured temperature gradients with numeri-
specific thermal capacity of the oil together with all cally determined 2D and 3D CFD results in both investigated passes at
losses below the investigated pass into account. The main various oil flow rates and inlet temperatures with specific losses per winding
turn of Plosses, i = 360 W/turn [3]
advantage of the temperature gradients in comparison to
absolute temperatures is the improved comparability of
related results at different inlet temperatures. Taking a close look at the presented measurements for
Pass 1, displayed in Figure 9b), d), f) and h), missing data
from some sensors becomes apparent. In particular, the
second conductor in disc 1, the third conductor in disc
2 and the second and third conductor in disc 5 are not
included in the respective plots. Because the referenced
sensors gave strongly fluctuating signals, they were
found not suitable for further analysis. In addition, also
the sensor inside the fourth conductor of disc 2 in pass
2 displayed a decreased accuracy. However, since the
connected deviations remained within a range of 2 K,
the respective results are retained for post analysis.
With regard to the assessment of the carried out numerical
modelling approaches, the simplifications applied to
accomplish a reduction of the modelling domain into a
2D geometry result partly in a significant overestimation
of the predicted conductor temperatures. This becomes
Figure 8: Temperature distribution determined with the CFD winding
models at
most notable at ϑin = 40°C and oil = 18.0 kg/s as well
as at ϑin = 80°C and oil = 9.0 kg/s with maximum

Cigre Science & Engineering • N°7 February 2017

21
temperature gradients up to twice as high as suggested
by the respective measurements. Also at the other two
investigated operating points, overestimations for the
maximum temperature gradients up to 50 % arise.
With regard to the 3D CFD results, a very good
agreement with the experimental findings throughout the
entire spectrum of investigated operating conditions can
be observed. Deviations are in most of the cases below
1 K. Small deviations between 3D calculations and
measurements only occur at the last conductor in oil flow
direction in each disc of both passes at ϑin = 40°C with
slightly underestimated temperature gradients. However,
due to the primary interest in hot spot temperatures and
a common focus on high oil temperatures, these local
underestimations at low oil temperatures are valued
uncritical. Consequently, the 3D numerical modelling
approach allows an accurate prediction of all relevant
conductor temperatures at operating conditions that are
representative for OD-cooled disc windings. In addition,
the application of 2D winding models might lead towards
a conservative thermal design leaving room for a more
cost effective winding layout.
5. Conclusion
The thermal investigation of an oil-cooled transformer
winding is carried out by means of an experimental and
a numerical CFD winding model. The experimental
model comprises two passes of a zig-zag cooled disc type
winding. Losses are impressed by heating cartridges in
each winding segment. The temperature of each turn is
measured by a precise temperature sensor. The model is
integrated in a laboratory setup, which allows an exact
control of the boundary conditions, e.g. the oil flow rate
and inlet temperature.
A two dimensional (2D) and a three dimensional (3D)
model, resembling both the experimental winding
geometry were created in order to investigate the
implications resulting from a 2D simplification of the
winding geometry. Several conclusions can be drawn
from the comparison of the 2D and 3D numerical
model with the measured temperature data. Because
of the thermal dependency of the oil properties, the
Cigre Science & Engineering • N°7 February 2017
22
laminar character of the oil flow is being affected at
high oil temperatures and oils speeds. For that reason,
the presented numerical investigations also include the
application of turbulence models at operating conditions
where a laminar consideration of the oil flow becomes
questionable.
The location of the hot spot is dependent on the taken
modelling approach. According to the 2D calculations,
the first disc after an oil washer appears to be most
critical at the last conductor in oil flow direction,
while the 3D results identify the second disc past the
oil washer as the disc holding the hot spot at the last
conductor in oil flow direction not facing the vertical
channel. The 2D simplifications result partly in a
significant overestimation of the predicted conductor
temperatures especially at high oil inlet temperatures
and flow rates. The calculated maximum temperature
gradients are up to twice as high as suggested by the
respective measurements. With regard to the 3D CFD
results, a very good agreement with the experimental
findings throughout the entire spectrum of investigated
operating conditions can be observed. The error between
measurement and 3D simulation is below 1 K, which
confirms the validity of the simulative approach in an
excellent manner.
6. Bibliography
[1] A. Tanguy, J. P. Patelli, F. Devaux, J. P. Taisne, and T. Ngnegueu:
‘Thermal performance of power transformers: thermal calculation
tools focused on new operating requirements’. Session 2004, CIGRE,
Paris, 2004
[2] P. Picher, F. Torriano, M. Chaaban, S. Gravel, C. Rajotte, B. Girard,
“Optimization of transformer overload using advanced thermal
modelling,” CIGRE 2010, Paris, A2-305.
[3] N. Schmidt: “Simulation and Experimental Validation of Forced Oil
Flow in Power Transformers” (PhD thesis to be published in 2017 at
the University of Stuttgart)
[4] N. Schmidt, S. Tenbohlen, S. Chen, C. Breuer: “Numerical and
Experimental Investigation of the Temperature Distribution inside
Oil-Cooled Transformer Windings”, 18th International Symposium
on High Voltage Engineering, ISH 2013, Seoul, South Korea, 25.-30.
August 2013, Paper OG1-03
[5] F. R. Menter: “Two-Equation Eddy-Viscosity Turbulence Models
for Engineering Applications”, AIAA Journal 32 (8): 1598–1605,
August 1994
Arresters with advanced cooling
performance for protection of valves
in HVDC converters
R. Göhler*, J. Shaikh, Siemens AG Berlin, Germany
V. Brendler, M. Weyer, Siemens AG Erlangen, Germany
M. Giessel, V. Hinrichsen, TU Darmstadt, Germany

Abstract environment, though. Such non-housed valve arresters

Arresters used in HVDC applications are specifically
dimensioned based on standard arrester types. Arresters
usually consist of an active part and a housing, where the
active part is made of metal oxide (MO) resistors connected
in series and in parallel, and the housing in most cases
to date is a polymer housing. Protective level and design
of an arrester for the different applications have to be
determined such that after an operation with rated energy
stress the arrester will stay thermally stable at continuous
operating voltage. Particularly for the valve arresters in a
line commutated converter (LCC), their protective level
decides about the number of thyristors to be connected
in series. Reduced protective levels will result in a lower
number of thyristors and reduced costs, accordingly.
Depending on the valve configuration in some cases
(with additional measures) a landslide reduction of costs
is possible. Prerequisite for an optimization of thyristor
valve parameters (number of series connected thyristor
levels, snubber circuit parameters) and the valve arrester
is a detailed investigation of the relevant operating
conditions. This has been achieved with improved digital
simulation for evaluation of the energy dissipated in the
valve arrester.
Beneath the properties of the MO resistors like non-
linear voltage-current characteristic and energy handling
capability, the overall thermal behavior of the arrester is
of main importance. Since lower protective levels will
result in increased heating under continuous operating
conditions, improved cooling is a main concern for
the valve arresters. A simple approach is, for instance,
replacing the hollow core insulator of a tube design
arrester with a cage design housing where silicone rubber
is directly attached to the MO resistors. The best solution
in terms of cooling would be an arrester without any
housing, which can only be used in a controlled indoor
have been in service for many years. To even further
improve cooling of the MO resistors, active cooling
elements – metal parts with cooling fins – have been
developed. To fulfill the creepage distance requirement of
14 mm/kV, silicone rubber sheds are pushed directly over
the MO resistors. Geometry of the cooling parts has been
optimized by thermal simulations of a complete arrester
consisting of MO resistors, polymer sheds and metallic
cooling elements. For verification of its improved cooling
behavior, a number of tests were performed on real
size, completely assembled single- and multi-column
arresters. To find the limits of thermal stability and the
lowest possible protective level, the test voltage was
varied around the assumed continuous operating voltage.
The test procedure comprised test voltage application for
several hours up to steady state temperature conditions
or thermal runaway, respectively. During some tests, the
temperature was further increased by a certain amount of
energy input up to the maximum specified temperature
where thermal stability is guaranteed, to demonstrate
cooling at the defined continuous operating voltage.
These tests were carried out at ambient temperatures of
20 °C and 60 °C, the latter taking into account the highest
possible temperature in a valve hall.
Finally, a complete type test according to the new standard
IEC 60099-9, published in June 2014, was successfully
carried out using the developed parameters. Some
observations on the applicability of the new standard are
reported.
1. Introduction
High voltage equipment is usually protected by metal
oxide (MO) surge arresters, which also applies to
thyristor valves in LCC HVDC (line commuted converter
high-voltage direct current) systems [1] [2]. Hence, the

*reinhard.goehler@siemens.com

KEYWORDS
MO arrester – metal-oxide resistor (MOR) – HVDC – valve protection – thermal behavior – equivalent continuous
operating voltage (ECOV) – energy handling – creepage distance – insulation withstand

Cigre Science & Engineering • N°7 February 2017

23
maximum voltage stress on the valves with respect to
surge voltages and short-time voltage stress is determined
by the valve arresters. The type test voltages for the surge
voltage tests of the valves are derived from the arrester
protective level (+15/20 %).
The withstand voltage of thyristors is different for direct,
alternating and surge voltages and may be also different
for the forward and the reverse direction. Nevertheless,
the protective level of the valve arrester is often a
decisive criterion for determination of the number of
series connected thyristors for a thyristor valve in a given
application.
Since the costs of thyristor valves predominantly depend
on the number of series connected thyristor levels there is
a strong motivation to minimize the protective level of the
valve arresters to the greatest possible extent.
The criteria for the choice of the protective level of the
valve arresters include switching surges transferred from
the AC system, for long distance transmission faults at
the HV-bushing and consequently discharge of the line
or cable, as well as the energy dissipation caused by
commutation transients in short-time and normal steady
state operation. In many cases the latter criterion is the one
that is decisive for the choice of the minimum possible
protective level. The commutation transients are mainly
damped by the snubber circuits within the thyristor valves.
Different cases have to be taken into consideration for the
evaluation of the stresses during commutation operation
under steady state conditions. Generally, the stress on the
arrester generated by commutation transients increases
with increasing firing angle.
The most obvious case is normal rectifier or inverter
operation, where the overshoot at turn-off generates the
relevant losses in the valve arrester (Fig. 1 a: rectifier
Figure 1: Valve voltage; a) rectifier operation (turn-off), b) rectifier opera-
operation). Another case, which could be relevant, is tion, c) inverter operation (transferred turn-off overshoot)
an operation where the transferred transient(s) from the
turn-off of other valves is decisive (Fig. 1 b: rectifier voltage (ECOV), which generates the same power losses
operation). Also at inverter operation the transferred turn- in the MO resistors as the actual operating voltage. This
off overshoot may be relevant (Fig. 1 c: inverter operation facilitates design and testing of the arresters.
- transferred turn-off overshoot). Once the decisive case
for steady-state operation has been found, the energy Additionally, the stresses during short-time operation with
dissipated in the valve arrester protecting a thyristor increased firing angle (e.g. at firing angle of 90°) have
valve in operation according to IEC 60071-5 [3] can be to be evaluated. As for the above mentioned operating
used to determine an equivalent continuous operating and fault cases this is covered by the adiabatic energy

Cigre Science & Engineering • N°7 February 2017

24
 Figure 2: Multiple valve unit Hudson with conventional valve arrester Figure 3: Multiple valve unit Hudson with HCA

absorption capability of the arresters (their repetitive
charge transfer rating, respectively, acc. to the arrester
standard IEC 60099-9 [4]).
After all the design process is an optimization between the
valve arrester and the thyristor valve design parameters
(e.g. number of series connected thyristors and layout of
the snubber circuit).
It is obvious that in general valve arresters with increased
energy absorption capability render possible designs
with lower number of series connected thyristor levels.
Therefore, an arrester design without housing and
additional design features for better heat transfer has
been introduced, designated as “High Cooling Arrester”
(HCA). The pilot application has been the HVDC system
Storebaelt where the arresters are in operation since 2009.
Ever since the HCA has been installed in a growing number
of valves in HVDC systems without any problems.
Under certain circumstances the protective level of a valve
arrester can be reduced by more than 10 % compared to an
arrester with housing, and the number of series connected
thyristor levels can be reduced accordingly with the
corresponding effect on costs. For the back-to-back
system Hudson the protective level of the valve arrester
could be reduced by 10 %. Since only discrete numbers
of series connected thyristor levels can be realized within
the thyristor modules this resulted in 13 % reduction of
the number of thyristor levels. A multiple valve unit in the
Hudson system comprises four valves belonging to one
AC phase of the 12-pulse group. In this arrangement it
was possible to reduce the number of thyristor modules
from 6 to 4 resulting in a height reduction of the multiple
valve unit from 9 m to 7.5 m (Fig. 2 and Fig. 3).
2. High cooling arrester design
Valve arresters without housings have been in service for
many years. Their better thermal properties are obtained
by simply omitting the housing. To further improve an
arrester’s thermal behavior, enforced heat transfer can
be considered by using e.g. compressed air of controlled
temperature or water cooling as it is used for thyristor
cooling. Both will be available in converter halls, but it
is very complicated including these measures into the
arrester. A specific design of passive cooling elements is
much easier to handle in the overall arrester design. In
addition to the cooling improvement specific creepage
distance requirements have to be fulfilled.
2.1. Design of insulation
According to IEC 60071-5 [3], a creepage distance of
14 mm/kV is required for indoor insulation. This means
for the MO resistors used for the HCA application that
their intrinsic creepage distance (i.e. the direct path on the
coating between the metallic end faces) must be doubled.
This is achieved by sheds made of silicone rubber, which
are slipped over the MO resistors as can be seen in Fig. 4.
Figure 4: MO resistor with slipped-over silicone rubber shed and cooling fins
2.2. Design of active cooling element
To find the best geometry of cooling elements, thermal
simulations have been carried out for complete arrester
assemblies consisting of MO resistors, polymer sheds
and metallic cooling parts. The simulations were carried
out with the FEM software ANSYS. In general, heat

Cigre Science & Engineering • N°7 February 2017

25
 Figure 5: Cooling elements and simulated temperatures; a) simple shape, b) with vertical fins

transfer is constituted by heat convection, radiation 20 % higher compared to that of each column of HCA1-3.
and conduction. In order to determine the simulation This increase was chosen to achieve the same equivalent
parameters, preliminary simulations and tests on different continuous operating voltage (ECOV), determined
models were carried out. With the final parameter set according to IEC 60099-9 [4], for both arrester designs.
of  convection  coefficient  α,  emissivity  e, specific heat The tests were performed under applied alternating
capacitance and thermal conductivity of the three materials voltage in a closed test chamber with a volume of 200
– varistor ceramic, silicone rubber and aluminum – the m³. Its internal temperature can be controlled by heaters
geometry of the cooling elements was optimized. The and air condition up to a maximum of 60 °C. The test
simulations were started with simple aluminum parts transformer is located in the chamber. As the test voltage
as can be seen in Fig. 5 a). Different steps enlarging the is limited by the transformer rating, the test samples were
metallic cooling area including horizontal fins lead to only chosen to have an ECOV of 94 kV. The whole test setup
minor improvements. The final step, introducing cooling is shown in Fig. 6.
elements with vertical fins as shown in Fig. 5 b) results in a
cooling behavior which is about 45 % better than the initial
one. Electrical behavior, such as dielectric strength and
absence of partial discharges (electrical field simulation,
partial discharge and withstand tests, see Section 4) and
the reduced overall length of the arrester were taken into
consideration. For evaluation the temperature increase
within the MO resistors was taken at constant power loss
and heat generation in the MO resistors.

3. Thermal investigations
To investigate the thermal behavior of the HCA and to
compare it with that of a conventional housed arrester
(HA) four different arresters are considered:
1. Housed arrester with one MO resistor column (HA1)
2. High cooling arrester with one MO resistor column
(HCA1)
3. High cooling arrester with two MO resistor columns
(HCA2)
4. High cooling arrester with three MO resistor columns
(HCA3)
Figure 6: High Cooling Arrester (HCA3) installed in the test chamber for
The number of MO resistors (MOR) used for HA1 is thermal testing at ambient temperature of 20 °C and 60 °C

Cigre Science & Engineering • N°7 February 2017

26
3.1. Cooldown behaviour at different temperatures For multi-column arresters different heat-up times were
tested as well. It could be observed that longer heat-up
In a first test program, the cooling behavior of both arrester
times are preferable to keep the deviations between the
designs (conventionally housed and high cooling type)
temperature increases of different columns small. The
is compared. For both designs the comparison is carried
deviations are limited to 20 K at 140 °C in the standard.
out on the single column arrester, HA1 and HCA1. After
This finding is evident when considering that electrical
installation of the arrester under test, the test chamber is
differences between columns, built with MOR of the
heated up or cooled down to 20 °C or 60 °C over night.
same make, will result in a parallel shift of the U-I
Thus all arrester components are at ambient temperature
characteristic in direction of higher voltages. In Fig. 7
when a test starts. In a second step the arresters are heated
fictitious U-I-characteristics of two arrester columns
up to 150 °C by application of a voltage above ECOV. In
with different residual voltages are shown (differences
case of 60 °C ambient temperature, this voltage is varied
are usually very small and have been exaggerated here
in order to investigate the influence of different heat-up
for better visualization). If the heating of an arrester
times on the cooldown behavior. To minimize possible
comprising both these columns in parallel is accelerated
long-term side effects, such as small voltage fluctuations,
by increasing the voltage by ∆U the difference in current 
the arresters are cooled down without any electrical stress.
peaks is increased from ∆/1 to ∆/2. Thus the heating of the
In Fig. 8 b) the blue and red curves show the cooling of
column with the lower residual voltage (red curve) will
HCA1 after heat-up times of 17 min and 74 min, and the
speed up compared to that of the column with the higher
green and cyan curves those of HA1 with heat-up times of
residual voltage (blue curve).
12 min and 36 min, respectively. Other heat-up times were
investigated, but are not presented here. Additionally, for
better visualization, the diagram has partly been zoomed
in. It can be seen that no significant influence of the heat-
up times can be observed. Due to the small height of the
tested arresters, the influence of stray capacitances on
axial voltage, power loss and temperature distribution is
negligible. A taller arrester with uneven axial voltage and
temperature distribution might lead to different results.
However, the test conditions are in agreement with the
requirements of the “Test to verify thermal equivalency
between complete arrester and arrester section” given in
standard 60099-9 [4], which allows for heat-up times of a
maximum of 1 h and for using only a single unit even in
case of multi-unit arresters.

Figure 8: Arrester cooling curves measured at ambient temperatures of 20 °C

Figure 7: Schematic visualization of the increase in current difference ∆/ of  (a) and 60 °C (b), zoomed segment showing the time interval from 0.5 to 1
two arrester columns with different U-I-characteristics (blue and red curves) h. Black dotted curves show the cooling behavior of an ideal single thermal
when the applied voltage is increased by a value of ∆U capacitance with only one thermal time constant fitted to the measured data

Cigre Science & Engineering • N°7 February 2017

27
For quantitative comparison of HCA1 and HA1 the time designs are compared for 60 °C ambient temperature and a
instants where the over temperature (Tarrester – Tambient) has voltage, which is 10 % above ECOV (I in Fig. 9): Though
reached 37 % of the initial value are compared. This time power losses of HCA1 are 2.8 times higher than those of
instant represents the thermal time constant under the HA1 under these conditions, the steady state temperatures
assumption of exponential cooling with only one time of both arrester types are identical due to the improved
constant, which is justified at least for the HCA, which has cooling behavior of the HCA1.
a “housing” in direct contact to the MO resistors, and in
general in case of an even axial temperature distribution.
Simple exponential cooling is expressed by
T = Ti × exp(-t/(RC)th)
where T is the actual and Ti is the initial over temperature
The resulting curves are depicted in Fig. 8 as black colored
curves with thermal time constants of 64 min and 136 min
in case of 20 °C ambient temperature and 65 min and
130 min for 60 °C ambient temperature. While the
match in case of the HCA1 is rather good, deviations
can be clearly seen in case of HA1 in both diagrams.
These deviations are due to the air gap inside the arrester
housing, which does not allow regarding the system as Figure 9: Steady state arrester temperatures of HCA1 and HA1 for voltages in
the range of 1.0…1.1 times ECOV for ambient temperatures of 20 °C and 60 °C
a lumped system with only one thermal time constant.
Despite these deviations, the difference in time constants
of the idealized curves indicates that cooling in case of 3.3. Influence of multi-column design on the steady
HA1 design takes at least twice the time than it is the case state arrester temperature
for HCA1. Due to the higher load of the test transformer in case of
3.2. Steady state arrester temperatures at different multi-column designs the deviations of Upeak/Urms from
operating voltages the  value  of  √2  are  higher  than  2  %  for  voltages  above 
1.05 times ECOV. These increased deviations result in
To compare the thermal behavior of the arresters in the different resistive components of the current and, therefore,
range of possible service conditions, the arresters are influence steady state temperatures. However, for a direct
energized at voltages greater than or equal to ECOV and comparison of arresters with different numbers of columns
operated up to thermal steady state conditions. In Fig. 9 the ratio of peak voltage to RMS value is decisive, and
the steady state temperatures are depicted for voltages, therefore, a comparison is only meaningful for voltages
where the ratio of peak voltage to RMS value deviates were the ratio Upeak/Urms at given voltage levels is, except
by less than 1.7 % from the value of √2, and the voltage 
for small deviations, the same for all arresters compared.
varies by less than ±0.5 % during the tests. These test
Hence the single column arrester is tested in a changed test
conditions are in line with IEC 60099-4 [5], Clause 8.7,
configuration compared to that presented in the sections
where a deviation of less than 2 % from the value of √2 
above. In the changed set-up an additional two column
and a voltage fluctuation of less than 1 % is allowed. The
arrester is connected in parallel to the single column arrester,
highest deviation from the value of √2 at a fixed voltage 
with the result that the test voltage shape is basically the
level between the different arrester designs is 0.4 % points.
same as in case of testing the HCA3. Due to practical
The resistive peak currents and, therefore, the power restrictions, in case of testing the HCA2 the transformer was
losses generated are different for HCA1 and HA1. loaded only by the two columns of the HCA2. Therefore,
Nevertheless, the steady state temperatures, pictured in the deviations, which are shown in Fig. 10 a), are different
Fig. 9, are similar in the observed range. Exemplarily the for HCA2 in comparison to HCA1 and HCA3.

Cigre Science & Engineering • N°7 February 2017

28
 voltages above ECOV, the good agreement of steady state
temperatures at applied ECOV shows that the arresters’
performance is not influenced for the voltages applied in
real service, which are equal to or less than ECOV.

Type test
A complete type test according to standard IEC 60099-
9 [4] was successfully carried out using the developed
parameters. The steady state temperature during the tests
performed on real size, completely assembled single-
and multi-column arresters was approximately 80 °C.
Therefore, the preheating temperature during the thermal
recovery test was chosen as 80 °C instead of 60 °C.

Figure 10: Ratio of û/√2 to URMS (a) and steady state arrester temperatures

(b) for HCA1-3 for voltages tested in the range of 1.0…1.14 times ECOV
and an ambient temperature of 60 °C; test configuration for HCA1 changed 
compared to Sections 3.1 and 3.2

The highest deviations occur between HCA1 and HCA2,

with a value of 0.6 % points at ECOV.
In Fig. 10 b) the steady state temperatures of all HCAs
are presented. In case of multi-column arresters only
the values for the column with the highest steady state
temperature are shown. The last depicted measuring point
Figure 11: Single unit of the surge arrester prepared for a dielectric test
for each arrester is the highest voltage applied that did not
lead to thermal runaway. It is evident that the number of Insulation withstand tests were performed on the longest
columns has a negative influence on the thermal behavior arrester unit, see Fig. 11. According to IEC 60099-9 [4],
of an arrester. This is well known and due to the heat the minimum required test voltages are
exchange between the columns, where heat radiation has
a major impact. Thus the effective heat dissipation per - LIWV = 518 kV
column decreases with increasing number of columns. - SIWV = 514 kV
Although the worse cooling behavior of the multi-column The tests were passed at 526 kV lightning impulse voltage
designs has an impact on the steady state temperatures for and 515 kV switching impulse voltage. The dielectric tests

Cigre Science & Engineering • N°7 February 2017

29
shall be performed on the empty housing of an arrester.
However, in this case, there is no housing and the
arrester is built with a stack of MO resistors and cooling
elements. Furthermore, the dielectric tests cannot be
performed on the complete arrester due to the voltage
limiting characteristics of the MO resistors. In order to
perform the dielectric tests, either the MO resistors must
be replaced by insulating cylinders to fill the space, or
the arrester must be built with MO resistors having a
higher voltage gradient (i.e. residual voltage per unit
length). Here, the latter approach was chosen. The U-I-
characteristic of the MO resistors was adjusted to the
test voltage to be applied. The resulting current during
impulse voltage application was in the range of 1 A in
order to not influence the impulse voltage generator.
Fig. 12 shows the comparison of the standard protection
characteristic and the increased characteristic used for
the switching impulse withstand voltage test at 515 kV.
Figure 12: U-I-characteristic of the MO resistor column used for insulation
withstand test in comparison with the standard protection characteristic
Summary and conclusion
A special design of a valve arrester for protection of
the valves in a line commutated converter (LCC) has
been developed, designated as “High Cooling Arrester”
(HCA). The arrester has an extremely low protective
level, which helps saving overall valve costs, but for the
price of high power losses and thus heating under the
typical operating conditions. To manage this its cooling
behavior has been improved not only by completely
omitting the housing but additionally by introduction of
Cigre Science & Engineering • N°7 February 2017
30
thermally optimized metallic “active” cooling elements.
The creepage distance requirements of 14 mm/kV have
been fulfilled by silicone rubber sheds, which are directly
pushed over the metal oxide resistors. The arrester has
been designed and optimized by electro-thermal 3D-FEM
simulations. The resulting protective level is some 10 %
lower than that of a comparable conventional arrester
with housing. Considerable cost savings in the valve
design can consequently be achieved. This report has
demonstrated how the Equivalent Continuous Operating
Voltage (ECOV) of the arrester has been determined
by heating tests at 20 °C as well as at 60 °C, which is
the maximum specified temperature in a valve hall. For
the type tests the new standard for HVDC arresters,
IEC 60099-9, has been applied. Some necessary and
reasonable adaptions had to be introduced, e.g. the start
temperature in the thermal recovery test was chosen to
be 80 °C rather than 60 °C, as is required in the standard,
and the dielectric tests on the “housing” (which does not
exist) were performed using a completely assembled
arrester with metal oxide resistors of higher protective
level in order to allow application of the required test
impulse voltage levels. All in all, by optimizing an
apparently simple device like a surge arrester by modern
simulation and test approaches a notable cost benefit
could be achieved for the valves of LCC converters.
In principle this arrester can also be applied in other
(indoor) locations of HVDC systems of both LCC and
VSC (voltage source converter) types. This will be
further investigated in the future.
6. Bibliography
[1] K. Steinfeld, R. Göhler, D. Pepper, „High Voltage Surge Arresters
for Protection of Series Compensation and HVDC Converter
Stations”, The 4th International Conference on Power Transmission
and Distribution Technology, October 11-14, 2003, Changsha,
China
[2] CIGRÉ Working Group A 3.17, “MO Surge Arresters - Stresses and
Test Procedures”, CIGRÉ Tech. Brochure 544, 2013, pp. 15–23
[3] IEC 60071-5, Ed. 1.0, 2014-10: “Insulation co-ordination – Part
5: Procedures for high-voltage direct current (HVDC) converter
stations”
[4] IEC 60099-9, Ed. 1.0, 2014-06: “Surge arresters – Part 9: Metal-
oxide surge arresters without gaps for HVDC converter stations”
[5] IEC 60099-4, Ed. 3.0, 2014-06: “Surge arresters – Part 4: Metal-
Oxide surge arresters without gaps for a.c. systems”
7. Biographies
Reinhard Göhler was born in Reinsdorf near Hannover,
Germany on January 20, 1955. He received the Dipl.-
Ing. degree in 1980 in electrical engineering from
Technische Universität Braunschweig, Germany. He
joined Siemens AG, Switchgear Works, Berlin in the
same year, where he began as a test and development
engineer for surge arresters. In 2001 he became director
of the sales department for special surge arresters and in
2004 director of the R&D department. He is member
of IEC TC 37 MT 4, IEEE SPDC and arrester working
groups of CIGRE A3.
Junaid Shaikh was born in Mumbai, India on October
22, 1983. He received the B.E. degree in 2005 in electrical
engineering from Mumbai University, India and the
M.Sc. degree in 2010 in electrical power engineering
from Technische Universität Darmstadt, Germany. Since
2010, he is working as a test and development engineer
for surge arresters at Siemens AG, Switchgear Works,
Berlin.
Volker Brendler was born in Nürnberg, Germany on
June 18, 1960. He received the Dipl.-Ing (FH) degree
in 1984 in mechanical engineering from Georg Simon
Ohm Fachhochschule Nürnberg, Germany. He joined
Siemens AG, Gerätewerk Erlangen in the same year,
where he began as a development engineer for high
voltage converters (mechanics). From 1987 he developed
calculation tools and did electrical design and project
management of converters for drive systems, static VAR
compensators and HVDC valves. Since 2006 he works
Cigre Science & Engineering • N°7 February 2017
31
in the system design department for HVDC systems now
as a senior expert.
Michael Weyer was born in Aachen, Germany on
August 16, 1979. He received the Dipl.-Ing. degree in
2007 in electrical engineering from RWTH Aachen. He
joined Siemens AG, Erlangen in 2008 as an engineer
for HVDC components. In 2012 he started to design
thyristor valves for HVDC classic projects.
Moritz Giessel was born in Marburg an der Lahn,
Germany, in 1985. He received the Dipl.-Ing. degree
in 2012 in electrical engineering from Technische
Universität Darmstadt (TU Darmstadt). Since 2013 he
is working as research associate towards a PH.D. degree
with high-voltage laboratories, TU Darmstadt, Germany.
Volker Hinrichsen was born in 1954. He received the
Dipl.-Ing. and Dr.-Ing. degrees in electrical engineering
from Technische Universität Berlin, where he studied
and worked as Research Assistant. He joined Siemens
Power Transmission and Distribution (Berlin) in 1989
as test engineer in the high-voltage laboratories. In 1992
he became Director R&D of the Surge Arrester Division.
In 2001 he joined Technische Universität Darmstadt as
full professor in high-voltage engineering. He is head of
the High-Voltage Laboratories, including an accredited
800 kV test laboratory, and is active in education,
research and testing. He is chairman of IEC TC37 (surge
arresters), convenor of IEC TC 37 MT 4, chairman of
DKE K 441 (surge arresters), convenor and member,
respectively, of several Cigré Working Groups in SC A3
and C4, and member of IEEE SPDC.
Development of the riser cable system for
offshore floating wind power project
S. Fujii, H. Tanaka, Y. Tateno*
Furukawa Electric Co. Ltd.
Japan
Summary
The research and development for the renewable power
energy to utilize marine energy, such as an offshore
wind power generation and an ocean thermal energy
conversion, are promoted with Industry-Government-
Academia collaboration.
The authors have proceeded with a Fukushima floating
offshore wind farm demonstration project (Fukushima
FORWARD Project) as a partner of the commissioned
project under the Ministry of Economy, Trade and
Industry and are responsible for a transmission system
(including a power and an optical transmission lines) for
the whole project.
The cable installation and connection to the offshore
substation and wind turbines were successfully
completed, and it is under operation now.
1. Introduction
An offshore floating wind turbine is a promising power
generation method in Japan, because Japan has the 6th
largest exclusive economic zone (EEZ) in the world and
has few suitable shallow shores for the implantation
of the offshore type wind power generations. Offshore
floating wind power generations are very few in the
world. The construction of a wind farm which consists of
floating wind turbine and an offshore floating substation
was the first experience in the world.
The authors have developed the dynamic cable systems
for power transmission from offshore wind turbine
generators to the shore system grid, and completed the
cable installation in Fukushima FORWARD Project.
This paper reports the high voltage dynamic cable,
submarine joint and installation method which have
been developed for offshore floating system [1].
*yuji.yt.tateno@furukawaelectric.com
KEYWORDS
Renewable energy - Floating wind turbine - Dynamic cable -Riser cable - Submarine joint
Cigre Science & Engineering • N°7 February 2017
32
2. Over view of the Fukushima
forward project
2.1 Project Overview
Fukushima FORWARD Project is consisting from
the 1st stage (2011-2013) and the 2nd stage (2014-
2015). In the 1st stage, a 2 MW down-wind type
offshore floating wind turbine generator and a
25 MVA offshore floating substation, for the first time in
the world, were constructed. In the 2nd stage, a 7 MW
new floating wind turbine generator, as the largest in the
world, was constructed [2] [3].
The authors installed dynamic cables (referred to as riser
cable) and a long distance submarine cable to connect
from the offshore floating bodies to onshore facilities
and power line. Fig. 1 shows the project overview [4].
2.2 Transmission and Substation System
The wind power generation capacities are 2 MW and 7
MW, and a 22 kV cable (inter-array cable) was selected
for the transmission line from the generators to the
offshore substation.
The route distance from the proposed shore position to a
wind turbine installing position was approx. 25 km long
and those were connected with a submarine cable.
An offshore substation was constructed to raise the cable
(export cable) voltage to 66 kV for power transmission
from the substation to the shore port because a 22 kV
cable had higher transmission loss. For grid connection,
the export cable on shore was connected with an
existing 66 kV overhead transmission line which had
been operated by a power utility. Fig. 2 illustrates the
transmission and substation system [4].
 analysis and a fatigue analysis are in turn conducted with
combinations of the various parameters in accordance
with the riser cable design flow illustrated in Fig. 3.
The major profiles of riser cable shape like Free Hanging,
Lazy-Wave, and Lazy-S shown in Fig. 4 are in practical
use in development of riser pipes for the marine oil and
gas industries. It is important to select the best design of
riser cable system out of those applicable systems. For
Fukushima FORWARD Project, the riser system in Lazy-
Wave was selected in consideration of analytic condition
and water depth at the installing position.
The analyses and feedbacks have been reiterated and
Fig. 1 Project overview
finally the most suitable submersible shape design and
its detailed riser cable design were defined. The required
characteristics for the riser cable are shown in Table 1.
Table 1 Required characteristics of riser cable

Analysis Condition Target characteristics

- Meet allowable tension and
Sea condition at
minimum bending radius in
Static / installation site
floating condition
Floating structure
Dynamic - Floating part of the cable
rolling condition
should not touch the sea bed
Cable parameters
- Cable should not be kinked
Cable strain fatigue
property - Similar to wind turbine or
Fatigue
Current wave floating structure
statistic data
Fig. 2 Transmission and substation system

3. Designing of riser cable system

3.1 Design flow of Riser Cable System
Fukushima FORWARD Project is designed with a
submarine riser cable consisting of optical fibers to
control a wind turbine, a floating body and also to transmit
measured data.
The riser cable connected to an offshore floating body
shows quite complicated movements due to shakes of
floating body (irregular motion on 6 degrees of freedom)
and fluid forces. To design the riser system, therefore,
the riser shape and accessories such as a bend stiffener,
distributed buoyancy modules (DBM) and protection tubes
have been designed by analytic simulation of underwater
cable behavior [5]. A riser shape should be selected at first,
and then, a static behavior analysis, a dynamic behavior Fig. 3 Design flow of riser cable

Cigre Science & Engineering • N°7 February 2017

33
 Fig. 4 Riser system cable profiles in practical use
3.2 Cable Behavior Analysis
The oceanographic condition such as height and frequency
of significant wave and a floating body’s dynamic response
are needed at first for external conditions.
In addition to the basic physical properties of cable
diameter and weight, and the mechanical parameters of
cable’s longitudinal rigidity (EA), bending rigidity (EI) and
torsional rigidity (GJ) are required for the behavior analysis.
The bending rigidity is the most important to check the
cable curvature. As the mechanical parameters the actual
measured values have been used in those analyses. The
analytic model of Lazy-Wave is shown in Fig. 5
In static analysis, the riser cable configuration, curvature,
maximum load (tension, shearing force, bending moment)
while the floating body has a maximum displacement
with a maximum current are checked. Fig 6 shows some
simulation results of riser cable configuration in case
that the floating body moves towards three directions
from a neutral position.
Fig. 5 Analytic model of Lazy-Wave
Fig. 6 Simulation results of riser cable configuration (sample)
Cigre Science & Engineering • N°7 February 2017
34
In dynamic analysis, the riser cable configuration, curvature,
maximum load are checked while the floating body in a
neutral position is affected by irregular waves corresponding
to wave-fluctuating amplitude. It is necessary to verify the
following aspects in the riser cable configuration: “The
catenary lowest point is above the ground.” “DBMs are
below the surface of sea.” “The ground touching position is
within the protected section.” It is also important to confirm
that the tension and the curvature at the top of riser cable
are within allowable values. Fig. 7 shows simulation results
of dynamic variation for tension and curvature at the top
of riser cable. One of difficult aspects was that the floating
body’s moving distances were larger than the water depth
in some cases.
Fig. 7 Simulation result samples of dynamic variation for tension
and curvature
3.3 Cable Fatigue Analysis
The cable life time against repeated bending and tension
variations is assessed in fatigue analysis after the
proposed riser system is confirmed to satisfy the design 3.5 Structure of Riser Cables
criteria in static and dynamic behavior analysis. The fatigue Based on those studies, the riser cable structures are
characteristics of riser cables and floating body’s dynamic
defined as shown in Fig. 10 and Table 2. The cables have
response for each wave-direction and wave-cycle are
double layers of strand armor wires twisted in opposite
required for fatigue analysis. For Fukushima FORWARD
Project the long-term wave statistics around the installation directions in order to prevent cable torsion.
area were provided. It has been confirmed that the material,
mechanical and functional components in designed riser
systems have more than 20-year life time which is also
requested to floating bodies and wind turbines.
3.4 Evaluation of Cable Properties
To improve the accuracy of analytic simulation, every
input data of riser cable’s mechanical properties
such as EA, EI, GJ used in those analyses are based
on the experimental results with actual cables.
Fig. 8 shows the actual evaluation of cable EA.
Fig. 9 shows tension bending test condition to evaluate
tensile strength and lateral pressure resistance of a riser
cable. The cable fatigue performances have been also
evaluated with actual cables.

Fig. 10 Structure of riser cables

Fig. 8 Experimental evaluation of cable EA
Table 2 Structure of riser cables

Nominal
66 kV 22 kV
voltage
Conductor 3 x 100 mm2 3 x 150 mm2
Insulation
11 mm 6 mm
thickness
Double layers of galvanized steel
Armor
wires (6.0 mm)
Outer diameter 175 mm 147 mm

Fig. 9 Riser cable tension bending test

Weight 53 kg/m in air 43 kg/m in air

Cigre Science & Engineering • N°7 February 2017

35
 Fig. 11 Assembled 22kV submarine joint

4. Designing of submarine joint

The authors developed the submarine joints to connect
cables having different designs (a riser cable and a
submarine cable) for Fukushima FORWARD Project.
Based on JEC-3408, CIGRE TB 490 and CIGRE Electra
No.171, the required characteristics of the submarine
joints are stipulated as shown in Table 3.
The 66 kV submarine joint involves three splices of
power cable connection and three sets of optical cable
splice box inside the 7 m protection tube. The 22 kV
submarine joint involves three splices of power cable
connection and one set of optical cable splice box inside
the 4m protection tube. It is designed that the protective
tube is for external damage resistant to be buried and
the structure of cable joint box should have a waterproof
function. It is also designed to have proper strength to
withstand mechanical stresses during both installation
methods of horizontal hanging style and catenary shape
style. A pre-molded joint for 66 kV system and a taped
joint for 22 kV system have been selected for power
cable connection and water-tight structures have been
Fig. 12 Cable laying vessel
provided in a protective copper tube. Bend restrictors
have been attached on both edges of the submarine joint
to avoid excessive cable bending. It was ensured that
the prototype 22 kV submarine joint could pass a water
permeability test under 1.3 MPa water pressure. The
assembled 22 kV submarine joint is shown in Fig. 11.
5. Cable and joint installation
The vessel equipped with a turntable was used for cable
installation and operated with a function of Dynamic
Positioning System (DPS). The cable laying vessel is
shown in Fig. 12.

Table 3 Required characteristics of submarine joints

Item Requirements
Nominal voltage 66 kV 22 kV
Connection It enables to connect a submarine cable and a riser cable.
- AC: 130kV x 3 hours (at R.T.) - AC: 57kV x 3 hours (at R.T.)
Withstand Voltage - Lightning Impulse: - Lightning Impulse:
±485kV x 3 times (at R.T.) -230kV x 3 times (at R.T.)
130m of maximum depth
Bearing Water Pressure 130m of maximum depth Water permeability
≤ 1×10-7 [g∙(cm/cm2)∙day∙mmHg]
Bearing Tensile Strength 93.1kN approx. of immersing tension
Assembling It can be assembled on the vessel.

Cigre Science & Engineering • N°7 February 2017

36
The cable installation procedure in 2nd stage is illustrated
in Fig. 13. It started with pulling up the first riser cable to
the floating substation, and DBM and other accessories
were fixed on the cable with unreeling. The riser cable was
installed to the expected position with a care of the touch
down point, and the first submarine joint was assembled
on the vessel. That joint was installed in catenary shape
style and the submarine cable was installed to the location
at which the wind turbine would be placed.
Then, the second riser cable was installed in same manner
between the floating wind turbine and the submarine
cable end. The submarine cable end was hoisted up and
the second submarine joint connected the submarine cable
and the second riser cable on the vessel. This joint was
installed in horizontal hanging style.
Fig. 14 Cable linear shape monitored by ROV
As after installation test, DC voltage withstand test
to the power cables and an OTDR measurement to the
optical cable were carried out to check the soundness.
It was confirmed by Remotely operated vehicle (ROV) 6. Current status of
monitoring that the riser cable shapes and positions were
meeting the designed values. Fig. 14 shows cable linear demonstration project
shape monitored by ROV. To monitor the cable movements, triaxial accelerometers
were attached on some riser cables. Those data have
been collected continuously and it isn’t considered that
any sever cable bending with not-allowable curvature
has taking place. Keeping those data collection and
statistical data analysis, monitoring of riser cable should
be still going on.

7. Conclusion
The cable installation and connection to the world’s
largest floating 7 MW offshore wind turbine in the 2nd
stage were successfully completed, and it is already
under operation.
There is, so far, not enough experience in riser cable
transmission system which is required for the floating
power generation system. With the facts found
throughout the Fukushima FORWARD Project, the best
transmission system design and the maintenance method
will be established.

Fig. 13 Installation procedure of 22 kV riser cables and submarine cable
8. Acknowledgments
This research has been carried out as a part of Fukushima
floating offshore wind farm demonstration project
funded by the Ministry of Economy, Trade and Industry.
The authors wish to express their deepest gratitude to the
concerned parties for their assistance during this study.

Cigre Science & Engineering • N°7 February 2017

37
9. Reference
[1] S. Fujii, Y. Tateno, K. Tomikuda, Y. Tominaga, N. Horiguchi,
H. Nakano, H. Kon, T. Kagoura, T. Yamaguchi, H. Sakakibara,
“The Development of Power Transmission System for Fukushima
FORWARD Project”, Furukawa Review, No.43 pp.28-33, 2013.
Available: http://furukawa.co.jp/review/fr043.htm
[2] Y. Tominaga , H. Nakano, K. Tomikuda, M. Toyoda, S. Torii, K.
Yagihashi, M. Minamide, Y. Tateno, «Dynamic Cable Installation
for Fukushima Floating Offshore Wind Farm Demonstration
Project», B1-1090, AORC CIGRE 2014.
[3] K. Yagihashi, Y. Tateno, H. Sakakibara, H. Manabe, «Dynamic
Cable Installation for Fukushima Floating Offshore Wind Farm
Demonstration Project», B4.3, Jicable 2015.
[4] Fukushima Offshore Wind Farm Consortium. Available: http://
www.fukushima-forward.jp/english/
[5] K. Kagoura, «Ocean Power Transmission System», No.62, The Japan
Society of Naval Architects and Ocean Engineers: KANRIN, 2015 in
Japanese.
10. Biographies
Shigeru Fujii was born on 26 January 1966 in
Kanazawa. He graduated from Department of
Nuclear Engineering in Nagoya University. He
joined Furukawa Electric in 1988 and engaged in
the development of distribution cable accessories.
Cigre Science & Engineering • N°7 February 2017
38
In 2011, He engaged in the Fukushima Floating Offshore
Wind Farm Demonstration Project and he is project
leader on Furukawa Electric.
Hideo Tanaka was born on 21 October 1963 in Tokyo.
He graduated from Master Course of Electric Engineering
in University of Tokyo. He joined Furukawa Electric in
1988 and engaged in the development of 500kV XLPE
cable system, water treeing diagnostic system and
HVDC XLPE cable system.
In 2001, He was transferred to VISCAS Corporation,
which is joint venture on insulated power cable system
between Furukawa and Fujikura and engaged in cable
system engineering and R&D. In 2016, he came back to
Furukawa and he is now the General Manager of Cable
System Engineering Department. He is a member of
Cigre as well as Japanese regular member of Cigre SC
B1 and he is also joining IEEE and Institute of Electric
Engineering in Japan.
Yuji Tateno was born on 21 August 1969 in Tochigi,
Japan. He graduated from Master Course of Electric
Engineering in Nagoya University. He joined Furukawa
Electric in 1995 and engaged in the development of high
voltage submarine cable system.
Calculation accuracy of high-temperature
sag for ACSR in existing lines
D. Loudon, EFLA, Norway
D.A. Douglass*, DPC, LLC, USA
R.G. Stephen, ESKOM, South Africa
G.C. Sibilant, EPRI/UKZN, USA
Abstract
According to a CIGRE questionnaire completed in
2003, over 80% of existing lines were built with steel
reinforced aluminium conductors (ACSR). Therefore,
when attempting to increase the thermal rating of existing
High Voltage (50 kV to 345 kV) transmission lines, low-
cost line uprating methods often involve allowing the
existing ACSR conductors to operate at an increased
Maximum Allowable Conductor Temperature (“MACT”)
temperature. Of course, minimum electrical clearances to
ground, buildings and other lines must be maintained and
the conductor system unharmed by operating at higher
temperatures. If these criteria are not met, the line must be
reconductored or rebuilt.
For ACSR, all sag-tension calculation methods assume
that the unstressed length of aluminium and steel layers
is the same before and after installation. After initially
sagging the conductor, high line currents cause the
conductor temperature and sag to increase. CIGRE
Technical Brochure 324 suggests two linear and one
non-linear conductor model. The Experimental Plastic
Elongation (EPE) model considers the aluminium layers
and steel core separately. The paper explains and quantifies
the various factors that influence the calculation of ACSR
sag at high temperature.
The thermal elongation rate of the aluminium layer(s) is
twice that of the steel core, and, while the total tension
decreases with increasing temperature, the tension in the
aluminium layers decreases faster than the tension in the
steel core. For strong ACSR conductors in short spans
operated above 75oC, the aluminium layer tension may
reach zero at a “knee-point temperature” (KPT) which
is less than the line’s actual or proposed templating
* Da.douglass@ieee.org
KEYWORDS
High temperature sag, knee-point temperature (KPT), aluminium layer axial compression, ACSR, LiDAR, Linear
Elastic, Simplified Plastic Elongation, Experimental Plastic Elongation.
Cigre Science & Engineering • N°7 February 2017
39
temperature. Since, above the KPT, the rate of sag increase
with temperature is reduced, the determination of KPT is
a significant factor in evaluating line uprating methods
using the existing conductors.
If the aluminium layers of ACSR undergo plastic
elongation due to metallurgical creep or high tension load
events, then the knee-point temperature of the conductor
is reduced, typically decreasing over the life of the line.
In calculating the conductor sag at temperatures above
the KPT, the possibility of compression forces, residual
manufacturing stress, and greater than anticipated plastic
elongation of the aluminium layers must be considered.
1. Introduction & Background
In designing a new overhead transmission line, one of
the most basic and necessary calculations is that of sag-
tension. CIGRE Technical Brochure 324 [1] explains
the calculation of sag and tension for bare overhead
conductors under various ice/wind conditions and at
the high conductor temperatures produced by the line’s
highest power flow. Based on these calculations, span
lengths and structure heights are chosen to meet the
structure loads and electrical clearance requirements over
the life of the line.
In evaluating existing lines, power flow constraints
on the AC Transmission System are often the result of
inadequate thermal rating on older existing High Voltage
(50 kV to 345 kV) transmission lines, built with one
ACSR conductor per phase. Increasing the thermal rating
of these existing lines can often be accomplished by
increasing the maximum design temperature of the lines
if this can be done while continuing to maintain minimum
electrical clearances to ground, buildings and other lines.
As described in [2], if the sag of these ACSR conductors
at the higher temperature is not adequate, then the line’s
conductors may be re-tensioned, the support heights raised,
the line reconductored with High-Temperature, Low-Sag
(HTLS) conductor of the same diameter, or the line simply
replaced by a new line with adequate thermal capacity.
The initial step in determining the choice of thermal
uprating method for an existing line, is to make a detailed
measurement of conductor position and terrain survey
for the entire line. Increasingly, this is done with airborne
LiDAR (“Light Detection and Ranging”). In most cases,
these older lines are lightly loaded (electrically) when
the system is operating normally. Even with the line in
service, the conductor temperature is typically within 5oC
to 15oC of air temperature since the current in most lines
is less than 0.5 amps/mm2. At higher current densities, the
conductor temperature which corresponds to the measured
sag must be calculated.
2. Errors in High Temperature Sag
TB244 [2] includes a useful table estimating high
temperature sag errors due to various factors. The table is
reproduced here (see Table 1) with certain modifications.
All three conductors are strung to an initial tension equal
to about 25% of their breaking strength.
Notice that the sag in Table 1, at 20°C, is approximately 5
m, and that the sag increases by about 3 meters when the
conductor temperature increases to 120°C. The calculated
high temperature sag can vary by as much as a meter due to
imperfect tension equalization [3] at supports for all three
ACSR conductors. Indeed, sag errors due to imperfect
tension equalization, occur for all aluminium conductor as
well as ACSR.
As per Table 1, the second largest variation in calculated
sag at 120°C is due to the choice of conductor stress-strain
model. The models are discussed at length in [2] and also
in this paper.
Other factors, including additional heating of three-layer
ACSR at high current due to the steel core [4], variation
in the conductor temperature when manufactured, creep of
aluminium at high temperature, and variation in thermal
elongation rate due to high stress, produce relatively small
errors, typically less than 0.15 m, but can be important in
certain situations.

Table 1: Sag Errors at High Temperature for three Different ACSR Strandings

Typical Errors for High-temperature Sag Calculations

ACSR Drake ACSR Condor ACSR Tern
Aluminium area (strands) 403 mm2 (26) 403 mm2 (54) 403 mm2 (45)
Steel area (strands) 66 mm2 (7) 53 mm2 (7) 28 mm2 (7)
Final tension at 20oC 25 800 N 23 150 N 19 100 N
Equivalent span length 250 m 250 m 250 m
Sag at 20°C 4.84 m 5.06 m 5.36 m

Effect of Conductor Stress-strain model on final 120 ºC sag:

Single conductor modulus & CTE (LE or SPE) 7.76 m 7.78 m 8.53 m
Separate modulus & CTE (Graphical method) - no Residual stress (EPE) 7.00 m 7.53 m 8.53 m
Graphical method with typical 20 MPa residual stress offset. 7.32 m 7.73 m 8.53 m
Additional sag errors at 120 ºC :
Temperature difference core/surface +0.03 m +0.05 m +0.06 m
Change of elastic modulus vs. temperature +0.15 m +0.11 m +0.06 m
High temperature creep 0 0 +0.50 m
Multiple span effects +0.6 to -1.0 +0.5 to -0.9 m +0.5 to -0.8 m
Effect of core magnetisation losses 0 + 0.07 m +0.05 m
Effect of manufacturing temperature +/- 0.14 +/- 0.12 0

Cigre Science & Engineering • N°7 February 2017

40
3. Stress-strain models for
sag-tension calculation
The stress-strain mechanical models for ACSR, used in
sag-tension calculations, assume that: (1) the manufactured
lengths (i.e. “unstressed lengths) of the steel core and
the aluminium layers are the same; (2) the mechanical
behaviour of the steel core is close to elastic and yields only
a small amount of plastic elongation at high load events,
and; (3) the thermal elongation of both the steel core and
the outer aluminium layers are linear with temperature. In
addition, the common conductor elongation models can
consider the thermal and mechanical elongation of the
steel core and the outer aluminium layers separately.
The models for conductor elongation of ACSR, differ
in how the plastic elongation and, to a lesser extent, the
linear elongation of the aluminium layers are modelled. At
high mechanical loading and high temperature, the choice
of model affects the calculated conductor sag. CIGRE
Technical Brochure 324 [1] suggests three alternative
mechanical conductor models, each models the steel core
similarly (e.g. a “linear spring”) and the aluminium layers
differently.
The Linear Elastic (LE) model [5] assumes that there is
no plastic elongation in the aluminium layers (the steel
core and the aluminium layers are modelled as parallel
springs). The Simplified Plastic Elongation (SPE) model
allows for plastic conductor elongation but on the basis
of field experience rather than calculation. As explained
in the next section of this paper, for both the LE and
SPE conductor models, the conductor is represented
mechanically with a single composite conductor modulus
(EAC) and thermally with a single composite conductor
Coefficient of Thermal Elongation (CTEAC).
Also, in reference [1], a third conductor model, the
Experimental Plastic Elongation (EPE) model, is defined.
The EPE model is very similar to the “Varney graphical
method” [6]. It models the mechanical and thermal
behaviour of the aluminium layers and the steel core,
separately. The mechanical stress and strain behaviour
of the aluminium layers and the steel core are measured
experimentally as is the plastic elongation of the conductor.
The thermal elongation of the aluminium and steel core
are modelled separately but not typically measured.
The generally non-linear mechanical “stress-strain”
behaviour of the conductor components is represented by
a polynomial equation.
4. Linear conductor thermal
elongation model
The initial sag and tension of bare overhead transmission
conductors change with the conductor length and with
conductor weight per unit length due to ice and/or wind
loading. Conductor length changes as the result of plastic
and elastic elongation and due to thermal elongation.
Elastic elongation and thermal elongation are reversible
(returning to the initial tension and temperature yields the
initial length). Plastic elongation occurs due to permanent
elongation during high tension events and, at normal
everyday tension, metallurgical creep elongation of
aluminium. Thermal elongation is the result of variations
in air temperature, solar heating, and line current.
For bare overhead conductors that are entirely stranded
from aluminium wires (e.g. All Aluminium Conductor),
the conductor’s coefficient of thermal elongation (CTE)
is taken equal to that of the aluminium wires (23E-6 per
oC). For bare overhead conductors that are stranded with
one or more layers of aluminium wires surrounding a
steel (or composite) core, the conductor’s CTE can be
calculated on the basis of the CTE’s and Elastic Modulii
of aluminium and the core material accounting for the
cross section areas of each component.
Where the subscripts A refers to Aluminium, C refers to
the Core, and AC refers to the complete conductor.
Note that for most ACSR conductors, the composite CTE
is closer to aluminium than to steel. For 26/7 ACSR, the
composite CTE is 18.8 whereas the aluminium CTE is 23
and the steel core CTE is 11.5.
5. Non-linear coefficient of
thermal elongation model
The equation for CTEAC is derived by assuming that (1)
both the aluminium layers and the core are under tension

Cigre Science & Engineering • N°7 February 2017

41
 Figure 1: Initial component tensions in aluminium layers and steel core for 54/7 ACSR
in a 300 m span as a function of conductor temperature.
and that (2) the length of the aluminium layers and the
core must remain equal. For an aluminium conductor
with a steel core (i.e. ACSR), suspended in a catenary, the
aluminium strand layers of ACSR elongate at twice the
rate of the steel core.
If the conductor is modelled with the EPE method, the
tension of the steel core and the surrounding aluminium
layers are modelled separately. Therefore, as the ACSR
conductor temperature increases, the total conductor
length and sag increase and the total conductor tension
decreases. However, for the lengths of the steel core
and the surrounding aluminium layers to remain equal,
the greater thermal elongation of the aluminium must
be offset by reduced elastic elongation and the lesser
thermal elongation of the steel core must be increased
by increased elastic elongation. In other words, as the
conductor temperature increases, the percentage of total
tension in the aluminium must decrease.
At a temperature referred to as the “knee-point
temperature” (KPT), the tension in the aluminium layers
of the ACSR conductor has decreased to zero (see Figure
1) and all of the conductor tension is in the steel core.
Above the KPT, the aluminium wire layers go into
compression. If the elastic modulus of the aluminium
layers in compression is equal to that in tension (think
of the behaviour of Aluminium-clad steel wire), then the
CTE remains equal to that at temperatures below the KPT.
If the elastic modulus of the aluminium wire layers is
zero, then the conductor’s composite thermal elongation
rate decreases to that of the steel core alone, depending
upon the compression modulus of the aluminium layers.
In any stranded aluminium conductor, the aluminium
wires undergo both elastic and plastic elongation. With
an all-aluminium conductor, plastic elongation due to
high load events and metallurgical creep, cause both the
everyday sag and the high temperature sag to increase
over the life of an overhead line.
Cigre Science & Engineering • N°7 February 2017
42
With ACSR, the core usually undergoes only a small
amount of plastic elongation during high load events and
the impact of plastic elongation in the aluminium layers is
different. Using the EPE stress-strain model, the amount
of plastic elongation in the aluminium layers can be
calculated as a function of both high tension load events
and creep elongation over time. All three stress-strain
models allow one to calculate the KPT of ACSR, but LE
assumes that it remains constant over time and loads,
SPE assumes that KPT decreases a set amount over time,
and EPE can be used to calculate the KPT of ACSR for
user-specified loading events and, given creep elongation
models, the variation of KPT over the life of the line.
Other factors can also influence the KPT. The initial
KPT of ACSR (or other reinforced conductors) can be
reduced by pre-stressing the conductors prior to sagging
and clipping. Rawlins [7] suggests in some detail that the
KPT can be higher than expected due to residual stranding
stresses in the aluminium wires of ACSR made in modern
rigid-frame stranding machines, which effectively makes
the zero-stress aluminium wire layers shorter than the
zero stress length of the core. Finally, long pulls and
high tensions during tension stringing can produce non-
negligible plastic elongation of the aluminium prior to
sagging. These phenomena are best studied using the EPE
model.
Above the KPT, the continued thermal expansion of
aluminium yields increasing aluminium compression but,
if the aluminium compression modulus is low, the rate of
sag increase with temperature is sharply reduced to near
that of the steel core alone as shown in Figure 2.
The advantage of using the EPE conductor model is that
it yields lower calculated high temperature sags (about a
meter less at 150oC as shown in Figure 2). On the other
hand, because it is not as conservative as the LE/SPE
conductor model, there must be increased concern with
those factors that affect the modelling.
 Figure 2: Sag vs temperature for ACSR with LE/SPE and EPE conductor models

There are three primary concerns with regard to sags
calculated by the EPE conductor model. They involve the
conductor manufacturing details (lay lengths and bobbin
tension), the conductor installation procedures, and the
occurrence of severe mechanical ice & wind load events
prior to the field sag measurements of the existing line.
The KPT depends on the ACSR stranding (Figure 3) and the
effective span length (Figure 4) as summarized in Table 2.
Table 2: “Final” knee-point temperature as a function of ACSR
steel content and ruling span length.
6. Compression of helical
aluminium strand layers
For single-layer and multi-aluminium layer ACSR
conductors, at conductor temperatures above the KPT, the
aluminium wires continue to expand faster than the steel
core wires putting the aluminium layers into compression
and increasing the elastic elongation of the steel core.
Given the physics of the ACSR conductor, aluminium
compression must occur above the KPT.

Ruling Span Length - meters

The aluminium wires in each layer are helices. Under
tension, the modulus of the aluminium wires is close to
Type* of ACSR 180 300 425 10 Mpsi and the strength of these layers is a significant
7 (e.g. 45/7) 90°C 130°C 160°C portion of the composite conductor’s tensile strength.
14 (e.g.54/7) 75°C 95°C 110°C
Under compression, an unconstrained aluminium wire
helix produces a much lower modulus because the wire
23 (e.g. 30/7) 30°C 30°C 40°C bends rather than compresses. That is, the helix expands
* - Type = 100*[AS/AA] radially and the resistance to axial strain is about 5% of
that in tension.
Clearly, high-temperature sag accuracy is more of an
issue with existing lines having relatively short spans and Rawlins paper on High Temperature Sag Calculation [7]
for any line having high steel content ACSR (e.g. 30/19). derives the axial compression modulus (spring constant

Figure 3: Comparison of sag vs temperature for 45/7, 54/7, and 30/19 ACSR Figure 4: Sag vs conductor temperature for 180, 300, and 425 m spans.
with a 300 m ruling span.

Cigre Science & Engineering • N°7 February 2017

43
 Figure 6: Radial expansion of unrestrained helical aluminium wire layers of Drake ACSR
as a function of axial strain.

per unit area of strand), Eff for a free helically-wound

aluminium strand layer as follows:

Where:
D = Overall diameter of strand layer
Where: E is the axial elastic modulus of aluminium d = Strand diameter in layer
(slightly less than 60 GPa). R = Radial expansion of the strand layer
n is the number of strands in the layer. Considering the number of strands in the layer, n, as a
α is the lay angle of the layer. function of strand and layer diameters:
Eff is a weak function of the lay angle and is typically
slightly less than 4 GPa. Therefore, the axial modulus
of the aluminium layers in compression is less than 10%
of the axial modulus in tension unless the layers are Then:
constrained radially.

The relationship between the total length (L) of a strand

and its lay length (λ) is shown in the following diagram. The strain, S, is the same for all layers of the conductor, so
a comparison between the radial expansion rates of each
layer can be made by
using:
Figure 5: Two-dimensional representation of aluminium wire helix dimensions

Where: Where subscript “i” denotes the specific strand layer.

θ = Lay angle For an ACSR conductor with two layers of aluminium
R = Radius of strand helix wires, if the radial expansion rate of the outer layer is greater
L = Helical wire length than or equal to that of the inner layer, then both are free to
λ = Axial length of one turn of the helix (the lay length) expand radially and the modulus of each in compression is
Now, low. Given the range of recommended lay ratios from IEC
or ASTM production standards, use of preferred ratios will
allow free expansion of the aluminium layers.
With 2 or more layers of aluminium, each layer will
The conductor strain, S, is defined as the ratio of the expand similarly if free to do so. Figure 6 shows the
change in lay length divided by the lay length and the dependence of inner and outer aluminium layer expansion
radial expansion of the strand layer due to a change in lay for 26/7, 2-layer Drake ACSR as a function of lay ratio. At
length due to conductor strain is: the preferred ratio, the layers are free to expand radially.

Cigre Science & Engineering • N°7 February 2017

44
 Figure 7: Influence of aluminium compression force and residual aluminium layer stress on EPE
high-temperature sag calculations

Rawlins developed an approximate aluminium layer 7. Experimental measurements

compression modulus for multi-layer ACSR which was
implemented in the SAG10 ™ numerical sag-tension
calculation method. Using the maximum recommended
inner layer lay ratio (16) and the minimum recommended
outer lay ratio (10) for 2-layer, 26/7 Drake ACSR, the
compression modulus was sufficiently high to change
the thermal elongation rate beyond the KPT as shown in
Figure 7.
Note that the existence of significant axial aluminium
layer compression stress changes the slope of the sag
versus temperature curve beyond the KPT while the
existence of residual stress from manufacturing shifts the
KPT value itself but does not change the slope. Higher
than estimated plastic elongation of the aluminium layers
would have an effect similar to that of residual stress but
would reduce the KPT rather than increasing it.
of High Temperature Sag
Field measurements of high temperature sag in operating
lines are extraordinarily difficult primarily because very
few lines carry currents close the line thermal rating under
normal system conditions and because weather conditions
are normally more favourable to heat loss than those
assumed for thermal line rating calculations. In addition,
the small differences in sag produced by variation in
CTE at high temperature are easily overwhelmed by
small variations in end-point movement. Consequently,
experimental measurements have been attempted for
indoor and outdoor test spans.
The point of this paper is to explain and quantify how the
user’s choice of conductor model (LE, SPE, and EPE)
influences the calculation of ACSR sag at high temperature
based on field survey measurements. The result of these

Figure 8: Comparison of measured and calculated sag vs temperature for 26/7 ACSR

Cigre Science & Engineering • N°7 February 2017

45
 Figure 9: Comparison of experimental measurements of sag vs temperature vs
conductor average temperature.
calculations can be important in establishing the thermal
capacity of existing clearance-limited overhead lines and
in determining the best method for increasing the thermal
capacity of such lines.
Two careful laboratory experiments are discussed in the
following paragraphs. Figure 8, shown in Reference [7],
is based on tests performed as part of a study of high
temperature thermal elongation with an indoor, 90 meter
test span. The conductor is 26/7 Hawk (240 mm2) and it
was pre-stressed to simulate the aluminium layer plastic
elongation that would occur as a result of creep and high
load events in an installed line.
This experimental data was analysed by Rawlins [7] and
presented in an IEEE paper which elicited a good deal
of interesting discussion. In essence, Rawlins proposed
that the experimental data could best be explained by
“residual stress” in the aluminium layers produced as a
result of stranding in a rigid-frame stranding machine.
He supplemented this data by measurements of residual
stress in various ACSR conductors prior to installation
and concluded that the KPT should be adjusted upward
by assuming a residual stress of 15 MPa.
A second experimental study was performed in an outdoor
test line (2 spans) at the Oakridge National Laboratory in
Oakridge, Tennessee. The analysis was reported by Seppa
[8] and his plot of the test data is reproduced in Figure 9.
The test facility consisted of two 180 m spans, carefully
terminated to allow accurate sag measurement and with
many thermocouples embedded in the conductors.
The laboratory experiments demonstrate that the measured
sags are less than those calculated with the LE or EPE,
constant CTE model and greater than that calculated
with the EPE/Graphical method where residual stress is
ignored.
Cigre Science & Engineering • N°7 February 2017
46
The experiments at both locations indicate that high
temperature sag calculations using the EPE conductor
model should use an assumption of 15 to 20 MPa for
residual aluminium stress to be conservative on the basis
of this limited experimental data.
8. High Temperature Sag in new
and existing lines
it may be observed that sag at high temperature is seldom
measured in either new or old lines. In both cases, high
temperature sag is calculated based upon sag measurements
made at everyday line currents and weather conditions.
For new lines, field measurements are made after tension
stringing as part of line construction. For existing lines,
measurements are made with the line carrying normal
current (or none), by LiDAR or conventional survey
methods.
With new and existing lines, given non-homogeneous
conductors (both conventional ACSR and HTLS), the
tension distribution between core and aluminium layers is
estimated at the time of measurement.
The EPE analysis presented in this paper is also applicable
to the analysis of high-temperature, low-sag conductors
having either zirconium or annealed aluminium layers
surrounding a core which has high strength and low
thermal elongation.
9. Selection of High-Temperature
Sag model for new lines
when designing new overhead transmission lines,
the height and the placement of support structures is
dependent on wind and ice loads (primarily on the
conductors) and on the sag of the phase conductor at the
templating temperature. If the reduction in slope of sag
versus temperature beyond the knee-point temperature
is neglected, then the structure attachment heights and
spacing will be somewhat higher than need be. The added
cost of somewhat higher sag clearances may be less
onerous in new lines compared to reconductoring existing
lines. For example, the difference in total line cost when
increasing the templating temperature from 50°C to 80°C
has been found to be around 5% in certain projects in
Eskom, South Africa. This is not a generic finding and
the cost will differ depending on the terrain, tower family
being used, conductor types, etc.
In determining the model to use, it is important to
determine the application of the conductor and the likely
operating temperature. If the power on the line has a
fairly flat load profile, the cost of losses will likely lead
the designer to select a larger aluminium area. This will
imply that the operating temperature of the conductor will
be quite low. In this case the linear model may suffice as it
will result in a slightly conservative sag but will not result
in high line costs as the difference between the models
below the knee point is small.
If the line is used to supply a peak load for a short duration
or is designed to operate at a high temperature under n-1
conditions, it is important to understand the exact sag-
temperature relationship of the conductor. If the linear
model is adopted, the cost of the line may be excessive
due to the conservative sag calculation, because it ignores
the non-linear CTE behaviour of ACSR.
10. Selection of High-
Temperature Sag calculation
method for existing lines
The initial step in evaluating methods of uprating existing
overhead lines consists of making detailed measurements
of conductor position and terrain survey for the entire
line. Increasingly, this is done with airborne LiDAR,
though traditionally, by conventional survey methods.
In most cases, these older HV lines are lightly loaded
(electrically) when the system is operating normally, so
the measured conductor position in each span corresponds
to a conductor temperature within 5oC to 15oC of air
temperature. It is necessary to calculate the conductor sag
for the line’s maximum conductor design (“templating”)
temperature and for any higher temperature considered
for uprating purposes.
Cigre Science & Engineering • N°7 February 2017
47
In this case, the choice of high temperature sag model
can be quite important. If the knee-point behaviour is
ignored altogether, then the resulting “conservative”
calculation of sag at high temperature may rule out the
use of existing ACSR conductors at higher temperature
and reconductoring may seem to be the only choice. If
the reduced thermal elongation beyond the knee-point
temperature is considered and properly modelled, then
it may be possible to choose a relatively inexpensive
method of uprating the line by raising certain supports or
re-tensioning the line.
On the other hand, when doing calculations to allow thermal
uprating of an existing line without reconductoring, one
must be careful to recognize the need to model the sag
behaviour correctly and to allow for both calculation
errors and measurement tolerances, both of which are
impossible to avoid.
If the thermal rating is to be determined in real time, it
is important that the actual sag-temperature relationship
is determined by calibration and measurement. This is
covered in TB 498 [9].
11. Use of High Temperature
Low Sag (HTLS) conductors
High temperature low sag conductors are often designed
so that the conductor operates above the knee-point
under normal conditions. This is the case with annealed
aluminium, gapped and some polymer cored conductors.
The rate of sag as a function of temperature is therefore
dependent on the core only and not the aluminium
component. This allows for a high temperature of
operation without exceeding the permissible sag.
If the linear model is chosen with regard to HTLS
conductors, it may indicate that the conductor is not
suitable for certain applications which may not be the
case. This is demonstrated in Technical Brochure 244
[2]. The type of HTLS also needs to be chosen to suit the
application.
In determining the initial cost of the HTLS installation
it is important that an accurate sag-temperature model
be used. This should take the knee-point into account.
Failing to do this may result in the HTLS option being
found less than optimal as the increase in thermal rating
may not negate the higher initial cost.
12. Conclusions
When calculating the sag of ACSR at conductor
temperatures above 75°C, the thermal elongation rate
of the conductor should model the mechanical and
thermal behaviour of the aluminium layers and the steel
core separately. If the conventional ACSR temperature-
independent CTE formula is used, the sag at maximum
line temperature will be overestimated. In designing new
lines, allowing for the modest amount of excess high
temperature sag which results from the linear CTE method
is of minor importance. However, in evaluating options
for thermal uprating of existing lines, the difference in
the linear and non-linear CTE methods can have a large
impact on the uprating method chosen.
If the conductors are intended to operate at temperatures
above 75°C, it is recommended that a model be used that
takes into account the knee-point of the conductor. Failure
to do this may result in a suboptimal conductor being
chosen for the line.
The temperature gradient also needs to be considered as
the average temperature determines the conductor sag and
not the surface temperature [10].
As the conductor temperature increases above ambient,
the aluminium layer thermal expansion rate is twice that
of the steel core and at the “knee-point temperature”
(KPT), the tension in the aluminium layers is zero and
all the tension is in the steel core. Beyond the KPT, the
CTE of the conductor is reduced and the conductor sag
which corresponds to the line’s maximum temperature is
reduced.
The issue of KPT and its impact on the line sag at high
temperature depends on a number of line design factors.
The issue is of minimal concern when:
1. The maximum operating high temperature of the line is
determined by loss of strength in the aluminium layers
rather than sag.
2. The spans are long (>400 m).
3. The conductor has a steel core area less than 7% that of
the aluminium layers.
4. Compression modulus of the aluminium layers is
negligible.
Cigre Science & Engineering • N°7 February 2017
48
The issue is of primary concern when:
1. The maximum operating high temperature of the line
is determined by the clearances produced when the
conductor is at the templating temperature.
2. The spans are short (<400m).
3. The conductor has a steel core area greater than 10%
that of the aluminium layers.
4. Lay ratios are near recommended values.
When the sag of ACSR at high temperature is essential
to the selection of line uprating method, the KPT should
be determined by use of the EPE mechanical conductor
model (including the impact of metallurgical creep of
aluminium and plastic elongation for high tension load
events) and a residual stress of 15 to 20 MPa should be
specified for the aluminium layers.
13. Bibliography
[1] CIGRE Task Force B2.12.3, “Sag-tension Calculation Methods for
Overhead Lines”, Technical Brochure 324, June, 2007.
[2] CIGRE Task Force B2.12, “Conductors for Uprating of Overhead
Lines”, Technical Brochure 244, April, 2004,
[3] IEEE Subcommittee 15.11, “Limitations of the Ruling Span Method
for Overhead Line Conductors at High Operating Temperatures”.
Report of IEEE WG on Thermal Aspects of Conductors, IEEE WPM
1998, Tampa, FL, Feb. 3, 1998
[4] CIGRE WG B2.12, “Alternating Current (AC) Resistance of
Helically Stranded Conductors”, Technical Brochure 345, April,
2008.
[5] IEC 1597, “Overhead electrical conductors – Calculation methods
for stranded bare conductors”, First Edition, 1995.
[6] Varney T., Aluminium Company of America, “Graphic Method
for Sag Tension Calculations for A1/S1A (ACSR) and Other
Conductors”, Pittsburg, 1927
[7] Rawlins, C.B., “Some Effects of Mill Practice on the Stress Strain
Behavior of ACSR”, IEEE Transactions on Power Systems, Vol. 14,
No. 2, April, 1999.
[8] T. Seppa “Fried Wire?” (Public Utilities Fortnightly, December 2003,
pages 39-41)
[9] CIGRE Working Group B2.36, “Guide for Application of Direct Real-
Time Monitoring Systems”, Technical Brochure 498, June, 2012.
[10] Clairmont, B, Douglass, D., “Radial and Longitudinal Temperature
Gradients in Bare Stranded Conductors with High Current
Densities”, CIGRE Paper B2-108, Paris, 2012.
Current situation and recent challenges in
asset management of aging T&D substation
facilities in Japan

T. Kobayashi*, H. Nakakoji, TEPCO Power Grid, Inc., Japan

S. Noguchi, Chubu Electric Power Co., Japan
S. Iwasaki, Kansai Electric Power Co., Japan
T. Takahashi, Central Research Institute of Electric Power Industry, Japan

Abstract the reliability of the electric supply. In order to achieve

In recent years, many initiatives for asset management of
substation facilities are reported from various companies.
These facilities have been installed to accommodate the
increasing power demand, however, most of them are
getting aged. It is required to maintain these facilities
in order to utilize until their end of lifetime and also
maintain the reliability of electric supply. In order to
achieve this, asset management is necessary, and this
includes the methods of decision making and optimizing
the priority for capital investments such as refurbishment
of aging facilities, due to the variation in the aging of the
equipment.
This paper reports current working for the asset
management, and also overview the surveys of CIGRE.
For example, organizing quantitative Health Index to
level the number of aging equipment to be replaced,
standardization of collecting format of failure data for
sharing among electric companies, prediction method of
the failure rate with limited data samples are reported.
Also we report the repair method corresponding to
challenges such as impact on the power system, capability
of construction, and maintaining of technical ability.
1. Introduction
substation facilities have been installed to accommodate
the increasing power demand, however, the timing of
which the facilities are installed varies widely and most
of them are getting aged. It is required to maintain these
facilities to utilize until their lifetime and also maintain
this, it is necessary to optimize the methods of decision
making and priority making for capital investments such
as refurbishment of aging facilities.
In this paper, we are organizing the technical trend
regarding the asset management of transmission facilities.
Also we are reporting the challenges of the asset
management and proposing new methods of the decision
making and the priority making.
2. The surveys of asset management
in Cigre
The surveys of 18 papers regarding the asset management
out of CIGRE technical report (A2, A3, B3, C1 and D1),
reports of CIGRE Paris session 2014, and WG reports
currently working have been carried out. The survey
results have been organized in Figure 1 according to the
asset management process, which consists of 4 phases;
maintenance, evaluation of current situations, prediction
and strategy.
The works of Service Provider are sufficiently reported
including technologies of maintenance/ inspection,
evaluating current situations and lifetime, and surveys of
analysis of faults based on CIGRE technical reports and
current WG report and so on.
Meanwhile, the roles of Asset Manager, such as the risk
evaluation of entire facilities (i.e. Health Index), the
evaluation of economic life, and the selecting methods
of refurbishment or partial repair, are introduced through
some applications.

*kobayashi.takayuki@tepco.co.jp

KEYWORDS
Asset management, Aging transmission facilities, Health Index, Economical life time, Investment strategy

Cigre Science & Engineering • N°7 February 2017

49
 Figure 1. Asset management process and current technical trend
In the asset management, the data of maintenance/
inspection and evaluation of current situation which
are prepared by Service Provider are utilized by Asset
Manager and Asset Owner in their policy decision. Also,
degree of communication between Service Provider and
Asset Owner has to be reinforced.
The challenges in the asset management are identified as
follows;
a. Arrangement of Health Index showing condition
evaluation of each facility
Recently, examples of actual applications examples
have been reported from some countries. Evaluation
items are being identified gradually, however, the way
of weighting them are not developed. Also the Health
Index items of transfomers are practical, whereas those
of circuit breakers are not, due to the lack of practical
external diagnostic technologies.
b. Collection of fault data for failure rate prediction
Although the fault records is important, those data can
not be shared among users because each of them use
original data collecting format.
As the failure rate of transmission facilities is very
low, a large amount of data samples are required to
accurately predict the failure rate in the future. And it is
necessary to have a standardized format to collect and
share failure records.
c. Prediction of failure rate
It is necessary to know the timing and the speed of
incline of the failure rate bathtub curve for an accurate
prediction. However, the prediction method of the failure
rate based on limited data samples is not developed.
Cigre Science & Engineering • N°7 February 2017
50
d. Setting of life expectancy and determination of
refurbishment and repair
Except for transfomers, the lifetime evaluation method
of facilities has not been developed. The lifetime
expectancy of facilities can be empirically estimated as
shown in Table 1 [1], but there is a gap between 10 to
40 years against the real value. In addition, quantitative
determination criteria tor refurbishment and repair has
not been developed.
Table 1 Lifetime expectancy of each facility [1]
3. Challenges of asset
management in japan
As discussed in the last chapter, there are still challenges
in asset management such as development of the Health
Index system, standardization of data collecting format,
and formulation of a prediction method for failure rate.
Also, to perform the asset management for substation
system as a whole, it is necessary to consider challenges
such as the impact on the power system, capability of
construction, and maintaining of technical ability.
In Japan, electric power companies, manufacturers,
universities and research institutes are cooperating to
respond to the challenges mentioned above. We will
report the activities in this paper.
3.1. Index showing condition evaluation of each On one hand, the failure rate of transmission facilities is
component (Health Index) low, and to be able to predict the failure rate, large amount
of facility data is needed. Accordingly, it is necessary
Transmission facilities, even if they are the same type, to collect and analyze data from not only one power
are subjected to different stresses caused by not only company but all power companies worldwide.
the differences of their rated values and duties, but also
their environmental conditions (impact of ambient In Japan, 11 users and 9 manufacturers are cooperating in
temperatures, salt pollution and so on), load factor, and the standardization of input data, collecting and analyzing
the number of operations. In the asset management, facility/fault data based on a unified definition. Table 2
transmission facilities have to be evaluated quantitatively shows the template and items of a maintenance record
and prioritized using an index considering not only the for a transfomer which was standardized in 2009 [2]. It
aging of facilities, but also their current conditions and includes items such as specification, operation record,
inherent risks. failure record, repair schedule, dissolved gas analysis
data, inspection results, test/diagnosis results.
Table 2 shows the Health Index for 66 kV distribution
transformer used in a Japanese electric company. In In order to predict future failure rates accurately, it is
addition to the conditions of current facilities such as age, necessary to standardize a platform internationally, and
result of dissolved gas analysis and oil leakage, evaluation to consolidate the each of the collected data as showing
Table 3.
is done for specific facilities, using index points, set based
on possible repair cost and impact of inherent risk items Table 3 Items for database of transfomers
according to its failure record.
Category Items for database
Table 2 Health Index items for 66 kV distribution transformers Facility data Manufacturer, type, serial number, year
Index Points Notes of manufacture, rated value (voltage,
capacity, etc), type of conservator, amount
Age n 1 point/operating year of insulating paper, oil capacity
Dissolved 10 Over Alarm level II Operation record Load factor, ambient temperature,
gas analysis environment
in oil
Trace PCB in 6 Exceeding defined limits Inspection data Results of inspection (oil leakage, crack/
insulating oil deformation, oil level, protective relay
Oil leakage 5 Leakage from one area operation),
from Results of dissolved gas analysis
10 Leakage from several areas Results of electric test (ratio, winding
transformer
20 High loss transformer using Hycar resistance, low voltage excitation current,
cork for gasket, and has leakage from partial discharge, characteristic of insulation
several areas. oil, LTC operation, insulation resistance,
Specific 47 Winding configuration with high charged potential, FRA)
equipments probability of lightning-caused failure
Repair record Details of repair
5 Risk of core corner damage due to
aging Fault record Details of fault, cause of fault, measurements
10 RIP bushing of Company A
(Risk of insulation breakdown due to
aging) 3.3. Prediction method of the failure rate
3 RIP bushing of Company B
(Risk of insulation breakdown due to In order to grasp the failure rate and economic life
aging) (planned refurbishment more advantageous in regards
LTC with no 5 Transformer using LTC with no to cost compared to carrying out repairs by Corrective
replacements replacements
Maintenance (CM)) of the current facilities, it is required
to evaluate quantitatively the increase in failure rate
3.2. Consolidation of facility/failure data caused by equipment aging. Figure 2 shows the results
of fault analysis of a vacuum circuit breaker of a 6 kV
In asset management, it is necessary to predict the
switchgear in a Japanese electric company.
future failure rate transmission facilities, and to do that,
collection of failure data is required. The hazard function in Figure 2 is used to express the

Cigre Science & Engineering • N°7 February 2017

51
 (a) Hazard plot (b) Hazard and Failure rate (c) Cumulative Hazard and Failure rate

Figure 2 Fault analysis of vacuum circuit breaker of 6 kV switchgear

failure rate of an operating equipment at the referenced By multiplying repair cost to this result, it is possible to
year, and is approximated by the Weibull distribution as estimate repair cost by CM and evaluate the life cycle
per the following; cost, and in turn leading to strategic decision making.

dR(t)
λ(t) = ¯ ____ /R(t) λ(t): Hazard function 3.4. Asset managemant as a substation system
dt
R(t): Reliability function* When it comes to the asset management as a substation
R (t) = 1 – F (t) F(t): Unreliability function system, not only the current conditions and priorities of
each facility, but the typical challenges in substations
F (t) = 1 –exp{–(t/ts)m} F(t): Cumulative failure must be factored in as we. These challenges include
rate by Weibull function power system factors caused by plan or change in power
m: Shape parameter demand, coordination of the number of field workers and
ts: Scale parameter maintaining technological competance, and the increase
Current facilities have been operating for approximately in obsolete type equipment.
30-40 years and by applying the hazard function to the In Japan, for transformers and early type gas insulated
aging distribution, the future failure rate can be obtained switchgears (GIS), guidelines for life extension, partial
as a bathtub curve. In the case of fault example in Figure repair, and refurbishment of substation system are
2 (c), a rapid increase in faults after about 40 years of determined upon assessment of the substation’s impact to
operation is predicted. the power system.

Figure 4 Repair Guideline of early GIS

Cigre Science & Engineering • N°7 February 2017

52
 Table 4 Evaluation items for refurbishment priority

Figure 4 shows the guideline for the early type GIS
[3]. First step is assessment of equipment condition.
In the second step, life extension or refurbishment is
decided, considering the possibility of life extension.
In the next step, it is assessed decided once more from
the viewpoint of economic evaluation. Refurbishment
priority is decided based on evaluation items in Table 4.
In the refurbishment planning step, site condition, system
reliability, standardization of investment, and capability
of construction are considered.
4. Examples of asset management
Using the approach reported in Chapter 3, the priority of
transmission facilities can be decided by not only aging
but also by indexing the condition of facilities, impact of
faults and economic loss. Figure 5 shows an example of
adopting this idea to a GIS installed in a Japanese electric
company [4]. For each manufacturer’s GIS type, risk
items such as maintainability of aging facilities, presence
of implementation of countermeasures to same type
equipment, failure rate, and production of discontinued
parts, are evaluated by scoring points in Table 5. In
addition to leveling the number of planned replacements,
by utilizing the characteristics of each manufacturer’s
facilities, an appropriate maintenance is made possible
based on the facilities’ soundness. Furthermore,
cooperation of facilities countermeasures and regular
maintenance can be realized.
Moreover, in the future, combining the asset management
methods to the capital management system will link
Service Provider and Asset Owner/Asset Manager more
strongly.

Figure 5 GIS refurbishment schedule based on the risk management in Japanese electric company

Cigre Science & Engineering • N°7 February 2017

53
 Table 5 Examples of evaluation for each GIS type
5. Conclusion
In Japan, the economic lifetime is evaluated from
prediction of future failure rate and evaluation of Health
Index, and refurbishment schedule is formulated based
on these evaluations. By these methods of the asset
management, effectiveness of investment which takes
into consideration of life cycle cost can be visualized,
and quantitative explanation to Asset Owner and
Stake Holders can be made. In addition, the planned
refurbishment can be leveled, and limited fund and
human resources can be utilized. It is beneficial not only
for electric companies, but also for all stakeholders such
as manufacturers and construction companies.
In the 2015 report [5] published by IEC Market Strategy
Board, it is reported “A centralized international
database of electricity equipment failures and historical
performance records would bring massive benefit to
all stakeholders.” In addition to deepening technical
knowledge considering the asset management in
CIGRE, it is necessary to develop advanced proposals
by uitilizing knowledge of CIGRE.
6. References
[1] Working Group B3.03 CIGRE, TB300 “Guidelines to an Optimized
Approach to the Renewal of Existing Air Insulated Substations”
(August 2006, pages 56-58).
[2] Electric Technology Research Association, Vol.65, No.1, “Guideline for
Refurbishment of Electric Power Transformers” (September 2009,
pages 224-225, in Japanese).
Cigre Science & Engineering • N°7 February 2017
54
[3] Electric Technology Research Association, Vol.70, No.2, “Advanced
Maintenance Strategies for Gas Insulated Switchgear” (August 2014,
pages 17-18, in Japanese).
[4] CIGRE Paris International Conference in 2014, B3-213, “Optimized
Replacement Strategy of Substations equipment Considering Risk
Management” (pages 5-6).
[5] IEC Market Strategy Board(MSB), WP Asset Management, “Strategic
Asset Management of Power Networks” (October 2015, pages 6).
7. Biographies
Takayuki Kobayashi is a senior manager of substation
engineering of TEPCO Power Grid, Inc. He joined Tokyo
Electric Power Co., Inc., in 1984. He has worked to
develop 500kV substation equipment and the diagnosis
techniques for substation equipment.
Hajime Nakakoji joined Tokyo Electric Power Co., in
2003. He has been engaged in designing substations and
technical development work.
Shinki Noguchi, in 2006, he joined Chubu Electric Power
Co., Inc., where he has been engaged in maintenance in
substation.
Shinya Iwasaki, in 1995, he joined Kansai Electric Power
Co., Inc., where he engaged in maintenance in substation.
Tsuguhiro Takahashi is a senior research scientist of
the Electric Power Engineering Research Laboratory,
CRIEPI., where he has been involved in research on
optical sensing and diagnosis techniques for high-voltage
equipment.
Brazilian experience regarding interactions
between series capacitors and svcs – Main
challenges of Tucuruí-Macapá-Manaus
interconnection project
A.R.M. Tenório*, A.A. Nohara, A.F.C. Aquino
Operador Nacional do Sistema Elétrico - ONS
Brazil

Abstract erection of a 500 kV double circuit, on the same tower,

This article describes the issues related to all Tucuruí-
Macapá-Manaus Interconnection’s (TMMI) SVCs
involved and the main findings regarding the possible
instabilities caused by subsynchronous oscillations
(SSO) as well as the solution adopted by each SVC
manufacturer in a heavily series-, shunt-compensated
system. Although the solutions are conceptually
different, both of them fulfill the requirement for the
SVCs to stop responding to sideband frequencies as
far as possible in addition to being prepared for further
action if a new contingency occurs subsequently. The
strategy to prevent hunting among the SVCs has been
described and its deployment explained. One can state
that the solutions adopted have shown to be robust to
provide both a flexible and safe operation for those
SVCs running on TMMI. The operational experience of
TMMI with its SVCs has been shown to comply with
integration and real time simulation (RTS) studies.
connecting the Tucuruí Hydropower Plant to the city of
Manaus, forming the long 500 kV line Tucuruí - Xingu
- Jurupari - Oriximiná - Silves - Lechuga with a total
length of 1418 km. To supply the state of Amapá, a 230
kV double circuit, on the same tower, was recommended
from Jurupari to Laranjal and further to Macapá, totaling
334 km. The whole project, called Tucuruí-Macapá-
Manaus interconnection, was divided for the purpose
of auctioning in three lots involving 18 FSCs, with an
average series compensation level of 70%, and four
SVCs, three of them located on the 500 kV trunk (one
SVC rated at -200/+200 Mvar and two SVCs rated at
-200/+300 Mvar) and the last one (rated at -100/+160
Mvar) in the 230 kV Macapá Substation. Figure 1 shows
a power system map of this interconnected system.

1. Introduction
The power supply for both Amazonas and Amapá States,
located in Amazon region - Brazil, was done through
isolated systems under concession of Eletronorte, as
well as distribution and generation entities before 2012.
Several studies have been conducted in recent decades
to incorporate these isolated systems into the Brazilian
Interconnected Power System (BIPS). In 2003, a study
Figure 1 - Power system map of TMMI project highlighting
conducted by the then Planning Authority concluded the different auction’s lots
the attractiveness of the incorporation of these isolated
systems of the Amazon region to BIPS. Techno- A better visualization of TMMI can be shown in Figure
economic feasibility studies concluded later that the best 2. TMMI passes along three federation states, namely:
alternative for integrating these isolated systems was the Pará (PA), Amapá (AP) and Amazonas (AM). The supply

*ricardo.tenorio@ons.org.br

KEYWORDS
FSC - Hunting – Oscillations – Subsynchronous – SVC

Cigre Science & Engineering • N°7 February 2017

55
 Figure 3 – Simplified single line diagrams for both Jurupari and Silves SVCs

of 18 FSCs by a single manufacturer is considered one
of the world's largest to date. The support of 4 SVCs
adds some more control and flexibility for operation but
it comes along with some complexity to run and is prone
to some control instability due to SSO to be discussed in
this article.
Figure 2 – Single line diagram of TMMI
2. Considerations on TMMI SVCs
and FSCs
2.1. Basic characteristics of TMMI SVCs
The main characteristics of TMMI SVCs are as follows:
i. Jurupari SVC – 2 TCRs rated at 149.6 Mvar each;
2 TSCs rated at 134.6 Mvar each; 2 single tuned
filters (5th harmonic) at 20.6 Mvar each – 60 Hz
and 18.3 kV (secondary voltage). Output -200 to +
200 Mvar on 500 kV, 13.2% coupling transformer
reactance (200 MVA).
ii. Oriximiná SVC – 2 TCRs rated at 179.5 Mvar each;
2 TSCs rated at 212.5 Mvar each; 2 single tuned
filters (5th harmonic) at 23.9 Mvar each – 60 Hz
and 15.5 kV(secondary voltage). Output -200 to +
300 Mvar on 500 kV, 15% coupling transformer
reactance (300 MVA).
iii. Macapá SVC – 1 TCRs rated at 152.1 Mvar each;
1 TSCs rated at 190.6 Mvar each; 2 single tuned
filters (5th harmonic) at 25.7 Mvar each – 60 Hz
and 15.5 kV (secondary voltage). Output -100 to
+ 160 Mvar on 230 kV, 15% coupling transformer
reactance (160 MVA).
iv. Silves SVC - 2 TCRs rated at 147.6 Mvar each; 2
TSCs rated at 129.4 Mvar each; 2 single tuned filters
(5th harmonic) at 36.8 Mvar each – 60 Hz and 20 kV
(secondary voltage). Output -200 to + 300 Mvar on 500
kV, 15% coupling transformer reactance (300 MVA).
Figure 3 shows the simplified single line diagrams of
Jurupari and Silves SVCs. Note that Jurupari, Oriximiná
and Macapá SVCs, which belong to lots A and B, were
manufactured by one single manufacturer, identified
here as manufacturer I. Silves SVC, which belongs to lot
C, was manufactured by another manufacturer, named
here manufacturer II. Note that Jurupari and Oriximiná
SVCs are of similar type with two TCRs/TSCs, while
Macapá SVC has only one TCR/TSC. All 4 SVCs have
in addition to the main controller (PI controller) some
control features, as follows: i) stability controller; ii) gain
optimizer; iii) undervoltage strategy; iv) overvoltage
strategy; among others.
2.2. Basic characteristics of TMMI FSCs
It is worth noting that manufacturer I also supplied 18
FSCs that compensated the 500 kV TMMI transmission
lines by 70% on average. The FSC locations are shown
in Figure 2. All TMMI FSCs are equipped with MOVs
and triggered spark gaps, as shown in Figure 4. The
FSCs are rated at 35% of the lines reactance, totaling
70% per line, except the FSCs installed on the 500 kV
Tucuruí-Xingu lines (circuits #1 and #2), at Xingu side,
which are rated at 70% of the lines reactance.
Figure 4 – Simplified single line diagrams of TMMI FSCs

Cigre Science & Engineering • N°7 February 2017

56
 Figure 5 – Illustration of resonances in a typical series-shunt-compensated power system
2.3. General system requirements specified in
auction’s technical annex
It is important to highlight that some system requirements
have been specified in the Technical Annex related to
Lots A, B and C’s auctions. Those requirements have
ensured that additional studies had to be conducted to
better understand and enhance the SVCs performance
under system disturbances. The most important system
requirements are listed as follows.
i. The SVC control systems shall not compromise
the BIPS performance, either in normal operation
or under contingencies, emergencies and degraded
operation, for steady state and transient conditions.
ii. The SVCs shall not give rise to ferroresonance
conditions or asymmetric core saturation.
iii. The SVC control shall be designed to prevent
hunting with controls from other electrically near
SVCs. The switching operations of SVC branches
or external elements, ordered by the control system
including also on-load tap changers of transformers,
shall not lead to intermittent oscillations (hunting)
among these elements, nor between any of them
and the SVC control.
iv. The SVC control system shall be designed in order
to help minimize the disturbances in the electrical
system for a fault. The control system shall be
designed considering the need for single-phase
autoreclosure without tripping the SVCs.
v. The SVC control shall take into account additional
quantities (power flow, reactive power flow, etc.)
with the aim of phase-modulating (if necessary)
the SVC output towards dampening voltage
oscillations, power oscillations in the power grid
and subsynchronous resonances.
vi. The SVC control shall be designed in order not to
affect the stability of the grid voltage. This shall
be done by identifying the sensitivity of the grid
voltage to the variation of the SVC susceptance,
and taking corrective action to avoid instability
conditions.
Cigre Science & Engineering • N°7 February 2017
57
vii. It shall be demonstrated the performance of the SVC
for operation in normal and degraded conditions
through digital studies, such as electromagnetic
transients studies, and both transient and dynamic
stability studies, to be conducted by the transmission
entity.
3. Series-compensated AC
systems and SVCs – Possible
adverse interactions
3.1. An overview on main resonances in an AC
series-compensated power system
The series compensation by fixed series capacitors in extra
high voltage systems has been recognized as an efficient
and powerful means to improve transient stability and
thereby increasing the transmission capacity between the
power systems, particularly in inter-regional AC links.
However, fixed series capacitors cause subsynchronous
and supersynchronous oscillations in power systems due
to several resonances in power systems.
Figure 5 shows two series-compensated lines in a typical
series-shunt-compensated power system.
The part of a series-shunt-compensated power system
shown in Figure 5 is useful for illustrating the main
resonances [2], [3], namely;
i. Line shunt resonance (dotted black rectangles) – this
type of resonance is well known and low frequency
but yet above fundamental (synchronous) frequency.
It comes from a parallel resonance between series
inductances and shunt capacitances of transmission
lines.
ii. Series resonance (dotted blue rectangle) – this
resonance is due to the combined effect of series
inductances of lines, series capacitors and generators.
The frequency of this resonance increases with the
degree of series compensation. This series resonance
causes subsynchronous components that may impact
adversely turbo-generators. These phenomena are
named subsynchronous resonance (SSR).
iii. Shunt reactor resonance (dotted red rectangles) –
this parallel resonance is developed between shunt
reactors and series capacitors. This resonance creates
subsynchronous components that may impact
adversely on the SVCs. These subsynchronous
oscillations are usually poorly damped.
3.2. SVC voltage measurement under system
disturbances
After a disturbance electrical modes related to the above
resonances are excited and added to the system currents
and voltages at the fundamental frequency (fo). Note that
the power system starts developing a power swing, then
the voltage phasor related to the fundamental frequency
experiences an amplitude modulation process so that
if a frequency f1 has been excited, it will create two
sideband signals, i.e. fo-f1 (subsynchronous) and fo+f1
(supersynchronous). These sideband signals will be seen
by the voltage regulators of existing rapid controllers
such as SVCs, which should essentially see the 60Hz
signal (demodulation process). SVCs try to interact with
these SSO resulting sometimes in unstable behavior.
These SSO issues and interaction with SVCs have
already been investigated and reported in the 1990s on
the 735 kV network of Hydro-Quebec in Canada [2], [3].
3.3. Subsynchronous resonance and subsynchronous
oscillations
After a disturbance in a series-compensated power
system subsynchronous currents at electrical frequency
(fe) are excited in the stator, which give rise to currents
in the rotor in subsynchronous frequencies (fm1 = fo-
fe) and supersynchronous frequencies (fm2 = fo +
fe) [4], [5]. According to the IEEE Subsynchronous
Resonance Working Group [5], SSO is “an electric
power system condition where the electric network
exchanges significant energy with a turbine-generator at
one or more of the natural frequencies of the combined
system below the synchronous frequency of the system
following a disturbance from the equilibrium”. The term
SSR encompasses the electromagnetic subsynchronous
oscillations associated with turbine-generator shafts and
a series capacitor compensated power system when the
oscillatory energy exchanged tends to grow. Due to the
existence of several excitation modes on long mechanical
shafts of the turbo-generators this phenomenon requires
Cigre Science & Engineering • N°7 February 2017
58
careful analysis with thermal generators. Hydro
generators have a turbine-rotor system much simpler
usually not developing issues with series-compensated
power system.
SSR can be further subdivided into: i) torsional
interaction; ii) induction generation effect; iii) transient
torque. It is not the objective of this article to deepen
the discussion in terms of SSR, and even taking into
account that the generation entities were informed of the
potential risks with respect to their thermal generators;
two cases based on the ATP program were made
available to allow them to investigate into this matter.
However, so far no complaints have been received
about SSR issues on TMMI. In addition, these SSR
phenomena may involve risks on the generators and
possible life time abbreviation of their mechanical shafts
(fatigue phenomena). Due to this fact SSR studies are the
responsibility of the generation entities. Nevertheless, it
is believed that the risk of possible SSR issues is very
low as thermal generators are connected to different
voltage levels from the 500 kV compensated trunk. The
vast majority of thermal generators are connected to the
network at 69 kV through step-up transformers, except
Cristiano Rocha thermal power plant (5 generators rated
at 21.3 MVA) which is directly connected to the 230 kV
network of Manaus.
4. Identification of SSO issues on
TMMI
As a consequence of 70% series compensation on
TMMI and taking into consideration the technical
literature [2], [3], some interactions between SVCs and
passive reactive components, e.g. series capacitors, are
expected. The first sign of these issues came through the
transmission entity responsible for lots A and B in 2012.
In order to illustrate the SSO issues, the transmission
entity (lots A and B) showed some RTS results in which
one can see clearly poorly damped oscillations after the
stability controller of Jurupari SVC tried to decrease
the SVC gain to almost zero and oscillations remained
poorly damped. These simulations were performed as
part of the Factory Acceptance Tests (FAT) and they
used hardware-in-loop configuration, i.e. a replica of the
control cubicles interfaced with an RTS network. The
SVC had thus its PI controller gain almost zeroed due
 Figure 6 - Jurupari SVC response to a three-phase fault applied at Jurupari 500 kV terminal (standard filtering versus new filtering
strategy) - RTS study conducted by the manufaturer during FAT

to an instability caused by the SSO signal modulation

affecting the measured voltage seen by this SVC. In this
situation, the SVC did not have any strategy to return
to voltage control mode in a reasonable time after this
disturbance. It is important to highlight that during
these oscillations the Jurupari SVC took part interacting
adversely with these SSOs leading its gain to almost
zero (open loop control) with no strategy to return to
closed loop control. This is highly non-desirable and
unacceptable taking into account the auction’s technical
annex (see section 2.3) [4] and the Grid Code [1].
Figure 6 shows this SVC undesirable behavior depicting
variables with a red trace. The following variables
Figure 7 – Silves SVC response to a three-phase fault applied at Silves 500
are shown: VACT – measured system voltage, QREG
kV – RTS study conducted by manufaturer II during FAT
– SVC voltage regulator output, GAIN – SVC gain
(PI controller), IOF – initiation of fault. Note that in
The least oscillatory SVC response was verified for
Figure 6 is also shown the same variables when a new
alternative S3. However, bypassing all 8 series capacitors
filtering strategy is used to mitigate this problem. This
belonging to lot C is undesirable and unacceptable from
corresponds to the black trace and the variable VACT_ system point of view. It was stated that an adequate
SW shows precisely the instant at which this new solution would be required where lot C series capacitors
filtering strategy was activated. Detail of this strategy should remain in operation. The solution adopted by
will be seen in section 5.1. Silves SVC manufacturer is explained in section 5.2.
Regarding Silves SVC, adverse interactions were also
observed between this SVC and the series capacitors 5. Solutions adopted by SVC
particularly under lower short circuit power at Silves manufacturers
substation after clearing faults. This was seen during
FAT in 2012. These oscillations were poorly damped but The solution adopted by the manufacturer of Jurupari,
still stable. Oriximiná and Macapá SVCs was different from the
solution adopted by the Silves SVC’s manufacturer.
Figure 7 shows the Silves SVC response during FAT
using different alternatives before the activation of 5.1. A new digital filter for the measurement
the undervoltage strategy: S1 (standard) – 20 ms time of Jurupari, Oriximiná and Macapá SVCs
delay, Bref = 0 pu; S2 – 40 ms time delay, TSC blocked (Manufacturer I)
for 200ms, Bref=0.65pu, all series capacitors of lot C Due to the large number of series capacitors installed
bypassed; S3 – same as S2 in addition to reinserting on TMMI and considering the intact and degraded
series capacitors (lot C) 200 ms after the undervoltage conditions of this network, manufacturer I calculated
strategy is released. The variable shown are: UP1 – subsynchronous frequencies in the range of 2 Hz up to
primary voltage (pu), VRESP (measured voltage), BREF 50 Hz and supersynchronous frequencies in the range
(SVC susceptance), and REL_GAIN (pu, SVC relative of 80 Hz up to near 120 Hz. Taking into account these
gain). frequencies it was decided to use digital band-pass filters

Cigre Science & Engineering • N°7 February 2017

59
applied to VACT (three-phase average value used to
control the system voltage).
These new filtering responses are much slower than the
standard ones. Therefore, they shall be used only when
SSOs were detected in the network in such a way as to
try to destabilize the SVCs. Network instabilities are
detected through the stability controller behavior. As
shown in Figure 6, this is not effective to damp these
oscillations out in many different network conditions.
Therefore, the new band-pass filters shall be activated to
prevent instabilities to occur. The SVC instabilities are
evaluated on its system voltage measurement and voltage
regulator output. Therefore, when SSOs are detected the
new filtering is activated. The deactivation is done when
SSO dies out. However, the longest activation time
allowed for these new band-pass filters is 10 seconds.
The response time of the voltage controller is acceptable
since it has been applied for a short time under system
oscillations. As can be seen in Figure 6 (curves in black)
when these new digital filters are activated the SVC
stops responding to SSO.
5.2. An SSO function for Silves SVC
(Manufacturer II)
Taking into account that the need for Silves SVC
to possibly damp SSOs was not specified but it was
required that its control should not amplify these
oscillations, Silves SVC’s manufacturer developed an
algorithm called SSO function. This function had the
purpose of freezing the SVC voltage regulator by means
of a considerable temporary reduction of its integral
gain whenever SSOs start disturbing the SVC stability.
In this sense, during the SSO a natural damping process
is provided by the power system itself, if the SSO
function is active, and a major disturbance occurs, i.e.
500 kV system voltage outside 0.95-1.10 pu range, the
SVC shall be able to respond and help the system keep
controlling its voltage at PCC by immediately returning
to voltage control mode.
The SSO function calculates a signal representing the
amplitude of subsynchronous oscillations in the range 5
to 55 Hz, based on measurements of the 500 kV system
voltage. In the algorithm, the voltage regulator of Silves
SVC shall be frozen if all following conditions are met:
Cigre Science & Engineering • N°7 February 2017
60
i) this SSO signal exceeds a threshold value for longer
than a set time; ii) gain supervisor (stability controller)
starts operating; iii) the system voltage is within 0.95-1.0
pu range. Therefore, when the SSO function is active, it
stops any possible feedback between the SVC voltage
regulator and SSOs.
When the SSO signal returns to a value below the
threshold value for longer than a set time, the SVC
voltage regulator is thus released for normal operation,
its voltage regulator gain is then ramped to its original
value within 500 ms. The longest time the voltage
regulator can be kept frozen is 10 seconds. It is worth
highlighting that if during the time the voltage regulator
is frozen the voltage goes outside of the range 0.95-1.10
pu, the SVC voltage regulator returns immediately to the
voltage control mode.
6. Strategy to prevent hunting
among SVCs in nearby operation
This section describes the strategy implemented in all
three SVCs installed on 500 kV TMMI. All three SVCs
have gain optimizers in order to ensure stable operation
under lower short circuit capacity. These SVC voltage
regulators were set during real time digital studies to
control their own primary voltages. Consequently when
a disturbance occurs on TMMI, all three SVCs respond
to some extent according to their voltage regulators. In
addition, they may interact adversely and react amongst
themselves due to relatively high controller gains on
lower short circuit power configurations. This behavior
shall be prevented to ensure a proper and stable operation
of these SVCs. These adverse interactions, called
hunting, may severely have an impact on the SVCs
stability. To mitigate these interactions, when more than
one SVC is operating, it is necessary to implement a gain
reduction scheme that takes into account the number of
SVCs in nearby operation.
Therefore, it is necessary to keep communication
among all three SVCs. Particularly for TMMI this
communication is deployed through an EMS/SCADA
system that exchanges basically two signals among
SVCs: i) SVC is on and operating in voltage control
mode; ii) SVC life bit for supervision of communication.
This communication does not need to be fast and it
 Figure 8 – Silves SVC response to an external system disturbance
flashes in a 10 second-period. The gain reduction scheme
can be summarized in Table I.
Table I – Gain reduction scheme for SVCs on 500 kV system
Number of SVCs in nearby operation in 1 2
Voltage Control Mode
Gain factor applied to Jurupari and Oriximiná 1.00 0.60 0.36
SVC controllers (pu)
Gain factor applied to Silves SVC controller (pu) 1.00 k1
remark: k2=k12 and k1 is a function of the short circuit power at 500 kV
Silves busbar
7. TMMI operational experience -
Contingencies occured on TMMI
It is worth remembering that TMMI can supply up to
50% of the Manaus and Macapá load in order to ensure
that the underfrequency load-shedding scheme is able
to recover the load-generation balance of both Manaus
and Macapá systems. This requirement is associated to
the system dynamic performance regarding the balance
between load and generation during system islanding
after multiple contingencies.
Since the commissioning of TMMI SVCs, there have
been a large number of disturbances such as line outages,
internally and externally to this interconnection, resulting
sometimes in islanding of Manaus and Macapá systems.
Figure 8 depicts Silves SVC performance during a
system disturbance (severe fault outside TMMI).
Figure 9 – Oriximiná SVC response to a single-phase-to-ground fault
Cigre Science & Engineering • N°7 February 2017
Once the SVC controller identifies SSOs on its voltage,
it starts checking the condition for enabling the SSO
function. Note that the first gain reduction (RELATIVE_
GAIN_VREG) is caused by the stability controller
3
(0.70 to 0.44 pu) and an oscillation around 35 Hz is
seen by the voltage regulator (VRESP). Therefore, as
soon as the triggering conditions of the SSO function
are met, the relative gain is temporally reduced from
k2 0.44 pu to pratically zero, freezing the SVC controller
as a consequence. At the time that the SSO signal falls
below a set threshold, the Silves SVC’s relative gain is
ramped back within 500 ms to its original value set by
the stability controller (0.44 pu), enabling the Silves
SVC to operate back in voltage control mode. During
this return, the SSO signal is still monitored and if
it becomes higher than the threshold level, the SSO
function is resumed, i.e. the voltage regulator is frozen
once more.
Figure 9 depicts the Oriximiná SVC controller behavior
during a single-phase-to-ground fault on the 500 kV
Oriximiná-Jurupari line (circuit #1), which led to the
activation of the new digital filtering. An oscillation
around 12 Hz can be measured by the SVC voltage
regulator in the VACT signal. Therefore, as long as the
conditions to enable this new digital filtering are met,
Oriximiná SVC gain is reduced by 50% and it starts
running on this feature within the time frame of the
oscillography.
61
8. Conclusions
The TMMI project has shown to be a complex one due
to the integration of three transmission lots into BIPS by
utilizing a multi-vendor solution for SVCs. Therefore,
this interconnection supplies power to both Manaus
and Macapá systems, which were previously isolated
systems. This encompasses 1418 km and 318 km of
transmission lines (double circuit) on 500 kV and 230
kV respectively over an area of rainforest passing along
three federation states.
This article has described the issues related to all
TMMI SVCs involved and the main findings regarding
possible instabilities caused by SSOs as well as the
solution adopted by each SVC manufacturer. Although
the solutions are conceptually different, both of them
fulfill the requirement for the SVCs to stop responding
to sideband frequencies as far as possible in addition to
being prepared for further action if a new contingency
occurs subsequently. The strategy to prevent hunting
among SVCs has been described and its deployment
explained. One can state that the solutions adopted have
shown to be robust to provide both a flexible and safe
operation for those SVCs running on TMMI.
The operational experience of TMMI with its SVCs has
been shown to comply with integration and RTS studies.
Some real disturbances have been recorded and they
confirm the adequacy of SVCs to meet TMMI's system
requirements. After the commissioning work and start
of commercial operation of TMMI, no power transfer
limitation due to unforeseen aspects has been observed.
The results show the importance of both integration and
RTS studies, especially in a multi-entity environment,
as they provide the opportunity to identify and solve
systemic problems involving different transmission
entities and manufacturers.
9. References
[1] ONS (Brazilian Electrical System Operator), Procedimentos de
Rede – Submódulo 23.3 – Diretrizes e Critérios para Estudos
Elétricos – Rev. No. 2.0 (http://apps05.ons.org.br/procedimentorede/
procedimento_rede/procedimento_rede.asp).
[2] Larsen, E. V.; Baker, D.H.; Imece, A.F.; Gerin-Lajoie, L.; Scott,
G., “Basic Aspect of Applying SVC’s to series-compensated AC
Transmission Lines”, IEEE Transactions on Power Delivery, Vol. 5,
No.3, July 1990.
Cigre Science & Engineering • N°7 February 2017
62
[3] Sybille, G.; Gérin-Lajoie, L.; Giroux, P., “Interactions between Static
Var Compensators and Series Compensation on Hydro-Québec
735kV Network, Canadian Electrical Association, SVC & HVDC
Committee, March 1992, Vancouver.
[4] ANEEL (Brazilian Regulator), Edital do Leilão 004/2008 – Anexos
Técnicos 6A, 6B, 6C – Interligação Tucuruí-Macapá-Manaus (http://
www.aneel.gov.br/aplicacoes/editais_transmissao/documentos_
editais.cfm?IdProgramaEdital=62).
[5] IEEE Subsynchronus Resonance Working Group, “Terms,
Definitions and Symbols for Subsynchonous Oscillations”, IEEE
Transactions on Power Apparatus and Systems, Vol. PAS-104, No.
6, June 1985, pp 1226-1334.
10. Biographies
Antonio Ricardo de Mattos Tenório received his
B.Sc. Degree with honors in Electrical Engineering
from Federal University of Pernambuco, Brazil in 1982,
and his M.Sc in Electrical Power Engineering from
University of Manchester, UK, in 1995. In 2010, he did
an MBA at PUC-Rio (Pontifical Catholic University,
Rio de Janeiro, Brazil) in Energy Business. Mr. Tenório
joined CHESF (Brazil) in 1982, and in 2000 joined ABB
Power Systems in Sweden, moving onto to join ONS
in Brazil in 2004. His area of interest includes HVDC
links, FACTS controllers, Electrical and EMT studies,
and Power Quality.
Alexandre Akio Nohara received his B.Sc. Degree
in Electrical Engineering from Federal Fluminense
University, Brazil in 2011, and his M.Sc. in Electrical
Power Engineering from Federal University of
Rio de Janeiro, Brazil, in 2015. Mr. Nohara joined
ELETROBRAS FURNAS (Brazil) in 2008 and ONS
(Brazil) in 2011. His area of interest includes power
system stability and control, Power electronics, Electrical
and EMT studies.
Antonio Felipe da Cunha de Aquino was born in
Teresópolis, Brazil, on June 13, 1974. He received
his M. Sc. and D. Sc. degrees in electrical power
engineering from Federal University of Rio de Janeiro
(COPPE), in 2000 and 2012 respectively. From 1999
through 2000, he worked as a researcher for CEPEL
(Brazilian Electrical Research Center). He has been with
ONS since 2000 working on power system analysis. His
technical interests are power system stability, HVDC
links and FACTS controllers.
Operating experience with and future
challenges for coordinating power plant
and power system protection
Anton Janssen*, Liander N.V., the Netherlands
Qikai Zhuang, GEIRI Europe GmbH, Germany
Sean McGuinness, EPRI International, Ireland
Giorgio Poggi, EdF, France
Andreas Kubis, TU Dortmund, Germany
Mirko Palazzo, ABB, Switzerland
Konstantin Aprosin, Prosoft-Systems, Russia
Jean-Marc Willième, GE-Alstom, France

Abstract: During unstable power swings, system separation can

In 2014 the three CIGRE Study Committees A3, B5 and
C4 established a joint working group JWG A3/B5/C4.37
to study the system conditions for and the probability
of out-of-phase cases. One of the primary questions for
the group is whether out-of-step is still a risk for power
systems. The JWG had to investigate out-of-phase in
multiple aspects: service experience, its consequences,
the requirements for protection/control and circuit
breakers, etc. This requires the input from experts in the
field of power system studies, protection, generators,
power plants and circuit breakers.
The article addresses the coordination between power
plant and system behaviour under consideration of
transient and dynamic disturbances in power systems.
Much can be learnt from large disturbances including
system separation, which occur from time to time
around the world. For this reason, the article refers to
international regulations, publications and working
groups (e.g. CIGRE, NERC, ENTSO-e, IEEE) on the
coordination of protection and control.
The perspectives of power systems and power plants are
discussed with emphasis on coordination of protection
and control devices, especially with respect to the timing
of protection intervention, critical clearing time (CCT),
fault ride through requirements (FRT), reactive and
short-circuit power contribution. The article also gives
attention to synchronization problems after a system
separation and generator re-synchronization. Further,
overall risk perception is discussed and mitigation
strategies are addressed.
*anton.janssen@alliander.com
be restricted to predefined locations or be allowed to
occur naturally (mostly near the centre of oscillation).
The application of out-of-step protection and the usage
of power swing blocking and/or tripping functions by
several TSOs are reported as well as system integrity
protection systems. Regarding this topic the reliability
of such functions and their impact on the performance
of distance protection is briefly discussed as well as
considerations for developing relay settings, designing
test plans for lab-testing and on-site commissioning
of relays and synchronization apparatus, the overall
responsibility plus verification of load shedding,
curtailment, FRT, etc.
1. Introduction
For the purpose of this article the following definitions
have been used for “out-of-step” and “out-of-phase”
conditions.
An Out-of-step condition is a power swing that will
result in a generator or a group of generators experiencing
pole slipping for which some corrective action must be
taken [5]. The term is synonymous to unstable power
swing [5] and loss-of synchronism [43].
Out-of-Phase condition is an abnormal circuit condition
of loss or lack of synchronism between the parts of an
electrical system on either side of a circuit breaker in
which, at the instant of operation of the circuit breaker,
the phase angle between rotating vectors, representing
the generated voltages on either side, exceeds the normal
value [44].

KEYWORDS
Power system stability, power system control, power system protection, power system dynamics, power plant
protection, transient stability

Cigre Science & Engineering • N°7 February 2017

63
 Figure 1 - Typical power transfer corridor limits ([7], Fig. 6.5), with angular stability and PV-curve (voltage stability)
“Out-of-phase” is almost identical to “out-of-step”.
However “out-of-step” includes a pole slip of one
or more generators, while “out-of-phase” does not
necessarily require an out-of-phase angle larger than
180°. The term “out-of-phase” is typically used by
circuit breaker experts for whom switching under out-
of-phase conditions covers conditions with an out-of-
phase angle of, say, 90° and above. “Out-of-step”, used
by power plant experts, protection experts and system
study experts, covers a particular case of “out-of-phase”.
Power plants have to be protected from severe damage
and should therefore be equipped with fast acting
protection devices in order to detect faults inside the
power plants (generators, transformers, auxiliary
systems and relevant connection buses and cables). From
a power system point of view, it is essential that multiple
power plants do not trip at the same moment and the
number of power plants tripping for a single disturbance
is minimised. Therefore, protection equipment should
discriminate between disturbances within the power
plant that may damage the power plant itself, and
disturbances on the power system that may damage the
system’s integrity, but can be withstood by the power
plant for a limited amount of time. If tripping of power
plants (in the MW range or larger [1]) is unavoidable, it
is also requested that they keep running while isolated
from the main power network and should be capable to
Cigre Science & Engineering • N°7 February 2017
64
quickly reconnect once the power grid conditions have
stabilised.
1.1 System perspective
Major disturbances in power systems are engraved in
the societal mind as their effects are unacceptable to
modern societies. With a failing power system, society
quickly stops functioning. The larger the black-out
area and the longer the period without electricity is,
the larger the societal chaos and economic damage will
be. The economic losses outweigh by far the costs of
the electricity involved and the costs of the necessary
improvements in the local infrastructure. Resilience is
a major topic nowadays on the political agenda. The
scenarios described in this article are failures due to
natural catastrophes, human activities and also due to
the infrastructural system itself. The requirements for
a power system to prevent large disturbances due to
system instability, or to reduce its impact, are addressed.
A blackout may be caused by a lack of power and/
or reactive power in relation to the load or by a lack of
power transfer capacity. In the first case, the local network
will show serious problems of frequency, while voltage
instability might occur in the second case before collapsing.
In the latter case, the electrical power transfer will become
too large which violates angular and voltage stability as
well as thermal limitations of the transmission corridors
(lines, cables, transformers, combinations of these) as
depicted in fig.1. Here out-of-step conditions come into
play and the system will be split due to automatic tripping
of protection relays or special protection schemes. In
[30] three prominent cases of such large disturbances are
outlined, while [6] reports a recent system split. After
system separation power generation and load in each part
have to be matched, usually by shedding power plants in
an overshoot of power generation or by dropping load
in a shortage of power generation. A system that will
not adequately adapt its load and power in due time will
collapse [6][30].
One of the examples in [30] is detailed in [8] and
described as follows. On 4 November 2006 the European
continental power system split into three parts where over
10 million clients were disconnected due to automatic
load shedding. Resynchronization occurred within
38 minutes and the total system was back to normal
operation within 2 hours. A continental black-out had
been avoided. Nevertheless from a societal perspective
the disturbance was perceived as being a major incident.
Eight attempts were made to synchronize two of the three
electrical islands. In each of the seven failed attempts,
after the circuit breakers were closed to reconnect the
islands, large power swings occurred and protection
tripped immediately (first five attempts) or within
seconds and disconnected the islands again. From this
case it can be learned that power swings occur not only
during large disturbances but also during the restoration
process. So severe power swings, large currents, low
voltages, large phase angles between areas should be
anticipated.
For preventing power system instability and/or limiting
the impact of large disturbances, it is crucial that power
plants remain connected and support the grid with active
and reactive power as well as inertial energy, especially
in situations where the overall system is at risk. Power
plants should provide short-circuit power that allows
grid protection to detect faults in the network.
In many developed countries, a transition from
conventional fossil-fired power plants with synchronous
generators to inverter coupled generation facilities,
e.g. wind or photovoltaic, is taking place. This leads
to fundamental changes that need a rethink of current
Cigre Science & Engineering • N°7 February 2017
65
protection coordination concepts. Driven by this trend,
overall installed generation capacity is shifting from
transmission network into distribution networks. This
implies significant challenges for the protection of
distribution networks such as blinding or sympathetic
tripping, as short circuit contributions may not necessarily
originate from locations considered in the conventional
protection coordination. An in-depth analysis of this
challenges and recommendations of best-practice
examples for protecting distribution networks with large
share of renewables can be found in [9]. However, the
prospective behavior of distribution networks in terms of
stability and their impact on the overall power system is
still in focus of recent research and these developments
could influence the protection coordination between
power generators and transmission networks as well.
Furthermore, the short-circuit current contributed
by inverter coupled generation (non-synchronous
generators) is limited to the inverters rated current due
to thermal restrictions of the power electronics (PE).
Due to this, the available short-circuit power (SCP) in
grids with large share of renewables may behave very
different from today’s. The type and grid connection
of certain renewable plant (for example Double Fed
Induction Generators) can also affect the available SPC.
In literature, there are various opinions on how the
available SCP will develop. One says, the available SCP
will decrease due to the fact, that synchronous generators
with high provision capabilities will be disconnected [9],
while practical measurements in the Dutch transmission
network indicate an increase of SCP due to additional
network reinforcements [31]. One of the most detailed
studies concerning this subject is presented in [10] and
[32]. Here, the development of available SCP in the
German transmission and distribution network is jointly
evaluated based on the simulations of the existing
power system hourly in the full year 2011 (aggregated
transmission network of continental Europe equipped
with a detailed German distribution network model,
including conventional generation units and renewables),
in comparison with the simulations of the power system
expected in 2033 (including planned grid reinforcements
and further expansion of renewable energy sources).
The study identifies that the SCP in Germany tends to
increase by about 20% in average, while the bandwidth
of minimum and maximum available SCP will remain.
It is striking that the available SCP increases in some
regions, despite a reduced installed capacity. This
increase cannot be explained solely by a slight increase
in the SCP supplied from abroad, as it originates mainly
from German plants. It has been shown that this increase
is plausible, as a reduction of instantaneous short-circuit
current providing sources in the vicinity of the fault
leads to a flattening of the grid-side voltage gradient
during the fault. Thus, geographically more distant
generators will participate in feeding the fault. This
situation is exacerbated by additional grid enforcements
that shorten the electrical distance between distant
power stations and the fault location. Another significant
finding is that the SCP can be subject to heavy daily
fluctuations depending on the installed inverter based
generation capacity. In order to control this volatility,
one must consider forcing inverter based generators to
stay connected to the grid at any time for the provision of
SCP. As voltage source converters are mainly applied as
inverters for renewables, the technical feasibility should
be possible.
Besides the studies concerning the magnitude of the
available SCP, the duration of the short-circuit current
contribution also needs to be taken into consideration in
order to verify the correct operation of the grid protection.
This is particularly important for the operation of all
back-up protection of the system whose typical tripping
time is in the range between several hundreds of ms and
up to a few s.
1.2 Power plant perspective
Power plants have to be protected from severe stresses
to prevent damages, not only to avoid economic loss
of their operators, but also to be available for system
restoration. Special attention has to be given to safeguard
a proper functioning of the auxiliary plant (e.g. the feed-
water-pump, the lube-oil pumps, the excitation system)
during the transient grid voltage phenomenon. When a
fault occurs within the power plant and/or its connection
to the grid, the power plant has to be tripped as fast as
possible. For example, the protection of generator A in
figure 2 has to clear fault 1 as fast as possible in order
to isolate the fault and avoid damage, as well as the
protection of generator B has to clear fault 2. Moreover,
by the involved protection all faults in figure 2 have to
be cleared faster than the critical clearing time (CCT)
Cigre Science & Engineering • N°7 February 2017
66
of generator A and/or generator B and/or other nearby
power plants, in order to avoid loss of synchronism. The
protection systems of the power plants have therefore
to be engineered to achieve a fast tripping time in an
incident that strikes the plant itself, but a delayed
tripping time for faults in the grid in order to be selective
with the grid protection; in figure 2 for the protection of
generator A: a delay for the faults 2, 3 and 4. As a power
system should be designed to continue stable operation
even if the single largest infeed is suddenly lost, the
power system may face serious problems when multiple
power plants are tripped due to a single event.
Figure 2 - Fault locations nearby generator A
The critical clearing time (CCT) is defined as the
maximum duration of a fault until one or more generator(s)
or a part of a power system loses synchronism with the
rest of the grid. The CCT is typically much shorter for
three-phase faults than for two-phase or single phase
faults. It depends on the ratio of the rotational inertia
and the nominal power of the power plant, the actual
active power, the momentary excitation and the involved
reactances. For numerous fault locations and operational
grid conditions the CCT of a particular power plant
can be calculated, but usually one refers to the shortest
calculated value of all reasonable conditions.
The grid protection systems are engineered to clear faults
(short-circuits) quickly, reliably and selectively, so that
the fault duration is restricted. Differential protection
(including bus protection) and the first zone of distance
protection can usually act fast enough. Depending on
its design back-up protection may act within the same
time frame (e.g. redundant protection relays, under-
reach and overreach distance protection schemes
with telecommunication channels), or after a delay
time (second zone of distance relays, breaker failure
 Figure 3 - Typical timing of fault clearing and back-up fault clearing

protection). In the latter case the total clearing time This requirement can be expressed as a minimum value
may reach several hundreds of ms (fig.3) and selective of the CCT or as an under voltage ride through (Fault
clearing of the faults is not assured. Ride Through: FRT) curve. FRT requirements are set
forward either in (inter)national grid-codes or by TSOs,
Figure 3 is a general example of the main and back-up
or are negotiated between grid and plant operators.
protection fault clearing time. Often, the back-up clearing
Figure 4 extracts FRT-requirements from the recently
time of the grid protection is longer than in figure 3 in
approved European grid-code “Requirements for
order to avoid the risk of non-selective tripping. Its setting
Generators” [1]. TSOs can determine the precise values
depends on the characteristics of the network (including
between the minimum and maximum limits given by the
the automatic re-closure philosophy and timing) and the
code. Note that the requirements for non-synchronous
protection system (delay in telecommunication signals,
generators connected with inverters are higher than those
relay reset time, etc.). In addition, slower protection
for synchronous generators, as they can easily fulfil the
functions (third and fourth zone of distance, (directional-)
longer duration of voltage recovery.
overcurrent, etc.) are installed as back-up protection
functions for clearing the fault if the main protection fails
or clearing far reach faults.
Power plant protection should not trip on a grid fault
before the grid main and back-up protection clearing
time. Some TSOs further request that the generator
protection trips are even more delayed (for example in
order to be selective with the third zone of the distance
protection). A medium to large synchronous generator
generally contributes short circuit current in the range
of 2 to 4 p.u. The IEC [45] and IEEE Standards [46]
state that generators should withstand these fault
current levels for 1 s or more. A similar consideration is
applicable for the generator withstand capability to the
negative sequence fault current components caused by Figure 4 - FRT limits to the European Requirements for Generators
non-symmetrical faults in the grid. larger than 50 MW Minimum and maximum requirements are given for
synchronous generators and non-synchronous (PE)
To prevent power plants from tripping due to angular
Apart from such a coordination, some power plant
instability the grid protection must operate faster
protection should unambiguously detect faults within
than the power plant’s critical clearing time [1]. For
the plants’ territory, in particular differential protection
strategic parts of the network, back-up protection
(generator 87G*, step-up transformer 87T*, step-up
should act faster than the CCT in order to avoid angular
transformer earth restraint 87N*), stator earth fault
instability and disconnection of the power plant. When protection (59G* or 51G*), loss of field protection
the CCT is around or below 200 to 300 ms, it might be (40 G*), etc. Although protection cannot prevent the
challenging to respect this principle [33]. In such cases, dynamic stresses of the first peak of a sub-transient
applying a redundant instantaneous protection system short-circuit current, the cumulative damage caused
can be considered such as duplicate main protections by succeeding current peaks can be limited. When a
(differential protection or distance protection with tele- circuit breaker is closed under out-of-phase conditions
protection and with independent communication links). - for instance at false synchronization - the out-of-
Based on fault clearing times and the recovery of the phase currents may exceed the peak value of the sub-
voltage after fault clearing, TSOs request generators to ________________
withstand a voltage dip curve without losing stability. *ANSI-codes for protection functions; see figure 5 and [11]

Cigre Science & Engineering • N°7 February 2017

67
 Figure 5 - Example of power plant protection with ANSI-codes [11]

transient short-circuit current [34]. According to the
international standards for large generators [45][46], in
case of a generator terminal short-circuit or an out-of-
phase synchronizing occurrence, the generator stator
windings, rotor windings, generator and turbine shaft,
shaft couplings, shaft balance and foundation should
be checked carefully, in order to prevent consequential
damage in service.
At power plants an out-of-phase condition may occur
in case of mistakes in the wiring or the setting of
synchronization apparatus. Error in the connected
phases, including wrong rotation direction, leads to
false synchronization with a large difference in phase
angle between generator and network. This can happen
at the first synchronization of a new generator or after
important modifications of the power plant and/or the

Cigre Science & Engineering • N°7 February 2017

68
synchronizing circuits [20]. Via careful checking and
testing, such wiring errors can be detected.
In case of false synchronization, the out-of-step stresses
may become very large. It is not obvious which protection
will lead to the immediate interruption of the out-of-phase
currents or any protection will trip at all. Most power
plant protection relays are set with a certain time delay
in order not to trip in case of a grid fault (unbalanced
load, maximum current, distance protection, etc.) or not
to disturb the regular synchronization process (reverse
power, etc.), a typical sketch of which is represented
in figure 6 (power control of a power plant right after
synchronization). To address this issue some inadvertent
energization protection (where existing) could be used
and properly set in accordance with relay manufacturer’s
manual, in order to give an un-delayed trip command [11].
Figure 6 - Typical representation of a possible generator active power
exchange, right after synchronization
The inadvertent energization protection function is
primarily used to identify when a generator at standstill
or operating below nominal synchronous speed is
unintentionally connected to the transmission grid. This
has been experienced in the past during commissioning
when the generator circuit breaker close command contact
is accidentally bridged or activated. The protection is
armed (enabled) some seconds after the generator is
disconnected from the grid. Further, relays allow using
an under-voltage threshold as arming condition. Once
armed, an under-voltage protection logically combined
with an overcurrent function is enabled. Since the
generator is disconnected from the grid, current can only
flow if the generator circuit breaker is closed. When the
generator is scheduled to return to service the necessary
synchronising conditions must be met before closing the
generator circuit breaker. If the necessary conditions are
not met and large deviating frequency, voltage or phase
angle conditions exist when the generator is connected,
Cigre Science & Engineering • N°7 February 2017
69
large currents will flow and the inadvertent energization
protection will trip. The inadvertent protection function
is disarmed (disabled) some hundreds of milliseconds
after the generator is resynchronised to the grid in the
correct manner, in order to ensure that it does not trip
during grid fault conditions when low voltage and high
currents are to be expected.
1.3 Risk perception
The chance of losing synchronism within a synchronous
power system is greater when the system is loosely
meshed or power flows are high [12]. Utilities with long,
heavily loaded overhead-lines have put forward adapted
requirements to protection and control equipment,
including circuit breakers, where they identify credible
risks of out-of-step conditions occurring. For them, the
probability and the consequences of uncontrolled out-of-
phase situations are unacceptable.
In contrast, utilities with heavily meshed systems may
not have much experience with out-of-step situations
as severe grid disturbances with many simultaneous
outages are required for them to occur. Nevertheless, the
consequences are huge, as can be learnt from a number
of large disturbances in meshed grids in recent decades.
Moreover, due to the cascading effects of line trips at a
major disturbance, meshed grids metamorphose locally
into a more radial system. Managing such strategic risks
on meshed grids (i.e. risks with a very low probability but
with unacceptable effects) is rather difficult, since the usual
statistical approach for incidents cannot be applied. The
direct trigger of out-of-step outages is often not the reason
that the disturbance reached its scale and conventional
load-flow simulation tools are challenging to be applied
to the study of cascading outages and associated degraded
voltage profiles. The hidden factors which contributed
to the escalation of failures spread over a wide range of
technical areas, each of which requires a deep technical
knowledge. Generally speaking, the overview on the
complexity is missing and the risks associated with out-
of-step can be underestimated or not perceived.
The primary goal of the strategic level of Asset-
Management is to find out a feasible roadmap which
satisfies all stakeholders and provides a stable business
environment for the optimizations at tactical level. In
other words, the business factors threatening this primary
goal will initiate the “strategic risks”.
A risk becomes strategic, when its characteristics prevent other parties, since it normally means extra costs.
analysts from quantifying it. In risk management in • Multidisciplinary: Knowledge from various
general, specialists have identified these characteristics technical areas is necessary to understand how a
as uncertainty, complexity and ambiguity [35]. The local failure can escalate to out-of-step or a system-
meaning of the characteristics has been applied meanings wide disturbance. These areas include system
to Asset-Management as below [4]. stability, (power swing) protection and control of
synchronous and non-synchronous generators,
• Uncertainty: the (un)availability of knowledge for
PE converters of distributed energy resources and
the asset manager to identify, measure and quantify
circuit breakers.
the elements in the causal chain of the studied risk.
• Missing records: The low probability of out-of-
In Asset-Management, high uncertainty is frequently
step events prevents many utility companies from
caused by missing (the knowledge from) the experts
experiencing them. But even within those which
from a certain area. Such area can vary from
have experienced out-of-step, the precise record
technology, economics, to social science.
of how out-of-step has developed is frequently
• Complexity: the degree to which multiple causal
unavailable due to two reasons. Firstly, when out-
chains are interdependent. The straightforward
of-step happens, the utility company will be so busy
causal chains analyzed at the tactical level are based
with the recovery that most direct evidence of how
on a few conditions on, for example, the load profiles
out-of-step occurs will be put aside and eventually
on assets, the supply chain, the regulatory rules, etc.
lost. Secondly, being afraid of the responsibility
Once any of these conditions are broken, a strategic
of causing the out-of-step disturbance, a utility
risk of high complexity is likely to be formed.
company tends to conceal the fact of how out-of-
• Ambiguity: the variety of the interpretation of the
step developed.
same facts, as well as the consequent normative
• Delay effect: The direct trigger of an out-of-step
conflicts between different stakeholders. In Asset-
condition is often not the reason that the disturbance
Management, even when a technical stimulus is
reaches its scale. A decision can be made many
known to all internal and external parties, they
years before it is finally proven to be contributing to
can still disagree on the countermeasures, due to
a system-wide disturbance, while a specialist who
different personal experience or understanding on the
had a long-term foresight could not influence/reach
consequences.
the decision maker.
Asset managers should check the risk of power system
In other words, out-of-step situations in meshed systems
instability properly. However, when modelled as a risk,
have to be anticipated and controlled in a way that
power system instability has properties which make it a
prevents a cascading effect and reduces its consequences
“strategic risk”. Such properties include:
as much as possible. Consequently, strong and uniform
• Extreme: power system instability is an extreme requirements have to be formulated in the grid-codes
risk, i.e. of catastrophic consequence but very for protection and control equipment, for components
low probability. Many asset managers will not in power plants and transmission systems and for the
experience out-of-step in their whole life. Hence, involved circuit breakers.
some of them never have recourse to out-of-step.
That large disturbances with system separation occur
• Multiple parties: an out-of-step condition can occur
less exceptional than expected can be seen in the
between the grid and the generator, or between
following list with examples of the last decades (see also
interconnected grids. When tackling the out-of-step,
the CIGRE Workshops on Large Disturbances and [13].
the operator of each grid or generator has its own
interest. A TSO wishes to retain the stability of its • South Australia, September 28, 2016
grid, while a generator operator wishes to minimize • Turkey, March 31, 2015
the damage on its generators. An economically • The Netherlands, March 27, 2015
rational party prefers NOT to reduce the risks of • India North-East-West Region, July 30, 2012

Cigre Science & Engineering • N°7 February 2017

70
 • India North-East-West Region, July 31, 2012
• Argentina, April 4, 2012
• USA SouthWest, September 8, 2011
• Brazil, February 4, 2011
• Canada Toronto, March 10, 2010
• Brazil/Paraguay, November 10, 2009
• Republic of South Africa, February 26, 2008
• China Chenzhou, January, 2008
• Continental Europe, November 4, 2006
• Japan Tokyo, August, 2006
• New Zealand, June 12, 2006
• Republic of South Africa, February, 2006
• Republic of South Africa, January, 2006
• North-Western Germany, November 25, 2005
• Indonesia, August 18, 2005
• Greece, July 12, 2004
• Italy, September 28, 2003
• Sweden-Denmark, September 23, 2003
• USA Eastern Interconnection, August 14, 2003
• Brazil, January 21, 2002
• Brazil, March 11, 1999
• USA Western Interconnection, August 10, 1996
• USA Western Interconnection, July 2, 1996
• South China, May 25, 1994.
This far from complete overview shows that large
disturbances occur worldwide on a regular basis. There
is no exception for special regions or network topologies
and the consequences are large. Of course, the probability
of a large disturbance in a heavily meshed network is
relatively small and the probability is not equal for all
regions. Furthermore, the developments in the electric
power systems are in the direction of more volatile
power generation, more non-synchronous generators
(i.e. connected by a power electronic inverter), power
plants with less synchronous moment of inertia, large
power flows along larger distances, less corridors to
transfer the power and operating the system more to
the limits of stability. All these developments lead to an
increase of the probability of system instability.
2. System instability
System instability can be triggered by long fault clearance
times, sudden changes in the power balance or a change
in the system impedance [14]. System instabilities can be
divided into voltage instability when a deficit of reactive
power plays a role, frequency instability when the balance
Cigre Science & Engineering • N°7 February 2017
71
between power generation and load cannot be achieved,
and (rotor) angle instability when the power cannot be
transferred between centres of power generation and load.
A further division can be made into short term phenomena
(within seconds) and long term phenomena (within tens
of seconds to minutes). Instability can develop in a few
hundreds of ms in case of severe three-phase faults or
within seconds after a large disturbance or within tens of
seconds after a small disturbance (leading to inter area
oscillations). [15]
Along a reactance X, like an overhead-line, a transformer
or a generator, power is transmitted by the angle difference
θ between the voltage vectors at both sides of the reactance.
The power that can be transmitted as function of the angle
θ between the two voltages, is presented in figure 7 with
the left hand diagram giving the voltage/current vectors
seen at the receiving end (Er), while the right hand diagram
shows the voltage/current vectors related to the sending
end (Es). The latter is usually applied to generators. From
these diagrams, the relationship between transferred
power and the voltages can be calculated.
Figure 7 - Power transfer along a reactance X, being a generator, OH-line,
transformer or combination of these
From the figure it is clear that the relationship between
the transmitted power P and the receiving voltage, here
presented as Er, is complicated, even with a fixed power
factor and a fixed sending end voltage Es; see equation
(1). Despite the simple model, the formula that leads to
the PV-curve in figure 1 is rather complex (note that Q is
defined by the power factor, tg φ):
(1)
From figure 1, angular or angle stability, it can also be
learned that the maximum (rotor) angle θ to transmit power
is 90°. Beyond this angle the power to be transmitted
will decrease and a condition in which the power system
becomes unstable is met. A small overshoot power input,
leading to a small acceleration and increase of angle θ, swing oscillation. In this way, the number of pole slips can
leads to a decrease of power that can be transferred to the
other end. Consequentially, a further acceleration will take
place, θ increases further and so on. In figure 1 a situation moderate). The rotor angle can also be directly measured
is presented in which a fault between a power generation
site and a load site leads to a sudden decrease of the
voltage, while the power input to the system stays constant
(Pmechanical). The increase in (rotor)angle θ is presented by
the angle «before» up to «after» the short-circuit. After fault
clearance the voltage returns to a value between the voltage
before and the voltage during the fault. The power that is
transferred, (Pelectrical after short-circuit) is larger than the power
input (Pmechanical) and a deceleration takes place, so that the
angle returns to a new stable operating point. Beyond a
certain combination of voltage drop during the fault and
duration to clear the fault, it will not longer be possible to
prevent a pole slip.
After a pole slip, the generator, the group of generators or
the part of the power system may run again synchronously
with the remaining part of the power system (this condition
is not so common) or it may be separated from the remaining
part. System separation will result either in disconnection
of power plants (tripping to house load) or in an island, as
described in section 1.1.
System instabilities are very complicated. No disturbance
falls exactly into one of the categories for instability. The
complexity of the network behavior, the overwhelming
information generated by the control systems and the
involved legal aspects prohibit easily-accessible knowledge
sharing and statistical analyses.
To cope with power swings, large power plants are generally
equipped with Power System Stabilizers and out-of-step
protection (ANSI 78*). The latter may be a measurement
of the impedance seen from the generator terminals, with
a setting to trip when the measured impedance between
generator terminals and the electrical center of oscillation
is towards the generator or towards the step-up transformer
in the range of 70% to 100% of the transformer impedance.
There are also generator out-of-step protections that are
set to trip with two settings when the electrical center is
towards the step-up transformer. In this case the faster
threshold can trip when less than 100% of the transformer
Cigre Science & Engineering • N°7 February 2017
72
impedance is measured and the slower can trip when less
than 150% to 200% is measured. The impedance follows
a locus that circulates with the frequency of the power
be counted and used as a criterion to trip (when the center
of oscillation is further away and the out-of-phase currents
on the shaft of the generator.
The requirements for the number of pole slips that a power
plant has to sustain before tripping vary between grids. In
some regions no restriction for tripping applies (tripping is
permitted right after the first pole slip), while in other regions
generators must remain connected for up to 4 pole slips.
Where the grid operator requires a power plant to remain
connected for multiple pole slips, it requires the verification
and coordination of Multiple generator protection devices.
(single period) These include not only the pole-slip
protection itself, but also distance protection, overcurrent
protection, voltage protection, reverse power protection
and loss-of-excitation protection. Manufacturers prefer fast
tripping after the first pole slip in order to prevent immediate
and consequential damage. In severe cases, even auxiliary
equipment such as electric pumps (feed-water pump, lube
oil pumps, etc.) can be affected and cause a power plant
trip. This can be caused by too sensitive relay setting of the
motor feeder or by a strong variation of the process quantities
(pressure, level, flow) induced by the auxiliary voltage drops
and consequent variations of the motor torque.
Independent power producers (IPP) may lack adequate
information about the network or share little responsibility, so
that they prefer standard relay settings by the manufacturer.
Sometimes no deep verification of functioning and setting
of the relays is done by a certain TSO or regulator. It may
even be a conflicting responsibility for the IPP when a
manufacturer warranty requires tripping after the pole slip
to avoid damaging the power plant. This is a very delicate
topic, since responsibilities will cross the limits of property.
3. Unstable power swings
- protection concepts and
recommendations
CIGRE JWG A3/B5/C4.37 conducted a basic enquiry
among the TSOs represented in the JWG or associated to
the members. From the answers received, it is clear that
these utilities seldom apply a centralized system wide
protection system considering power swings. About half
of the TSOs have selected special locations where the
power system will be separated in case of detection of
an unstable power swing. Such unstable power swings
are mostly detected by dedicated out-of-step relays
which consequently split the system. These TSOs have
activated the power swing blocking function in their
distance protection relays. Meanwhile, a number of
utilities without predefined locations to split the power
system have also activated the blocking function. The
blocking is most of the time limited to zone 1 and 2, but
sometimes it is oppositely limited to the zones 2, 3 and
4. Some utilities even block all zones without applying
a dedicated out-of-step relay or the power swing trip
function in some of their distance relays. Details of their
policy are not known, but it could be that power swing
blocking is disabled after a short while or when a fault is
detected. Or the distance protection has merely a back-
up function.
Almost all TSOs reported that they do not put forward
requirements to power plants about the number of
pole slips that a generator has to be able to withstand.
Nevertheless some require that the power plants out-of-
step relays are set with a delay of 2, 3 or even 4 pole
slips. Table I gives a summarizing overview of the
received answers
Since the loss of synchronism is accompanied by low
voltages and high currents, the grid distance protections
could recognize them as three-phase short-circuits.
Without activating any of the power swing blocking
and power swing tripping functions, distance protection
relays are likely to trip close to the electrical center [1]
[30]. As discussed in [30], such a policy is likely to
provoke that the electrical center will be detected within
the un-delayed zone 1. The disadvantage of such a
policy is the low predictability of the system cut-set that
goes with numerous and cascading line trips.
Some experts prefer distance protection to be applied
only for high performance fault clearing without
enabling any additional feature, which may influence the
clearance times for short-circuits [36]. The power swing
detection algorithms implemented in some protection
relays can result in a little delay to tripping of fast zones
in order to take an extra cycle of measurements to be
able to distinguish between power swings and genuine
faults. To protect a system against loss of stability,
it is recommended to implement dedicated system
protection schemes [16]. Many papers are published
on the detection of out-of-step conditions and system
protection schemes; for instance: [21]-[28], [37][38].
An example of a dedicated protection scheme is
described in Report [38], based on the prevailing system

Table I - Summarized overview of enquiry results among TSOs worldwide (1 to Z) on their power swing protection policy

TSO 1 2 3 4 5 6 7 8 9 A B C D E F G H I J K L M N O P Q R S T U V W X Z

predefined
N N N N N N N N N N N N N N N N Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
locations1
out-o-step
N N N N N N N N N N N N N N N N Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
function2
special
N Y Y Y Y Y Y Y Y Y Y Y Y Y Y
function3
blocking
N N N N N Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
distance4
which
2-4 1,2 2-4 all 2-4 all all all 1 all 1 1,2 1,2 1,2 1,2 1,2 all all 1,2 1 1 1 1 1 1 1-3 1,2 1,2
zones
1 Whether TSO applies pre-defined locations for system separation in case of unstable power swings
2 Whether TSO applies out-of-step relays ANSI 87 or out-of-step tripping function in distance relays ANSI 21
3 Whether TSO applies special protection systems, such as voltage swing relays, inter-trips, voltage phase angle comparison, line loading.
The application of centralized wide area protection systems has not been reported.
4 Whether distance protection relays are blocked for unstable power swings and which zones are blocked

Cigre Science & Engineering • N°7 February 2017

73
 Figure 8 - Power system test model for out-of-step relay (algorithm) certification tests; figure B.1. of [39]
conditions of TEPCO, Japan, where a tight power
generation capacity margin has arisen since nuclear
power plants were taken out-of-service. By using
diurnal energy storage (filling the water basins during
the night and releasing the water during the day), energy
demands can be met, but the loss of a transmission line
to the hydro plants would lead to stability problems.
Therefore, an automatic hydro plant shedding scheme
is being implemented to keep the system stable. This
is a new special protection scheme, which will operate
autonomously at each hydro plant without the need of
analyzing remote information. See for other applications
with respect to the state of the art [12][17].
According to the Russian grid-code [2] the control
centre intervenes manually by load shedding when long
interconnecting overhead-lines are loaded above 80%
of their secure power transfer stability limit, as defined
by the TSO. Apart from that, system wide contingency
analyses based on a state estimator are used to detect
possible stability problems and simulate the effect of load
shedding. Then, load shedding follows automatically by
the same technology as applied for under-frequency load
shedding. If automatic load shedding is not effective and
the power system falls into out-of-step conditions, the
network is split by special out-of-step tripping relays at
pre-determined locations.
The described system protection schemes are examples
of schemes used to prevent system splitting, but there are
also protection systems that act in order to keep parts of the
system stable after system separation. An example of this
is where tie lines between regions are tripped to preserve
the stable operation of the grid within each region. In
such cases, the regions do not support each other any
longer. This is a matter of policy: the grids support each
other so long as practical and when appropriate separate
the healthy parts to safeguard them. A similar choice in
policy could be made between autonomous synchronous
Cigre Science & Engineering • N°7 February 2017
74
areas that are coupled by HVDC links. These areas may
support each other through the HVDC connection with
a risk that the neighboring area will also lose stability or
HVDC connections can be blocked, thus restricting the
incident to one synchronous area only.
Not much has been published on the reliability of the
existing and the new (proposed) protection schemes or
on the reliability of distance protection with blinders
or other software tools to detect very slow or very fast
power swings. In Ireland and Denmark power system
oscillations with a frequency of only 0.4 Hz occurred. In
the UK the Grid-code requests that the generator power
system stabilizers are tested in order to prove damping in
the range from 0.3 to 2.0 Hz. Note that fast oscillations
between power plants could be in the range of 4 to 7 Hz
(and even more: in Russia 10 Hz and in [36] 11 Hz).
Close collaboration between the engineers who perform
the stability studies and the protection engineers will
contribute towards the development of good settings.
Power swings are complex events. While evaluation
of relay settings and configuration in simulations will
contribute to the development of secure and reliable
settings, additional understanding and confidence in
relay algorithm and setting performance can be gained
from lab testing or commissioning testing of the
relay using simulated measurements. These may be
generated from offline studies using conventional RMS
stability assessment programs (PSSE, PSLF, TSAT,
PowerFactory etc) or from on-line studies using real-
time digital simulators (RTDS).
An extensive type test of the algorithm to detect power
swings and to block or to give a (delayed) trip is
performed in Russia [39]. A network is used, as given
in figure 8 and the relays locate at the buses 2 and 5. A
number of cases with different load, faults and power
swings are simulated with an RTDS simulator and the
behavior of the protection algorithm can be extensively
 Figure 9 - Anticipated disconnection of DER power at frequency deviations up to ± 0.5 Hz (2014)

tested. But it is not possible to simulate the cascading a large retrofit program has been scheduled and it has
effects that occur in practice. been planned to be implemented before 2016, at least in
Germany and in Italy.
4. Dispersed energy resources In smaller synchronous areas, like on the Island of Ireland,
(DER) affecting power system the active power frequency response of renewable
stability power sources is very critical. This resulted in specific
requirements in the grid-code [3], illustrated in figure 10.
Conventional synchronous power plants control the system
frequency in such a way that deviations from the nominal
frequency are minimized through being immediately
compensated by (1) the inertial energy stored in the
rotating masses, (2) the primary governor response (droop
control), and (3) secondary and tertiary control regimes.
Nowadays frequency deviations occur more often, with
a longer duration and with a larger magnitude than in the
past. A main cause is varying energy exchanges forced by
the market prices that are adapted for each hour. There are
some disputes about the influence of increasing percentage
of DER. The renewable sources are volatile. Thus, by
their large numbers, could enlarge power frequency
deviations by certain changes in weather conditions,
unless suitable reserves are available. Nevertheless, more
relevant is their response to frequency deviations – before
relevant grid-code changes were (are) enacted they were
(are) permitted to trip for modest deviations of frequency.
Figure 9 extracted from an ENTSO-e report [18] shows Figure 10 - Irish gridcode requirements for active power frequency response
of windfarms [3]
that the large amounts of DER tripped before 49.5 Hz
resp. 50.5 Hz would be reached, while the European grid- At a major disturbance, it could happen that some power
code [1] requires operation at least between 49 and 51 Hz plants do not act as expected, and/or load shedding do not
for all power generating units. Meanwhile, the quantity of follow the rules that predetermined percentage of load
inverter-coupled generation found to have disconnected will be dropped immediately at a certain under-frequency.
in figure 9 exceeds 3000 MW, the level of a sudden power It is evident that the escalation of an incident could be
deficit up to which the European continental system has avoided by applying and verifying the strict rules which
been designed to control the frequency. have been agreed upon by the collaborating parties.
During the system split of the European network on 4 Another challenge is that certain types of renewable power
November 2006, a large number of wind power parks generation lack synchronous inertia and synchronous
not only disconnected automatically at a certain power torque [40]. In case that conventional generation is
frequency deviation, but also reconnected automatically replaced, an increasing share of inverter-interfaced DER
at a certain smaller frequency deviation. Figure 9 shows leads to a decrease of system moment of inertia necessary
the amount of prospective disconnected DER capacity in to achieve system stability. However, the situation is more
case of frequency deviations [18]. Due to this assessment, complex than at first glance, as conventional plants are

Cigre Science & Engineering • N°7 February 2017

75
still running at low loads, thus effectively contributing a angle between the involved voltage sources determines
higher moment of inertia per MW output than in the past. the out-of-phase current and therefore the TRV stresses.
Moreover, some DER make use of directly coupled (a)
At power plants the main circuit breaker may locate
synchronous generators and other DER emulate by their
at the HV-side of the step-up transformer or between
controllers a virtual synchronous torque. Nevertheless,
generator and step-up transformer, as shown in figure
numerous publications point out the risk of lack of
11. When a MV generator circuit breaker is applied,
synchronous moment of inertia in future system conditions.
another circuit breaker will be necessary at the HV side
Report [40] gives an example of how to manage this risk
by advanced unit commitment in the Irish synchronous of the step-up transformer, but only for regular switching
area network. In this case, a paid system service includes (e.g. maintenance) and transformer protection plus back-
both synchronous moment of inertia and fast frequency up protection. The HV circuit breaker is also used for
response from non-synchronous power plants. resynchronising the power plant to the grid after house-
load operation.
5. Circuit breakers
A command to open the circuit breaker results in a
physical separation of the contacts within about 40 ms
for all three poles. But, the load or fault current will
continue to flow, as the electromagnetic force of the
power system will not allow a current to be interrupted
without approaching zero. Therefore, modern AC
breakers will wait until the next natural power frequency
current zero to really stop the current from flowing.
Meanwhile, the current will flow through the electric arc
that is established between the departing contacts. As the
circuit breaker may not be able to interrupt the current
at the first power frequency current zero after contact Figure 11 - Location of main generator circuit breaker
separation, the circuit breaker pole has to wait for the
next opportunity. The duration after contact separation Different standards are applicable for HV (generator)
up to the final interruption is called the arcing time (or circuit breakers and MV: [44][47] resp. [48]. These
arc duration). Modern HV circuit breakers face an arcing standards cover the fact that during a number of power
time of 10 to 20 ms. Thus, the time from command to frequency cycles fault currents and out-of-phase currents
open until current interruption in all three phases is may show no zero crossing. The lack of current zero’s
about 60 ms. makes it very hard for a circuit breaker to interrupt [20];
figure 12. More detailed information on MV generator
The current to be interrupted (a load current, a fault circuit breakers is given in [34][41].
current, an out-of-phase current) is determined by the
driving voltage(s) in the loop formed by circuit breaker The out-of-phase switching requirements in the
and the relevant impedances up to the driving voltage Standards for HV circuit breakers seem to be taken
source(s) [19]. It means that the current is always close from a proposal in an AIEE paper in 1952 [29], in which
to inductive with respect to the driving voltage. For a calculations, simulations and real tests in a power system
defined system, the response of the system to the change have been analyzed. The rated out-of-phase current
in dI/dt at I(t)=0 is only dependent on and proportional has been proposed to be 25% of the rated short-circuit
to dI/dt = 2πf*√(2)*Irms. The response is the transient current. For economic and statistical reasons, minimum
recovery voltage (TRV) across the circuit breaker peak values from the TRV analyses have been proposed:
contacts. For circuit breaker design both the current a power frequency recovery voltage of 2.0 p.u. and an
to be interrupted and the TRV are the most important overshoot of 25%. The factors add up to a TRV peak
parameters. In case of an out-of-phase condition, the (Uc) of 2.5 p.u., while the analysis showed peak values

Cigre Science & Engineering • N°7 February 2017

76
varying between 2.5 and 4.0 p.u. for effectively earthed
networks. The TRV peak value of 2.5 p.u. has been
introduced in the Standards [44][47] and corresponds to
out-of-phase angles as large as 105° and 115°. But, when
such information is available, large disturbances show
out-of-phase angles that may become much larger than
105° to 115°. For instance, in the Report of the Enquiry
Committee on the Grid Disturbances in India, June 30th
and 31st, 2012, from Figure S-1 (page 17) an out-of-
phase angle of 135° can be deduced.
At the same time the large disturbances show voltage
instability problems. At the moment of system separation
large voltage dips, depressed voltages and dramatic
voltage drop have been reported together with huge
power flows. For instance with the separation of the
Italian grid from the European continental grid (2003),
for the depressed voltage a figure less than 80% has been
published.
Large out-of-phase angles belong to system conditions
in which low operating voltages occur at the same
time, as can be learnt from the large disturbances. The
combination of a large out-of-phase angle and a low
operating voltage give TRV peak values comparable
with those mentioned in the Standards for situations with
a relatively low out-of-phase angle at the rated voltage
(i.e. the maximum operating voltage).
Figure 12 - Prospective fault current considering the moment of inertia of
the synchronous machine and resulting from synchronizing under out-of-
phase conditions (out-of-phase angle φ0 = 90°)
6. Conclusions and observations
• Since long time distance protections are provided with This seems to be more a matter of discipline than of
power swing blocking and tripping options. In the case
Cigre Science & Engineering • N°7 February 2017
77
of fast evolving faults and very slow or very fast power
swings, some questions could be raised about the impact
of such features on the reliability of the protection
functions, on the reaction time of the relays and on the
proper functioning of the blinders (under all prevailing
conditions).
• About half of the utilities in contact with CIGRE JWG
A3/B5/C4.37 have arranged predefined locations for
system separation in case of unstable power swings.
These utilities have activated power swing tripping at the
desired locations and power swing blocking elsewhere.
(Some of the TSOs who do not have predefined
locations for system separation activate power swing
blocking as well). Most of them apply special out-of-
step relays to achieve the separation. International
bodies, manufacturers and utilities are dealing with the
setting of the relevant relay parameters, even recently
(e.g. IEEE Power System Relaying Committee Working
Group J5, NERC Reliability Standard PRC-026-1 and
PRC-023-3, [42]). Continuous evaluation of the relay
setting policy, especially after system disturbances, is of
prime importance.
• Type testing of distance relays and out-of-step relays
for their behavior under stable and unstable power
swing conditions is a difficult task. At low power swing
frequencies relays may not act properly and, when
required, the range of frequencies to be tested is quite
different per TSO. An example from Russia of type
testing relays in a network model is given in figure 8.
However, here as well as elsewhere, it is not possible to
simulate the cascading effects that occur in practice.
• Apart from type testing (that should not be limited to
testing the algorithms), routine testing and regular
verification of the relay settings is an important activity.
The involvement and split roles and responsibilities of
TSOs, power plant operators and regulators in these
activities are not always well established. Because
of the split responsibilities and the limited available
knowhow, this is of large concern.
• Verification of the correct wiring and proper functioning
of synchronization and synchro-check apparatus is
required to prevent power plant (and system) false
synchronization and the consequential damage.
complexity. Taking false synchronization into account
 in the design and specification of primary plant and of
protection systems is still deemed necessary.
• Load shedding and generation curtailment at deviating representatives of the more than 30 utilities that have
power frequency is a system wide necessity and
all involved parties have to fulfil their obligations.
Regular checking and testing is required and should be
obligatorily reported.
• There are a number of requirements for power plants (USA), Shinichi Suganuma (Japan), Franco de Villiers
related to their impact on the power system: FRT,
number of pole slips to be withstood, island operation,
operation at deviating power frequency, (reactive)
power contribution during transient phenomena,
coordination of protection, stability of the auxiliary
plant, black-start capability, etc. Network operators have
to fulfil obligations with respect to fast fault clearing
time, controlling power swings, controlling deviating
power frequency (e.g. load shedding) and voltage (e.g.
transformer tap changer control). An authority which
has the overall view, coordination of responsibilities and
is pro-actively verifying all commitments is missing.
• Unlike the USA, which has National Electric Reliability [4] Qikai Zhuang, “Managing Risks in Electrical Infrastructure
Standards per the North-American Electric Reliability
Corporation (NERC), Europe does not have over-
arching rules on the required response of generator and
grid protection to power swings. Instead the details are
left open to the TSOs in their respective grid-codes,
even where TSOs are connected within the same
synchronous area. It is technically beneficial to have a
unique, unambiguous set of rules within a synchronous
area. Such a ruling should not depend so much on
each TSO, but on uniform system wide and local
technical aspects, in order to achieve fairly distributed
responsibilities. Such rules can even cross synchronous
areas, depending on the way HVDC couplings are
operated under transient phenomena.
• In the international (generator) circuit breaker
standards, requirements for type testing under out-of-
phase conditions have been defined. Other prevailing
conditions, such as power frequency voltage deviations
from rated values and actual delay times in protection
and auto-reclosing settings, have to be considered in
project specific applications. For instance, whether and
which protection takes care for immediate tripping after
false synchronization or the relationship between out-
of-phase angle and the actual operating voltage.
Cigre Science & Engineering • N°7 February 2017
78
7. Acknowledgements
The authors express their sincere thanks to the
contributed to the enquiry on their policy with respect
to power swing protection, and the members of SC B5
and JWG A3/B5/C4.37 collecting this information,
especially to the corresponding members Kevin Jones
(South-Africa) and Ton Geraerds (Netherlands).
8. References
Gridcodes:
[1] ENTSO-E Network Code RfG, “Network Code for Requirements
for Grid Connection - Applicable to all Generators”, www.entsoe.
eu, 2013-03-08
[2] Regulation paper in Russian on the TSO official site, http://so-ups.
ru/fileadmin/files/laws/orders/pr277-300603me.pdf
[3] EirGrid Grid Code Version 6, www.eirgrid.com
Dissertations:
Assets from a Strategic Perspective”, PhD thesis Delft University of
Technology, 2015, ISBN 978-94-62331-25-9
Technical Reports:
[5] IEEE PSRC WG D6 Report to the Power System Relaying
Committee of the IEEE Power Engineering Society, “Power Swing
and Out-of-Step Considerations on Transmission Lines”, 2005-07-
19
[6] ENTSO-E/TEIAŞ publication, “Report on Blackout in Turkey on
31st March 2015”, www.entsoe.eu, 2015-09-21
[7] CIGRE Technical Brochure 316 (2007), “Defense Plan against
Extreme Contingencies”
[8] UCTE publication, “UCTE Final Report System Disturbance on 4
November 2006”, www.entsoe.eu, 2007-01-30
[9] CIGRE Technical Brochure 613 (2015), “Protection of Distribution
Systems with Distributed Energy Resources”
[10] Deutsche Energie-Agentur (dena), Berlin, “Ancillary Services
Study 2030”, www.dena.de,/en
[11] CIGRE Technical Brochure 479 (2011), “International Guide on
the Protection of Synchronous Generators”
[12] CIGRE Technical Brochure 187 (2001), “System Protection
Schemes in Power Systems”
[13] CIGRE Technical Brochure 608 (2015), “Lessons learnt from
Recent Emergencies and Blackout Incidents”
[14] Electra No. 50, 1977, pp. 77-92, WG 34.01, «Protection against
Out-of-Step Operation of Large Synchronous Machines»
[15] CIGRE TB 231 (2003), “Definition and Classification of Power
System Stability”, (IEEE/CIGRE JTF)
[16] ENTSO-E publication, “Report on Technical Background and
Recommendations for Defense Plans”, www.entsoe.eu, 2011-01-31
[17] CIGRE Technical Brochure 664 (2016), “Wide area protection &
Control technologies”
[18] ENTSO-E Position Paper, “Dispersed Generation Impact on
Continental Europe Region Security”, www.entsoe.eu, 2014-11-15
[19] E. Thuries, V.D. Pham, J. Vadaszi, G. Borgonovo, L. Barlusek, K.
Nakanishi, G. Balan (CIGRE TF 13.00.2), “Generator circuit-
breaker. Transient recovery voltages under load current and out-
of-phase current switching conditions”, Electra 126, pp. 54-60
Periodicals:
[20] I.M. Canay, D. Braun, G.S. Köppl, “Delayed Current Zeros due to
Out-of-Phase Synchronizing”, IEEE TEC Vol. 13, No. 2, June 1998,
pp. 124-132
[21] S. M. Rovnyak, C. W. Taylor, Y. Sheng, “Decision Trees Using
Apparent Resistance to detect Impending loss of Synchronism”,
IEEE TPD, Vol. 15, No.4, Oct. 2000, pp. 1157-1162
[22] K.R. Padiyar, S. Krishna, “Online Detection of Loss of Synchronism
Using Energy Function Criterion”, IEEE TPD, Vol.21 No.1, Jan.
2006, pp. 46-55
[23] K. Matsuzawa, K. Yanagihashi, J.Tsukita, M. Sato T. Nakamura,
A. Takeuchi, “Stabilizing Control System Preventing Loss of
Synchronism from Extension and its Actual Operating Experience”,
IEEE TPS, Vol. 10, No.3, Aug. 1995, pp. 1606-1613
[24] T. Minakawa, Y. Ichikawa, M. Sato, Y. Ishihara, “A New Method for
Detecting Loss of Synchronism Using Power and Current Measured
on a Line”, IEEE TPD, Vol. 14, No.1, Jan. 1999, pp. 68-73
[25] Ph.Denys, C.Counan, L. Hossenlopp, C. Holweck, “Measurement
of Voltage Phase for the French Future Defence Plan against Losses
of Synchronism”, IEEE TPD, Vol. 7, No.1, Jan. 1992, pp.62-69
[26] T. van Cutsem, C.D. Vournas, “Voltage stability analysis in
transient and mid-term time scales”, IEEE TPS, Vol. 11, No.1, Feb.
1996, pp. 146-154
[27] J.G. Rao, A.K. Pradhan, “Power-Swing Detection Using Moving
Window Averaging of Current Signals”, IEEE TPD, Vol.30, No.1,
Feb. 2015, pp. 368-376
[28] P.K. Nayak, A.K. Pradhan, P. Bajpal, “Secured Zone 3 Protection
During Stressed Condition”, IEEE TPD, Vol.30, No.1, Feb. 2015,
pp. 89-96
[29] W.M. Leeds, D.J. Povejsil, “Out-of-Phase Switching Voltages and
Their Effect on High-Voltage Circuit Breaker Performance”, AIEE
PAS January 1952, pp. 88-96
Papers from Conference Proceedings:
[30] A. Kubis, C. Rehtanz, J. Xu, Y. Li., “About the Relationship of Line
Overload Cascading and Loss of Synchronism in Electric Power
Systems”, PowerTech 2015, Eindhoven, Report 461936
[31] A.L.J. Janssen, M.v. Riet, R.P.P. Smeets, J. Kanters, W. v.d.
Cigre Science & Engineering • N°7 February 2017
79
Akker, G.L.P. Aanhaanen, “Prospective Single and Multi-Phase
Short-Cirquit Current Levels in the Dutch Transmission, Sup-
Transmission and Distributions Grids”, CIGRE SC A3 Session
2012, Report A3-103
[32] A. Kubis, S. Rüberg, C, Rehtanz, “Development of Available
Short-Circuit Power in Germany from 2011 up to 2033”, CIRED
Workshop 2014, Rome, paper 379
[33] G.D. Jennings, L.N.F. de Villiers, “Transient stability improvement
using series reactors: a Case Study”, IEEE/EPEC, London (2015),
978-1-4799-7664-5/15
[34] M. Palazzo, M. Delfanti, “New Requirements for the Application of
Generator Circuit-Breakers”, IPST 2013, Report 088
[35] Andreas Klinke, Ortwinn Renn, “A New Approach to Risk
Evaluation and Management: Risk-Based, Precaution-Based,
and Discourse-Based Strategies”, Risk Analysis, Vol. 22, 2002, pp.
1071-1094
[36] D.Z. Meng, “Maintaining System Integrity to Prevent Cascading
Blackout”, CIGRE SC B5 Session 2006, Report B5-207
[37] S. Agematsu, S. Imai, R. Tsukui, H. Watanabe, T. Nakamura,
T. Matsushima, “Islanding Protection System with Active and
Reactive Power Balancing Control for Tokyo Metropolitan Power
System and Actual Operational Experience”, IET: 7th Int. Conf. on
Developments in Power System Protection (2001), pp. 351 – 354
[38] T. Yasuda, Y. Suzuki, Y. Ishikawa, H. Oshida, Y. Kowada,
H. Nishino, «TEPCO’s Experience of Stability Problems and
Development of Wide Area Special Protection Scheme without
Telecommunication», APAP 2015, 6th  Int. Conf. on Advanced
Power System Automation and Protection, Nanjing, pp 306-313
[39] Релейная защита и автоматика. Автоматическое
противоаварийное управление режимами энергосистем.
Автоматика ликвидации асинхронного режима. Нормы и
требования / Стандарт организации ОАО «Системный
оператор ЕДИНОЙ Энергетической системы» / Москва
2015. Relay protection and automatics. Power system state
automatic operation. Out-of-step termination automatics.
Requirements and standards. / Company standard. System
operator of Russia / Moscow 2015
[40] P. Daly, D. Flynn, N. Cunneffe, “Economic Dispatch for Systems
with High Non-Synchronous Penetrations”, PowerTech 2015,
Report 462401
[41] M. Palazzo, M. Popov, A. Marmolejo, M.Delfanti, “Revision
of TRV Requirements for the Application of Generator Circuit-
Breakers”, IPST 2015, Report 221
[42] Kevin W. Jones, “Power Swing Relaying in the Texas Panhandle –
An Out-of-Step Odyssey”, A&M/IEEE Conference for Protective
Relay Engineers, Texas, 2015
Standards:
[43] IEV 60050-448-14-35, International Electrotechnical Vocabulary,
Chapter 448: “Power system protection” , 1995
[44] IEC Standard 62271–100, ed. 2.1 (2012), “High-voltage switchgear
and controlgear - Part 100: High-voltage alternating current circuit
breakers “
[45] IEC Standard 60034-3, ed. 6.0 (2007), “Requirements for
synchronous generators driven by steam turbine or combustion gas
turbine”
[46] IEEE Standard 50.13 (2014), “Standard for cylindrical rotor
50 Hz and 60 Hz Synchronous Generators rated10 MVA and
above”
[47] IEEE/ANSI Standard C37.06, “AC High-Voltage Circuit Breakers
Rated on a Symmetrical Current Basis Preferred Ratings and
Related Required Capabilities”
[48] Standard IEC/IEEE 62271-37-013, ed. 1.0 (2015), “High-voltage
switchgear and controlgear – Part 37-013: Alternating-current
generator circuit-breakers”

Cigre Science & Engineering • N°7 February 2017

80
Managing regional security of supply:
a case study from Scotland
S. Gill*, G.S. Hawker, K.R.W. Bell,
University of Strathclyde, United Kingdom
Abstract:
Security of supply within a region of a power system
can be achieved using generation within that region, or
transmission import capability coupled with generation
elsewhere. Great Britain, system planners cannot
directly influence the location in which generators are
built, and instead maintain regional security of supply by
developing the transmission system via a deterministic
standard which takes as inputs expected values of peak
demand and generation fleet parameters in each region.
This is in contrast to the probabilistic standard used to
define system wide generation adequacy through a Loss
of Load Expectation. This paper proposes a probabilistic
standard for regional security which is used to determine
the secure import required from the transmission
network for a given level of security. The method is
applied to Great Britain and Scotland using historical
data for demand and generation availability from recent
winters. The paper concludes that a probabilistic metric
provides greater information on the level of regional
security provided, and allows the impact of all types of
generation, including intermittent renewables generators
such as wind, to be properly accounted for.
1. Introduction
Maintaining security of supply is a major goal in
the planning and operation of electricity systems. It
involves ensuring that sufficient generation is available
across the power system – generation adequacy – and
that sufficient transmission capability is available to
transfer that electricity to where it is needed. Security
of supply within a region of a power system depends
on the type and geographical distribution of generation
*simon.gill@strath.ac.uk
KEYWORDS
Power System Security, Reliability, Security of Supply, Transmission Adequacy, Wind Generation.
Cigre Science & Engineering • N°7 February 2017
81
and sufficient transmission capability to allow access to
generation outside the region. In theoretical terms, the
problem is one of co-optimisation of generation and
transmission including investments and O&M costs in
which the collection of services offered by each type of
asset (including energy, capacity and various ancillary
services) are all fully considered. In the reality of
modern liberalised markets, the interests and split of
responsibilities between parties, and the imperfect nature
of the instruments available to influence investment
mean that such an optimal trade-off between generation
and transmission is difficult to achieve.
In Great Britain (GB) the current arrangements mean
that regional security of supply is defined in terms
of the required secure transmission capability in the
Security and Quality of Supply Standard (SQSS) [1].
The capability of transmission required is a function of
peak demand and the generation mix within that region
(and across GB) and there is no opportunity (with the
exception of some short term arrangements) to mandate
that generation should be built within a particular
region in order to support security. In addition, as the
GB electricity market is largely decentralised with no
consideration of transmission constraints in bilateral or
power exchange trades, there has been little incentive for
generators to consider regional security or congestion on
the transmission network when deciding where to locate.
The one instrument which does provide a locational
signal is the Transmission Network Use of System
(TNUoS) charging mechanism. Part of this charge on
users of the transmission network is location dependent
and is designed to reflect the cost of network required to
serve either generation or demand in particular regions.
The TNUoS methodology has recently been updated by
splitting generation charges into peak-security and year-
round components, the latter of which reflects the role
of transmission in facilitating an effective market [2],
[3]. The aim of the cost-reflective aspect is to provide
a location-incentive on generation connections, and the
splitting out of security and economic backgrounds has
the potential to encourage different types of generation
in different areas. Therefore low or negative peak-
security charges may encourage peaking plant to a
region dominated by intermittent generation. However,
the ability of TNUoS charges to do so depends on both
the size of the locational element and the split between
which generators pay which component.
A second development taking place in 2016 is the
Integrated Transmission Planning and Regulation
project (ITPR) [4]. The development and operation
of the transmission network has historically been a
collaboration between the System Operator (SO) –
National Grid – and Transmission Owners (TOs) of
which there are three in GB: National Grid in England
and Wales, Scottish Power Transmission in southern
Scotland, and Scottish Hydro Transmission in northern
Scotland (although National Grid’s SO and TO activities
in England and Wales have only recently become quite
distinct from each other). Two major changes under ITPR
are the enhancement of the SO’s role, and the opening up
of major new ‘high value and separable’ transmission
assets to competition. This will increase the number of
parties involved in maintaining security of supply as it
will, in the case of major transmission assets, introduce
third party TOs responsible for building and maintaining
parts of the transmission network.
Scotland is one part of the GB system where security
of supply issues have come to the fore in recent years
due to the closure of dispatchable fossil-fuel plant and
the anticipated end of life of baseload nuclear generators
over the coming decade. This is coupled with a huge
growth in wind generation. For many years, Scotland
has been a net exporting region, with transmission
upgrades tending to be driven by consideration of export
from Scotland. With the closure of a large fraction of
the remaining conventional capacity1 expected by 2030
Scotland will become more reliant on imports at time
__________________
1- ‘Conventional’ is used here to mean: coal, gas, nuclear, hydro and pumped storage generators.
Cigre Science & Engineering • N°7 February 2017
82
of low wind availability and not only the transmission
export capability but also import capability and security
of supply must be assessed to identify the need for
further investment. One aspect which will need careful
consideration is the role of wind power. The SQSS, the
TNUoS regime and new approaches via ITPR all explicitly
state that regional security of supply should not depend
on wind or other intermittent renewable generators. For
example, the Network Options Assessments (NOA)
introduced as part of the enhanced SO role within ITPR,
states that ‘[t]he security criterion is intended to ensure
that demand can be supplied securely, without reliance on
intermittent generators or imports from interconnectors’
[5]. As shown in [6] and in the work reported below,
whilst not ‘dispatchable’, when its impact is considered
from a probabilistic perspective, wind generation can
and does support regional security of supply by reducing
the probability that demand cannot be served.
The probabilistic nature of system planning is clearly
recognised by methods used to define the required
level of generation adequacy. In GB this is specified
by a maximum Loss of Load Expectation (LoLE)
index which defines the expected maximum number of
hours per year that generation availability across GB is
insufficient to meet demand. The current target value of
LoLE (3 hours) stems from a trade-off between the cost
of new capacity and the value consumers are judged to
place on being supplied [7]. Assessment of generation
adequacy for the capacity market takes no account of
generator location or transmission limitations. In respect
of transmission requirements, the SQSS identifies a level
of import that should be accommodated which is based
on pessimistic but not extreme assumptions about the
total available generation and demand in the importing
area [8]. However, this basic deterministic criterion
does not allow the specifics of particular regions to be
considered, nor the specific impact of particular types
of generators.
A probabilistic analysis would allow contributions to
security to be identified among multiple generators
within a region including wind power, and clarify the
remaining requirement for transmission, which might
then inform mechanisms to charge and reward those
 Figure 1: location of main generation plant within Scotland (as of December 2015).

parties reflecting their contribution. When comparing the Table I: breakdown of main generation plant in Scotland
as of December 2015.
level of imports required to the transmission networks
_____________________________________________________________
import capability, care should be taken to model the
network, the initial generation dispatch and outages Generation Main Installed Notes
Type Stations Capacity
appropriately as power is rarely shared equally between (MW)
________________________________________________________________
circuits on a region’s boundary and the critical outage or
limiting factor may be remote from that boundary. The Due to Close
Coal Longannet 2400
March 2016
deterministic rules in the SQSS relating to conditions at _____________________________________________________________
time of peak demand address either a relatively modest Operational capacity
Gas Peterhead CCGT 740
transfer with a double circuit outage, a single fault in 2015/16
_____________________________________________________________
outages on a system that already has one other network
Hunterston B 1016
outage, or a higher transfer with a single outage. The Nuclear
first two condition are typically the more limiting ones. Torness 1280
_____________________________________________________________
Thus, in this paper the term ‘secure capability’ refers to Pumped Cruachan 440
N-2 security, that is, secure after a double circuit outage storage Foyers 300
or one fault following some other prior outage. _____________________________________________________________
31 Locations with 39
This paper uses the example of Scotland as a region of BM wind a
Balancing Units 3142 As of Autumn 2015
the GB power system to consider the need for secure _____________________________________________________________
import capability in order to secure demand in the Other wind Distributed 2170
region. The impact of major plant closures and the role _____________________________________________________________

of the existing and an expanded Scottish wind fleet are
investigated, before discussing how the results suggest
various parties contribute to regional security of supply.
BM hydro a 896
_____________________________________________________________
a
BM generators are those that are active in the balancing mechanism with data on
availability at half hour resolution, it includes a mixture of transmission and distribution
connected generators. ‘Other’ generation is not monitored, and its output appears as a

2. Overview of the Scottish reduction in demand.

Whilst conventional plant is expected to close, wind

electricity system
The main existing power stations in Scotland are shown
in Figure 1 and Table I. In January 2016, four large
conventional power stations were operational. Of these,
Longannet closed in March in 2016 [9], Hunterston
nuclear power station is expected to close by 2023 [10].
The operators of Peterhead CCGT have been involved
in a project to develop CCS capability, however in late
2015 a major source of funding for CCS development
was cancelled by the UK government [11]. Torness
nuclear plant is expected to operate until 2030 [12].
generation capacity is expected to increase. The Scottish
Government has a target of annual renewable electricity
production equalling the equivalent of 100% of Scotland’s
electricity demand by 2020 [13]. Legally this target is
non-binding (and it should be noted that energy policy is
not devolved to the Scottish Government meaning that
its greatest influence in respect of energy arguably lies
within its role in the planning and consenting regime)
but it represents a strong desire for decarbonisation in
Scotland. Importantly, Scotland is attractive to wind
developers as it has high wind speeds and much of

Cigre Science & Engineering • N°7 February 2017

83
 Figure 2: (a) cumulative distribution for demand in 2015; and (b) cumulative distribution for import to Scotland
(negative values represent exports) estimated from demand, nominated generation and constraint actions.
it is relatively sparsely populated. In January 2016,
approximately 5GW of wind generation was operational
in Scotland, with a further 1.6GW in construction and a
further 6.8GW consented [14].
Scotland is connected to the rest of GB via two 400kV
double circuits (and some minor 132kV circuits)
crossing the ‘B6 Boundary’. Transmission boundary
capabilities are listed annually in National Grid’s
Electricity Ten Year Statement, and the value quoted
for export from Scotland in 2016/17 is 3.5GW [15];
the import capability is 2.6GW [16]. New transmission
infrastructure is expected to be commissioned in 2017
increasing both the import and export capability.
Replacement of conductors on circuits within Scotland
in the near future will increase the export capacity to
4.4GW; and a 2.2GW, 600kV embedded HVDC link
from Hunterston in central Scotland to Deeside in North
East Wales will increase the secure export capability to
6.6GW and the secure import capability to 3.8GW [15].
Figure 2 shows cumulative distributions for demand in
Scotland [17], and estimated boundary flows for 2015
based on nominated outputs from generators in Scotland
at gate closure known as Final Physical Notifications (1
hour ahead of delivery) and post-gate closure balancing
actions taken by the SO [18]. Peak demand in Scotland
over the winter of 2014/15 was approximately 5.5GW
[19].
Figure 2 (b) shows that Scotland currently exported
the vast majority of the time: 93% across the full year
of 2015. The two nuclear power stations operated as
baseload when technically available, and generation at
Longannet coal power station was used the majority
of the time. Periods of import were mainly during day
time, concurrent with summer maintenance outages at
the main power stations and coinciding with low wind
availability.
Estimated historic exports from Scotland in 2015
reached a maximum of 3.5GW consistent with the secure
export limit reported in [16]. However, ensuring that the
boundary flows remain within that limit requires action
by the SO to constrain down generation within Scotland
when flows would otherwise exceed it. Two additional
Cigre Science & Engineering • N°7 February 2017
84
curves are shown in Figure 2 (b): the situation if wind
generation had been allowed to run un-constrained,
and the situation if all Scottish generation had run
unconstrained – i.e. if the SO had taken no actions to
keep export levels within limits. Such unconstrained
operation would require significantly greater export
capability than is currently provided, with exports rising
to greater than 5GW during some periods.
3. Security of supply in Scotland
Security of supply calculations attempt to strike a balance
between the risk of needing to disconnect customers and
the cost of providing the capability to meet demand. In
this paper we calculate an estimate of the secure import
capability that would be needed in order to ensure that
demand in Scotland is met for all but a certain number
of hours within a winter. This definition is similar to
that of LoLE used in generation adequacy studies. To
differentiate the transmission adequacy calculation we
call this the ‘Regional Demand Reduction Expectation’
(RDRE) requirement. To illustrate the method we study
two levels of security: an RDRE requirement of no more
than 3 hours per winter (in line with the current GB
generation adequacy standard), and a stricter limit of 1
hour per winter. Four scenarios are investigated:
1. Base case: the intact system during the winter of
2015/16 including Longannet power station.
2. Longannet Closed: Longannet power station closed,
representing the system after March 2016.
3. Low conventional: Longannet, Peterhead and
Hunterston closed.
4. High wind future: Longannet, Peterhead and
Hunterston closed and wind scaled up to represent an
additional 5 GW of capacity.
In each case, demand in Scotland is distributed as it
was over the two winters of 2013/14 and 2014/15.
Generators including hydro and pumped storage are
modelled at their 2015 capacities, and for scenarios 1
– 3 two sensitivities are run to identify the role of the
existing wind fleet on security of supply:
a. Existing large Wind Fleet: uses historic data from
 Figure 3: (a) Winter historic generation availability in Scotland 2013 – 15. ‘Conventional’ includes coal, nuclear, gas, hydro and pumped storage;
‘Total’ includes all conventional plus wind. (b) Availability representing conventional generation in the main scenarios.

the GB balancing mechanism for availability at 4. Winter generation availability

transmission monitored wind farms (including some
embedded generation) during the winter months
of 2013-15; this was, on average, 2500MW. Other
embedded generation, typically small generators, is
not reported however its output is accounted for in
the fact that the demand data used is demand seen
by the transmission network and therefore net of un-
monitored generation.
b. Zero ‘BM’ Wind: the nominated output and
availability of wind farms reported in the Balancing
Mechanism – ‘BM’ – is removed; the impact of un-
monitored generation in the demand profile remains.
To estimate the import distribution for each scenario,
empirical distributions of demand and total generation
availability are created from historic data for Scotland
drawn from winter months. Whilst a purely time-series
based analysis over a sufficient time-horizon may give
an indication of the distribution of import requirements
which can occur in the future it is highly likely that it
will not capture the full range. This is particularly true
when considering periods of particularly high demand,
for example above the ACS peak demand, as these will
occur, at most, only a few times each winter. These
distributions are then convolved to give an estimate of
the distribution of imports, and this allows the secure
import requirement required to provide a particular
RDRE to be identified.
and demand in Scotland
Records of availability for each generation unit in GB
Balancing Mechanism in the form of Maximum Export
Limits (MEL) are available from [18] – that is, the
maximum level of output that the plant can provide at
a given time. Figure 3 shows the historic distribution of
generation availability in Scotland during the months
of November – February inclusive for the years 2013
to 2015. This has been split into ‘conventional only’
(which excludes wind) and ‘total’. The shape of the
conventional line shows that generation availability is
not smooth, reflecting the fact that whilst occasionally
generators may operate de-rated, most of the time they
are either fully available or fully unavailable, and this
is driven primarily by the rate of forced-outages (rather
than planned outages which are generally avoided during
high demand periods). Scotland, before the closure of
Longannet included a fleet of 9 large units, and the main
peaks of the distribution shown in Figure 3 represent
occasions where various combinations of those units are
available.
The addition of wind to conventional generation
produces a smoother distribution. The absolute lower
ends of the two distributions are similar, highlighting
the fact that there are occasions where very little wind
generation is available due to calm conditions across

Figure 4: Distribution of winter demand for GB and Scotland 2010 – 2015

as a fraction of weather corrected peak demand.

Cigre Science & Engineering • N°7 February 2017

85
the whole of Scotland. However, the relatively rare co-
incidence of low wind availability and low conventional
availability across fleet reduces the probability of very
low total generation availability.
Figure 4 shows the distribution of demand in Scotland
and GB during the months of November – February
inclusive across five winters. Demand for each winter
is normalised by the reported weather corrected peak
demand for that year [20] allowing demand from
different years to be compared; values greater than 1
represent periods where outturn was greater than the
weather corrected peak, for example during periods of
cold weather which can drive high demand peaks.
5. Modelling boundary requirement
distributions
Where independence between availability of generation
and the level of demand can be assumed, the distribution
of import required to secure demand in that region can
be calculated as the convolution of the demand and
generation distributions. There are some conditions
under which the assumption of independence is not valid:
importantly, there is some evidence that the availability
of wind has a negative correlation with demand for very
high demand levels (see for example [21]); and, away
from absolute peak demand, conventional un-availability
may include some discretionary maintenance. In the
analysis below the assumption of independence is used,
and further discussed in the Section 7.
Figure 5 to Figure 7 show distributions for the required
import into Scotland calculated via a direct time-
series analysis for outturn data for winter months
(November – February inclusive) during the calendar
years 2013 – 2015; and via the convolution of the
demand, conventional availability, and wind availability
distributions taken from Figures 3 and 4. Results for both
methods are presented for comparison whilst further
analysis uses the convolution results.

Figure 5: (a) cumulative and (b) density plots of convolved and time series estimations (denoted C and TS respectively)
of boundary import requirements for the base case scenario with and without wind.

Figure 6: cumulative distribution for ‘Longannet Closed’ and ‘Low Conventional’ cases estimated by convolved and time series analysis
(denoted C and TS respectively) with and without the existing large wind fleet

Cigre Science & Engineering • N°7 February 2017

86
 Figure 7: The impact of wind capacity on boundary requirements for the ‘Low Conventional’
and ‘High Wind Future’ scenarios.
The sign convention of each distributions is chosen so
that the need for import into Scotland is represented by
a positive number. For negative ‘import’ values there
is in effect zero necessity for import into Scotland
(Although the market may choose to import under some
circumstances, it is not required in those conditions
for security reasons. Negative values reflect the secure
boundary capability required in order to allow all
available Scottish generation to be used - maximum
unconstrained export).
Results show that there is very limited requirement for
import capability in the base case. Import is required for
security reasons during only 1.6% of the winter when
if wind is not considered, or 0.02% of the winter when
wind is considered.
The closure of conventional plant will increase the need for
import. Figure 6 shows results for the two scenarios with
station closures. The closure of Longannet increases the
fraction of time that Scotland is likely to require imports
for security reasons from 1.6% to 10.6% without wind and
from 0.2% to 1.8% with wind. The closure of Peterhead
and Hunterston further increases the fraction of time import
is needed to 69% without wind and 24% with wind.
6. Defining a secure import
capability requirement
The secure import capability required for a particular
RDRE requirements can be calculated from the
distributions shown above. The 1 and 3 hours RDRE
requirements used in this paper correspond to finding
the P99.90 and P99.96 value of the boundary flow
distribution. Table II shows estimated secure import
capabilities for the scenarios and sensitivities studied
calculated against the RDRE requirement and compares
these to values calculated using the current SQSS
deterministic methodology.
The results highlight several important issues. In the
base case the SQSS does not require a secure import
capability because Scotland is defined as an exporting
region. (The ‘planned transfer’, which can be best
Cigre Science & Engineering • N°7 February 2017
87
thought of as a market neutral median power flow at time
of system peak demand, is from Scotland to England).
However, a small import capability is required under
the RDRE calculations if wind generation is ignored.
With wind included the RDRE method agrees with the
SQSS. Whilst in this case the difference is marginal, it is
likely that in future it will not be appropriate to designate
regions as purely importing or exporting; with high wind
penetration and low conventional capacity in a region,
it will be important to define both import and export
requirements for each region.
Whilst the SQSS explicitly ignores wind capacity in
the security calculation, the impact of wind on the
distribution of boundary flows can be seen in the RDRE
values of Table II. With Longannet closed, results with
and without wind show that the 2.5GW of wind capacity
added between the two sensitivities reduces the required
import capability by 630MW and 590MW for the 3
hour and 1 hour cases respectively; this is, in effect, a
‘capacity credit’ of approximately 25%. Similar results
are seen in the Low Conventional scenario, although the
impact of the 2.5GW wind capacity is slightly reduced.
The addition of an extra 5GW of wind in the High Wind
Future scenario further reduces the required import
capability, however the impact is significantly smaller
and the reduction in secure import requirements due to
this extra generation is only 5% of its capacity.
The reason for the relatively high impact of the first
2.5GW of wind is that while the availabilities of wind
within Scotland is strongly correlated across wind farm
sites at any one time due to weather conditions across the
region. Wind availability, as a whole, is not correlated
with the availability of conventional units (and therefore
provides a source of generation-availability independent
of other sources) but new wind farms added to an
existing wind fleet are correlated to existing generation
and therefor have a smaller effect on secure import
requirements. In this respect new wind added to an
existing wind fleet contrasts with new conventional
units: an additional conventional unit provides a new
independent source of generation availability, whilst an
additional wind farm does not.
Table II: required secure import capacities (MW) in order to limit the but, by virtue of its location, would have contributed
expected RDRE within Scotland to less than 3 hours and 1 hour per winter,
and comparison against results using the current SQSS methodology. reactive power, black start capability and offset the need
for transmission reinforcements to support increased
3 hours 1 hour
failure to failure to Current
imports into an area. This suggests consideration should
Scenario Wind? supply supply SQSS be given to the procurement of bundles of services
regional regional method rather than individual services, though this has the
load load
disadvantage that the prices of specific services are
No 120 290 0
Base difficult to discover and, hence, market information is
Yes 0 0 0
less available to potential new providers.
Longannet No 1600 1750 1840
Closed Yes 970 1160 1840 Planning and charging methods continue to discount
Low No 2600 2710 3215 wind entirely from regional security of supply
Conventional
Yes 2050 2250 3215 calculation. In particular, the recent TNUoS revisions
(Longannet,
Peterhead and Yes + exempt wind from the ‘Peak Security’ charge. Whilst
Hunterston additional 1760 1950 3215 it is true that wind cannot reliably provide full power
closed) 5GW
on-demand to support security of supply, the same is
actually true of any individual conventional generator.
7. Discussion: multiple parties A CCGT generator, for example, has approximately
providing security of supply an 88% probability of providing full availability and a
12% probability of providing zero [18]. Availability of
The optimal situation for Scottish security of supply the Scottish wind fleet has a different distribution and
is to have the least cost combination of transmission analysis of historical meteorological data suggests an
and generation infrastructure combined with sufficient availability of at least 10% for 82% of the time - ignoring
generation adequacy across GB. However, this objective a large wind fleet entirely when calculating transmission
is embedded within a range of others including requirements or charging for access at peak times will
participation by generation in the GB electricity market ignore this contribution. With any regional generation
and the facilitation by transmission of competition fleet, the contribution of generation to security within
among generators. The latter is represented by the that region is not simply additive over individual units
capability to export power from a region that has the but arises from the probabilistic interaction of un-
cheapest generation and a surplus of available power correlated (or partially correlated) sources of generation.
relative to demand in that region. Particular generators
The results of the analysis presented here suggest that,
may also be able to provide ancillary services such as
to some extent, both conventional and wind generation
primary, secondary or tertiary reserve, reactive power
can support regional security of supply, although the
or black start capability. Although in respect of reserve
credit given to wind power should depend on the total
it is not always considered; a generator’s value in
size of the existing fleet. Further, a more detailed
respect of ancillary services will depend on location and
analysis of any correlation between wind availability
transmission network constraints. The various services
and demand may be required. Studies used to inform
are typically provided by multiple parties, and allocating
the use of ‘Effective Firm Capacity’ for wind power in
the proportion of each service provided by each party
the GB generation adequacy calculation have for several
is part of the regulatory regime. Different services are
years noted that there is some evidence of a negative
typically procured through separate markets. However,
correlation between wind and demand at high levels of
this runs the risk that, where a provider is tendering for
demand, for example where demand is greater than 0.95
multiple separate services it may appear expensive for
time the expected weather corrected peak demand for
each service in isolation; but may be capable of providing
that winter [22]. However, the authors of [21] argue that,
several services at a lower overall price that the cheapest
due to the relatively small number of times this occurs
bidder in each individual tender. An example of that
and the short time span over which significant wind
could be a generator that does not win a capacity contract

Cigre Science & Engineering • N°7 February 2017

88
generation has been installed, there is (and will continue
to be) insufficient data to draw meaningful conclusions.
In 2016 the GB system operator moved their baseline
assumption in this area to one which assumes some
level of correlation between wind and demand [23].
The results show that for a wind capacity in excess of
15GW across the whole of GB in the winter of 2016/17,
the range of assumption regarding wind and demand
correlation lead to very little change in the capacity of
conventional generation capacity required.
If wind is negatively correlated with high demand, the
true impact of wind on regional security of supply would
be reduced compared to the regional capacity values
calculated in this work, and such analysis should form
part of future studies. Whilst operational data will remain
insufficient to provide statistically significantly results,
it may be possible to construct the relevant correlations
through synthesis of wind power outputs at given times
in the past from historic meteorological records and
compare them with the demands at that time. (Such
synthesis methods are described in, for example, [24])
The combination of greater wind capacity and lower
conventional capacity in Scotland will mean that the
transmission network connecting Scotland to the rest
of GB will play more of a dual role, facilitating both
economic export and security related import. A limitation
of the current SQSS is that the required boundary
transfer capability is generally defined only in one
direction, determined by a calculation of the ‘planned
transfer’2. As has been shown above, an area with a
mix of wind and conventional generation can spend
almost equal periods of time importing and exporting
and it is important that these considerations are reflected
in the standards: both import and export capabilities
should be defined in respect of each region. However,
transmission expansions predicated in serving power
flows in one direction will generally also help in the
other. For example, the Western HVDC link embedded
within the GB synchronous AC system is driven by the
need for greater export capacity year-round to allow
wind generation in Scotland to meet demand across GB.
However, the link can also contribute to Scottish security
____________________________
2- To be more precise, the SQSS defines two ‘planned transfers’: one intended to reflect ‘economy driven’ need and the other ‘security driven’ need. Depending on
the ‘generation background’, it is possible that these would point in opposite directions.
Cigre Science & Engineering • N°7 February 2017
89
of supply. One issue to address though for this project, is
the risk of commutation failure in the absence of a strong
voltage source at Hunterston. On a wider point, where
transmission upgrades are driven from one reason (either
economic export or security related imports) it will be
prudent to identify where the upgrade can be designed
to ensure cost effective support for transmissions other
roles.
For some scenarios in which conventional generation
is closed in Scotland (and the direction of the planned
transfer flips relative to today), the transmission
capability required by the SQSS is comparable to that
suggested by the probabilistic assessment presented here.
The impact of wind is one area which the SQSS does
not take account of and this can lead to over estimation
of required transmission import capabilities. A second
effect, in the opposite direction, is that of large population
effects, i.e. the change to the shape of a probability
distribution when generation capacity comprises many
small units compared with that which results from a
few large units with same overall capacity. The SQSS
is predicated on the former whilst Scotland, particularly
after further closures of conventional plant, has the latter
and this can leads to the SQSS under-estimating required
transmission import capability. Taken together, there is
a danger that the two effects can at times cancel each
other out, giving the impression that the deterministic
calculation is given an appropriate result.
When new transmission assets are built and owned by
third party developers as proposed under ITPR it will
become even more important to clearly articulate the
requirements of each component of the system so that
assets designed for one function – such as export –
are capable of performing others – such as import at a
particular level. A number of risks have been identified
associated with increased numbers of transmission
developers and owners [25]. In particular, the risk
associated with development delays has the potential to
impact security of supply if transmission infrastructure is
not delivered in time for planned closure of generation.
During operation, whilst an asset may have been built
to facilitate an economic market, there is likely to be
a higher societal impact associated with unavailability
during periods when it is required to support regional
security of supply. Operation by a third party TO may
therefore require a carefully designed set of incentives
to encourage the owner to ensure availability at crucial
times.
8 Acknowledgements
The authors would like to acknowledge funding for this
work provided by ClimateXChange.
9. References
[1] National Grid. “NETS Security and Quality of Supply Standard
Issue 2.2”, 2012, Available: http://www2.nationalgrid.com/uk/
industry-information/electricity-codes/sqss/the-sqss/
[2] Ofgem. “Project TransmiT: Decision on proposals to change
the electricity transmission charging methodology”, 25th July
2014, Available: https://www.ofgem.gov.uk/sites/default/files/
docs/2014/07/ project_transmit_decision_on_proposals_to_
change_the_electricity_transmission_charging_methodology.pdf
[3] K.R.W.Bell, R. Green, I. Kockar, G. Ault, J. McDonald, “Academic
Review of Transmission Charging Arrangements”, 17th February
2011, Available: https://www.ofgem.gov.uk/sites/default/
files/docs/2011/03/e.on_comments_transmission_charging_
arrangements_0.pdf )
[4] Ofgem. “Integrated Transmission Planning and Regulation (ITPR)
project: final conclusions”, 17th March 2015, Available: https://
www.ofgem.gov.uk/sites/default/files/docs/2015/03/itpr_final_
conclusions_decision_statement_publication_final.pdf
[5] National Grid. “Network Options Assessment Methodology”,
December 2015, Available: http://www2.nationalgrid.com/
UK/Industry-information/Future-of-Energy/Network-Options-
Assessment/
[6] K.R.W. Bell, D.P. Nedic and L.A.Salinas San Martin, “The need for
interconnection reserve in a system with wind generation”, IEEE
Trans on Sustainable Energy, vol. 3, issue 4, November 2012.
[7] Department of Energy Climate Change. “Electricity Market
Reform Delivery Plan”, December 2013, Available: https://www.
gov.uk/government/uploads/system/uploads/attachment_data/
file/268221/181213_2013_EMR_Delivery_Plan_FINAL.pdf
[8] K.R.W. Bell, N. Green, D. Nicol and Chandra Trikha, “Security
criteria for planning and operation in the new GB market”, August
2006, paper C2-108, CIGRE 2006, Paris.
[9] Scottish Power, “Longannet closure marks the end of coal-fired
electricity generation in Scotland”, 24th March 2016, Available:http://
www.scottishpower.com/news/pages/longannet_closure_marks_
the_end_of_coal_fired_electricity_generation_in_scotland.aspx
[10] Office for Nuclear Regulation, “Sites that we regulate”, Accessed:
15/02/2016, [Online] available: http://www.onr.org.uk/regulated-
sites.htm)
Cigre Science & Engineering • N°7 February 2017
90
[11] Department of Energy and Climate Change, “HM Government
Statement to Markets Regarding Carbon Capture and Storage
Competition”, 25th November 2015, Available: http://www.
londonstockexchange.com/exchange/news/market-news/market-
news-detail/other/12597443.html
[12] EDF Energy, “Torness power station”, Accessed 13th December
2016, [online] Available: https://www.edfenergy.com/energy/power-
stations/torness
[13] Scottish Government. “Energy in Scotland 2015”, 2015, Available:
http://www.gov.scot/Resource/0046/00469235.pdf
[14] UK Renewables, “UK Wind Energy Database”, Accessed: 4th
February 2016, [online] Available: http://www.renewableuk.com/
en/renewable-energy/wind-energy/uk-wind-energy-database/
[15] National Grid. “Electricity Ten Year Statement 2016”, 2016,
Available: http://www2.nationalgrid.com/UK/Industry-
information/Future-of-Energy/Electricity-Ten-Year-Statement/
[16] National Grid. “Electricity Ten Year Statement 2015”, 2015, http://
www2.nationalgrid.com/UK/Industry-information/Future-of-
Energy/Electricity-Ten-Year-Statement/
[17] National Grid. “Historic Demand Data”, Accessed: 4th February
2016, [online] Available: http://www2.nationalgrid.com/UK/
Industry-information/Electricity-transmission-operational-data/
Data-Explorer/
[18] Elexon, “Balancing Mechanism Report Archive”, Accessed: 4th
February 2016, [online] Available: www.elexonportal.co.uk
[19] National Grid, SP Transmission, SHE Transmission. “Security of
Electricity Supply in Scotland”, 17th March 2015. Available: http://
www2.nationalgrid.com/UK/Services/Balancing-services/System-
security/Transmission-Constraint-Management/Transmission-
Constraint-Management-Information/
[20] National Grid, “Weather and Seasonally Corrected Demands. Issue
number 2”, 17th March 2015, Available: http://www2.nationalgrid.
com/UK/Industry-information/Electricity-transmission-
operational-data/Data-Explorer/
[21] S. Zachary, C.J. Dent, D.J. Brayshaw, "Challenges in quantifying
wind generation's contribution to securing peak demand", Power
and Energy Society General Meeting, IEEE , pp.1-8, 24-29 July 2011
[22] Ofgem. “Electricity Capacity Assessment Report 2013”, 27th June
2013, Available: https://www.ofgem.gov.uk/sites/default/files/
docs/2013/06/electricity-capacity-assessment-report-2013_0.pdf
[23] National Grid, “National Grid EMR Electricity Capacity Report”
31st May 2016, Available: https://www.emrdeliverybody.com/
Lists/Latest%20News/Attachments/47/Electricity%20Capacity%20
Report%202016_Final_080716.pdf
[24] G. S. Hawker, W. A. Bukhsh, S. Gill and K. R. W. Bell, "Synthesis
of wind time series for network adequacy assessment," 2016 Power
Systems Computation Conference (PSCC), Genoa, 2016, pp. 1-7.
[25] Frontier Economics, “A cost benefit analysis of the potential
introduction of competitively appointed transmission operators”,
January 2016. Available: https://www.ofgem.gov.uk/sites/default/
files/docs/ng_response_appendix_2_fronteir_economics_rpt-cato_
cba-08_01_16_-_final.pdf
Operational security challenges and tools
for a synchronous power system with high
penetration of non-conventional sources
Ivan Dudurych1* (EirGrid), Michael Burke (Saudi Aramco),
Louis Fisher (EirGrid), Martin Eager (EirGrid), Kieran Kelly (EirGrid)
Abstract
In this paper, we describe the effects of increased
renewable sources on the operational security of the
Ireland and Northern Ireland power system. Operational
security comprises a set of criteria that needs to be
met by the power system, in real time. Understanding
these effects allows us to develop and enhance metrics,
techniques and tools to deal with high and variable levels
of non-conventional generation (up to 60% at present)
while maintaining operational security.
The metrics that we use include a System Non-
Synchronous Penetration (SNSP), Rate of Change of
Frequency (RoCoF) and System Inertia. We also discuss
our experience with such tools as the Wind Dispatch Tool,
On-line Short Circuit Tool, and Wind Security Assessment
Tool (WSAT), as well as future developments in this area.
1. Introduction
The generation portfolio connected to many power
systems has changed significantly over the past decade.
Large amounts of variable and non-synchronous
generation technology such as wind and solar are
displacing more conventional generation power plants in
the normal running regime. As the penetration of these
technologies increases, a transmission system operator
has to deal with significant operational challenges. In
the synchronous system on the island of Ireland (Ireland
and Northern Ireland), with only HVDC interconnectors
to Great Britain, these include:
• Significant variations in power system inertia levels, penetrations of variable non-synchronous generation
with periods of very low inertia;
• Increased variations in system voltage profiles with the
creation of zones with elevated or depressed voltages;
* ivan.dudurych@eirgrid.com
KEYWORDS
Control centre tools, Non-conventional generation, Operational Security, Stability, Power transmission System
Cigre Science & Engineering • N°7 February 2017
91
• Increased probability of low-frequency and load
shedding events due to slow active power recovery
of some generation technologies following network
faults; and
• Increased need for real-time energy balancing due to
difficulty in accurately forecasting renewable resource
availability and due to the inherently variable nature
of the renewable resource and the associated ramp up/
ramp down of real power output.
In such variable conditions, operational security of the
power system cannot be adequately supported if based
only on off-line analyses and studies. In an attempt to
cover many different potential scenarios, estimated
security limits would become extremely conservative
for one set of scenarios, while being inadequate for
others. As such, the evolving power system requires new
principles and practices of operation, with a consequential
requirement for the development and implementation of
new tools for control centres that enable operators to
forecast, monitor and control the power system more
dynamically. Such developments are further advanced
by the advent of Demand Side Management (DSM) and
Smart Grid technologies.
This paper looks at the electrical power system of Ireland
and Northern Ireland, which currently operates with up
to 60% of its total instantaneous generation requirements
served by non-synchronous generation technology [1]. In
particular, the paper will examine:
• The impact on the operational security of significant
with a particular focus on thermal security, voltage
security, frequency limits and frequency security;
• The techniques required to deal with high and variable
 Opera&onal
 Security
 
 

Steady-­‐state   Dynamic  

Short-­‐circuit     Synchronous  (Rotor  

current  level   angle  stability)  
Thermal   Non-­‐synchronous  
(Fault-­‐ride-­‐through)  
Voltage  
Voltage  stability  
Frequency  
Frequency  stability  
Reserves  
Oscillatory  stability  
Ramping  

Sta&c  limits      Stability  limits  

Figure 1 Operational Security Components

levels of non-conventional generation, including the 2. Operational security definitions

monitoring of a range of new operational metrics to
help maintain operational security including system Operational security of the power transmission system can
inertia and rate of change of frequency (RoCoF); be defined in terms of steady-state and dynamic security
criteria.
• Experience using new tools in the control centres
including a new wind dispatch tool, an online short Steady-state criteria are defined by static limits (thermal
circuit monitoring tool, an online stability assessment limits, short-circuit current limits, voltage limits and
tool WSAT, and a scheduling tool which incorporates frequency limits). Active and reactive power reserves and
new constraints like inertia and RoCoF, as well as ramping reserves are also included in these criteria.
an Energy Management System (EMS) integration Dynamic criteria, on the other hand, are defined by
facilitating all island power system operation; stability limits and refer to transient stability (rotor angle
• Proposed developments in control centre tools, such stability for the synchronous plant, and fault-ride-through
as a forward-looking security assessment tool, system for the non-synchronous plant), voltage stability and
generation ramping assessment tool and a voltage frequency stability. Also included is oscillatory stability,
trajectory tool. which refers to the ability of the power system to dampen

Figure 2 Wind Generation as percentage of the total island of Ireland Generation

Cigre Science & Engineering • N°7 February 2017

92
 Figure 3 Changes in a CCGT scheduling due to increase of renewable generation

oscillations. Oscillatory stability encompasses small
signal stability, as well as transient and frequency stability
where disturbances could result in unstable oscillations
of active and reactive powers, frequencies and voltages.
Figure 1 shows the structure of operational security
components discussed above.
The components of the operational security and its criteria
are described in [2].
3. Impact of very high penetration of
renewable sources on operational
security
3.1. The Effects of Resource Variability
Renewable resource variability has a profound impact
on operational security. Figure 2 shows wind generation
in the all island (Ireland and Northern Ireland) power
system as a percentage of the total generation for October-
November 2015. It is seen that such generation is quite
variable and changes without pattern from almost zero to
55%. This means that scheduling of the conventional plant
will be significantly different from day to day and from
hour to hour, causing it to change outputs and even start/
stop much more often than in the past. Figure 3 shows
changes in the scheduling of one combined-cycle gas
turbine generator (CCGT) in Ireland between November
2009 and November 2015. The scheduling changes for
this particular CCGT are due to the combined effect of
increasing the renewable generation, connection of a new
CCGT, and cheaper coal prices.
With the advent of renewable and other distributed
generation, transmission systems became subject to a
much greater variety of power flows, while distribution
systems became subject to multi-directional power flows
depending on the prevailing generation conditions. Figure
4 shows two EMS screenshots displaying power flows
at some transmission substations, which have connected
wind generation. Not only can the level of power flow
vary, but also the direction of the flow: with low wind,
power flows from the transmission to the distribution
system, while with high wind it can change its direction
as seen in Figure 4 (right) and net power flow can be from
the distribution to the transmission system.

Figure 4 Power flows at the transmission substations with connected wind generation, with low (left), and high (right) wind conditions: Red arrows
indicate net flow from the transmission to the distribution system, and green arrows indicate net flow in the opposite direction

Cigre Science & Engineering • N°7 February 2017

93
Conventional power stations have generally been built as clearance for which rotor does not slip poles) is below
close to load centres as possible. In contrast, renewable 200ms (i.e. the Grid Code requirement at the time). It is
generation is normally located where the renewable seen from the figure that initially transient stability of
resources are best, which is often far from load centres. the system improves with wind penetration up to around
This can significantly change power flows. For example, 60%. Then it starts declining and sharply deteriorates at
if there is a lot of wind generation in the system remote wind penetration levels of 80%.
from major loads, power flows on the grid can reverse and
reach significant values that were not originally envisaged
for the particular circuits. Such circuits can be highly
loaded or even overloaded when there are outages of
parallel circuits. Therefore, the thermal security needs to
be monitored in real-time to make sure that no cascading
outages occur due to circuit overloads. In addition, short
circuit currents vary in a wide range, making it challenging
to set protection systems at very low short circuit levels.

3.2. Effect of renewables on transient stability

Transient Stability has two components: rotor angle
stability of the synchronous plant, and fault-ride-through
of the non-synchronous plant.
Synchronous (Rotor angle) stability of the synchronous
plant is affected by two phenomena operating with the
increasing penetration of non-synchronous sources. The
first concerns the loading patterns of the synchronous
plant as non-synchronous generation increases. Initially,
as long as the output from the synchronous plant is being
replaced with renewable generation without decommitting
a synchronous unit, the rotor angle stability improves due
to lower loading of the synchronous plant. However, as
more and more synchronous plant becomes decommitted,
the loading of the remaining synchronous plant starts to
increase, leading to a deterioration of rotor angle stability.
The other phenomenon relates to the spatial patterns of the
synchronous plant displacement with non-synchronous
plant. Traditionally, large conventional plants have been
placed near the load centres, but renewable plants are
normally placed far from the load centres. The energy
generated by the renewable plants must therefore travel
long distances. At certain very high non-synchronous
renewables penetration levels, this effect leads to the
avalanche-like deterioration of the rotor angle stability.
This discussion is illustrated in Figure 5, taken from
EirGrid’s Facilitation of Renewables Study [3], which A reduction in mechanical inertia will affect the frequency
shows the percentage of cases where critical clearance
time (CCT, the maximum time of three-phase short circuit
Figure 5 The percentage of cases where critical clearance time (CCT) is
below 200 ms versus wind penetration levels
Non-synchronous (fault-ride-through) stability of the
non-synchronous plant (ability to ride through a three-
phase fault) declines as renewable penetration increases
due to a reduction in system strength (in other words –
short circuit level).
Due to the variable nature of most non-synchronous
renewable sources, they cannot be directly connected
(“synchronised”) with the synchronous system. Instead,
they are coupled with the system partially or fully
via power electronic converters. There are two main
implications of this fact:
• Mechanical inertia of the system is progressively
depleted with increased renewable penetration, and
• System short circuit currents may be very low because
contribution from the converter- based renewable
sources is limited to their maximum load current.
security of the system as will be described in the section
3.4. The low short circuit current leads to the conditions

Cigre Science & Engineering • N°7 February 2017

94
 Figure 6 Active power recovery characteristics of conventional and wind generators

that voltage dips associated with faults (three- or even two-
phase) on the transmission system propagate to a broader
region of the system. This has a direct impact on the fault-
ride-through performance of non-synchronous sources
of generation. The larger the propagation of the voltage
dip into the system, the greater the proportion of non-
synchronous generation that may be exposed to an under-
voltage protection trip. In addition, unlike conventional
generation, some renewable generation, in particular wind
generation, has a slower rate of recovery of active power
following the clearance of the fault as illustrated in Figure 6.
This can lead to a new phenomenon – Voltage Dip Induced
Frequency Dip (VDIFD) – and operational security should
take account of the probability of such an event.
3.3. Effect of renewables on voltage stability
In the early days of renewable energy, renewable generators
did not have any voltage or reactive power control
capability. Soon it was recognised that the availability of
voltage control capabilities of the conventional generators
was increasingly depleted when more of them were
replaced by renewable ones. The Grid Code was updated
to introduce at least some of the capability requirements
of conventional generation for the non-synchronous
generation portfolio. For instance, the Irish Grid Code
requires that all transmission-connected wind farms have
the ability to control voltage and/or reactive power at
their terminals. The full-rated converter grid-connection
technology is best suited for this task as it allows
modulation of the grid-side three-phase voltage according
to the required reactive power. On the other hand, many
distribution-connected renewable sources on the Ireland
and Northern Ireland system are fixed power factor
machines and they do not regulate voltage. This means that
voltage regulation becomes increasingly more difficult in
a system with high penetration of renewables. In areas
with significant renewable generation, network solutions
are required to avoid voltage collapse or excessively high
voltages. It is important for the Distribution Code to be
updated to obtain the reactive power capability from those
wind farms connected to the distribution system without
compromising operations of the distribution system. This
is particularly relevant given that about a half of the total
wind capacity is connected at the distribution level.
3.4. Effect of renewables on frequency stability
The main effect of non-synchronous plant on frequency
stability is related to the fact that renewable generation
provides very little or no inertia to the system, because
it is partially or completely mechanically decoupled
from the power system. The system inertia is getting
smaller and smaller with progressive replacement of the
conventional generation by non-synchronous renewable
generation. This fact is clearly demonstrated at Figure 7
where inertia is represented as kinetic energy stored in the
rotating masses of the synchronous machines.
This means that a sudden imbalance of power will cause a
greater frequency deviation in a system with less available
inertia, as shown in the illustrative Figure 8. The curve
(c) in this figure corresponds to a situation where the
load-shedding scheme would activate. Load shedding,
the automatic disconnection of load, is a last defence
protection and its routine operation is not accepted in a
modern society.
From the above analysis, it follows that we need to
monitor the inertia and frequency stability in real time.
Figure 7 Decline of system Inertia with increasing of wind penetration as %
of generation

Cigre Science & Engineering • N°7 February 2017

95
 Figure 8 Frequency deviations due to sudden trip of a large generator for
(a) a traditional generation portfolio; (b) for a portfolio with a large non-
synchronous generation penetration; (c) for a portfolio with a significant
non-synchronous generation penetration

4. The techniques required to techniques, often referred to as operational metrics, as

deal with high and variable levels
of non-conventional generation
As mentioned earlier, the displaced conventional
generation leads to decreased levels of synchronous
inertia, system strength and voltage support, leading to
the critical need for robust techniques which monitor
operational security in real-time. We developed these
easily comprehensible parameters, which immediately
indicate, at a high-level, the health of the power system
in real time.
System non-synchronous penetration level (SNSP).
The concept of a SNSP level was developed in parallel
with the Facilitation of Renewables Studies [3]. The
objective of the SNSP metric is to have a parameter that,
at a high level, captures a range of operational issues. It is
calculated as

HVDC Interconnector Imports are included here as they
displace the conventional generation together with other
non-synchronous sources. At a constant demand, an
increase in non-synchronous generation would equate
to an increase in SNSP. Maintaining SNSP below a
defined value ensures the robust operation of the system
at differing levels of non-synchronous generation. The
existing maximum SNSP limit is set as 60% on a trial
basis [1]. The maximum SNSP limit is maintained by
curtailment of non-synchronous renewable generation.
System inertia. Monitoring system inertia from
synchronous generation guarantees a minimum level of
conventional generation is synchronised to the power
system. At present this minimum level of inertia is set for
the Ireland and Northern Ireland system as 20,000 MWs
[1]. System inertia is also maintained through an alternative
operational metric, which ensures a minimum number of
generating units are synchronised on the system. This
minimum level of conventional generation is set to provide:
• inertia to manage frequency control and stability, and therefore allow monitoring of this change.
• synchronous torque for system strength and to maintain
transient stability, and
• reactive power support (provided there is a suitable
geographical spread of generation) to manage voltage
control and stability.
Increasing nonsynchronous generation places pressure
on existing system inertia limits. A balance must be
maintained between levels of synchronous and non-
synchronous generation. Lowering the system inertia limit
would permit a reduction in the quantity of conventional
generators synchronised to the system. Maintaining the
existing inertia requirement level would at times increase
curtailment of non-synchronous sources on the system.
RoCoF. Rate of change of frequency (RoCoF) is the
measure of frequency over time following the sudden
disconnection of a generator or an interconnector.
Predominantly, the most vulnerable time for RoCoF to
exceed its limit is in the initial moments after the event. To
this end, Equation (2) is a technique to calculate RoCoF

Cigre Science & Engineering • N°7 February 2017

96
At present, we monitor the minimum RoCoF that could
occur due to the trip of a large generator or importing
interconnector, and the maximum RoCoF that could occur
due to the trip of the exporting interconnector.
As the operational metrics discussed in this section are
new in respect to power system operational practice,
we continue to analyse and modify their optimal limit to
ensure that the power system is securely operated with
increasing amounts of variable non-synchronous sources.
in Belfast. This tool allows the TSOs to directly control
wind generation to a secure level, as dynamically dictated
by various security indices. The wind dispatch tool
allows issuing of setpoint controls to both transmission
and distribution connected windfarms from the EMS [4].
The output of a subset of windfarms may be reduced to
avoid local transmission problems, called constraints, or
all windfarms on the island of Ireland may be reduced to
avoid system issues, called curtailment..

5. Control centre tools Through the tools’ user interface (see Figure 9), the grid
controllers specify the maximum wind production for the
system or a group of windfarms to ensure operational
5.1. Monitoring the new operational metrics security. The tool selects windfarms to be dispatched
The new operational metrics discussed above have been down and calculates setpoints based on a regulatory
included in the generation scheduling tool to ensure defined hierarchy. Such automation is essential for
that operationally secure plans are created and allowing efficient decision making in complex real-time situations.
decisions to be taken in a timely fashion e.g. day ahead. Constraint and curtailment can co-exist - for example,
For example, additional synchronous generation may be a thermal overload contingency may require local wind
required for inertia – optimising this in advance allows generation to be dispatched down while minimum inertia
the transmission system operator (TSO) take more requirements may require concurrent curtailment to avoid
economic actions. It also allows trading of displaced wind having to decommit a conventional unit. When removing
generation on HVDC interconnectors to Great Britain, curtailment, the tool respects active constraint setpoints to
thereby reducing curtailment. avoid exacerbating any local security issue.
Grid Controllers monitor these metrics on-line via
the Energy Management System (EMS), to cater for
unexpected scenarios such as higher than forecast wind
generation or reduced system inertia due to the loss of a
large synchronous generator. We implemented specific
calculations and displays, which show the prospective
RoCoF for the loss of any one of the large synchronous
generators or HVDC interconnectors. To allow pre-
emptive action, grid controllers are warned of situations
approaching operational limits (e.g. 0.45Hz/s RoCoF)
and second level alarms activate once limits have been
breached (e.g. 0.5Hz/s RoCoF).

5.2. Wind dispatch tool

The Grid and Distribution Codes require wind farms above
a minimum level to be controllable by the TSO. The Wind
Dispatch Tool (WDT) is one of the key tools introduced
into the two transmission system control centres on the
island of Ireland; the National Control Centre (NCC) in
Figure 9 Main page of the User-Machine Interface of Wind Dispatch Tool
Dublin and Castlereagh House Control Centre (CHCC)

Cigre Science & Engineering • N°7 February 2017

97
 Figure 10 On-line Short Circuit Calculation Tool

5.3. Online short circuit tool

We developed the EMS Short Circuit Tool (SCT) to
allow continual monitoring of fault levels on the Irish
transmission system. The tool employs the G74 Engineering
Recommendation for fault level calculation, derived from
the International Standard IEC 60909. The user interface of
this tool is shown in Figure 10. The SCT models three-phase
and single-phase faults at designated nodes on the network,
calculates the associated short circuit currents and performs
comparisons with equipment ratings. Alarms are generated
to inform Control Centre engineers when short circuit
currents are approaching limiting ratings. An offline study
SCT is also available which enables efficient assessment
of the impact of adjustments to network topology and re-
Figure 11 Functional diagram of Wind Security Assessment Tool (WSAT)
dispatch of generation.
Significant benefits have been realised following
implementation of the EMS Short Circuit Tool. The key messages from running this on-line real-time
Transmission substations and localised network areas, operational security analysis tool thus far are:
which had previously been sectionalised for short Results have indicated that, at current levels of wind
circuit purposes on the basis of offline studies, can now penetration on the system, the voltage and transient
be coupled as standard. This results in a more robust stability of the system is largely unaffected by the level of
transmission network, improved economy of power flow wind generation. However, this is likely to change as the
and smoother system voltage profiles. Furthermore, the power system evolves and wind generation penetration
continual monitoring of prospective short circuit currents levels increase.
improves system safety by ensuring equipment ratings
can be respected at all times. Grid Controllers have modified the output of generators
per the preventative control measures recommended
by WSAT. In one example, this resulted in the output
5.4. All island WSAT new developments reduction of a generator by 20 MW to avoid possible
Real-time Wind Security Assessment Tool (WSAT) [5] transient instability for a critical contingency. Other
has been operational in EirGrid’s Control Centre in Dublin measures include bringing the concerned generator into
since September 2010. In 2012 it was replaced with the less leading mode in terms of reactive power.
all-island version, and in November 2016 a new, faster WSAT has directly led to changes of transmission line
two-jurisdictional version of WSAT was simultaneously protection settings in areas of the system that are prone to
commissioned in both Control Centres – in Dublin and in severe transient instability problems.
Belfast. The WSAT now runs every 5 minutes assessing
such operational security components as transient, Grid Controllers have used WSAT to monitor an isolated
voltage, frequency, and thermal security performing in area of the system susceptible to voltage collapse; this
total more than 4.5 million tests daily. The functional was not part of WSAT’s original design scope but was
diagram of WSAT is shown in Figure 11, and the WSAT implemented based on increased real time operational
User interface in the electricity control centres in Dublin knowledge of voltage stability. This would not have been
and in Belfast is shown in Figure 12. possible with traditional EMS contingency analysis.

Cigre Science & Engineering • N°7 February 2017

98
 Figure 12 The main page of WSAT User Interface in the Electricity Control Centers in Dublin and Belfast

In a number of cases, the generation dispatch has been
changed to remove frequency insecurity risk (e.g. a large
frequency deviation following a trip of a unit or following
a system separation event between Ireland and Northern
Ireland).
In July 2015, we started providing the Grid Controllers
with real-time information (refreshed every 5 minutes)
about the largest frequency nadir and zenith that can
arise from an N-1 contingency, and from July 2016 we
also included information about maximum and minimum
RoCoF values as shown in Figure 13.
5.5. Integrated EMS
An EMS integration project, which combined the Ireland very fast reserves called Fast Frequency Response, etc.)
and Northern Ireland EMSs into one new system, has will require online monitoring to alert grid controllers to
recently been completed. This allows all-island power
system operation from either control centre, and provides
enhanced power system analysis functionality based on
an all-island network model. One of the main advantages
of the integrated system is the operational ability to fulfil
more complex real-time functions across two jurisdictions.
Specific applications facilitate easier scaling of numerous
wind generators in powerflow studies and improve
monitoring of negative reserve provision (capability
to reduce generation to prevent over-frequency) and
longer term reserve provision (e.g. ramping capability to
balance a renewable forecast error). Improved situational
awareness and all-island wind dispatch capability are
other features.
As a new set of system services products is being
introduced [6], some of them (e.g. ramping products,
inadequate service provision in real-time and to instigate
remedial action.

Figure 13 A sample table showing estimated largest frequency nadir and zenith and maximum and minimum RoCoF together with relevant contingency name

Cigre Science & Engineering • N°7 February 2017

99
6. Future Developments
A key enabler of renewable integration is the development
of innovative tools for the two transmission system
control centres on the island of Ireland. These tools will
help grid controllers to forecast, schedule, monitor and
control the complex power system of the future. Some of
the proposed tools we describe in the following section.
6.1. Forward-looking All Island WSAT
We identified that it would be beneficial to improving our
Unit Commitment Tool outputs to have comprehensive
WSAT-type operational security analyses ahead of time.
For this purpose, we plan to develop a Forward-looking
WSAT (FLWSAT) to predict the power system security
up to 24 hours ahead.
Initial input will be based on the dispatches from the Unit
Commitment Tool, where a number of rule-based security
constraints are used. Based on these dispatches, forward-
looking scenarios will be created in a 24 hours ahead time
horizon. These scenarios will be created within the new
EMS utilising its forward-looking feature.
The initial run of FLWSAT will produce the security
analyses for these scenarios. If the system is insecure,
FLWSAT will suggest preventive measures. Then
additional runs of FLWSAT for the scenario with
preventive measures implemented will take place until the
system becomes secure.
These outputs will be fed into the Unit Commitment Tool
to produce the final dispatches that have all operational
security criteria met.
6.2. Voltage Management
Voltage management is becoming more challenging due
to a vastly increased number of dispersed reactive devices,
each with different capability characteristics; and due
to increased variability in generation and demand. New
tools are required to manage the change, one of which
is a voltage optimisation tool to facilitate evolution from
the current heuristic methods of voltage control. Such a
tool will determine optimal reactive targets for different
types of devices, delivering voltage plans secured against
contingency events for both day-ahead within day re-
optimisation. More efficient methods of voltage control
will also be required. Whereas a few large synchronous
Cigre Science & Engineering • N°7 February 2017
100
generators may be easily dispatched manually, control
of many small devices may require some level of
automation, much like the wind dispatch tool provides for
active power control.
7. Conclusions
Significant experience with the integration of renewable
generation in the Ireland and Northern Ireland power
systems has been built up over the last decade. A thorough
understanding of the main phenomena associated with
the increasing penetration of renewable sources on the
operational security has allowed the development of new
operational metrics and techniques, and the enhancement
of the existing metrics, techniques, and tools to deal with
high and variable levels of non-conventional generation.
The metrics that were developed include: System Non-
Synchronous Penetration (SNSP); Rate of Change of
Frequency (RoCoF); and System Inertia. These metrics,
together with such tools as Wind Dispatch Tool; On-line
Short Circuit Tool; and Wind Security Assessment Tool
(WSAT), are used in the Control Centres in Dublin and
Belfast for monitoring and controlling the operational
security of our power system in real-time.
TSOs continue to develop these techniques and tools to be
able to accommodate ever increasing levels of renewable
sources onto our system that apart from wind will also
include solar, wave, and other technologies coming on-line.
8. Bibliography
[1] EirGrid. Operational Constraints Update 21st November 2016.
[Online]. http://www.eirgridgroup.com/site-files/library/EirGrid/
OperationalConstraintsUpdateVersion1_46_November_2016.pdf
[2] I. M. Dudurych, A. Rogers, R. Aherne, L. Wang, F. Howell and X.
Lin., “Safety in numbers,” IEEE Power & Energy Magazine, pp. 62-
70, March-April 2012.
[3] EirGrid, All Island TSO Facilitation of Renewables Studies. Dublin:
EirGrid, 2010.
[4] M Burke, R Lopez, C Martin, and M Gallagher, “Advanced EMS
Wind Dispatch Tool for the Power System of Ireland and Northern
Ireland,” in Proc. 13th Wind Integration Workshop, 2014.
[5] R. Aherne, J. Conroy, D. Connolly, R. Doyle, I. Dudurych, H. Jones
and A. Rogers,”Design and Implementation of a Tool for Assessment
of Secure level of Wind on the Irish Power System,” in Proc. 2010
CIGRE Session, paper C2-107.
[6] EirGrid Group. Delivering a Secure, Sustainable Electricity System
(DS3). [Online]. www.eirgridgroup.com
9. Biographies transmission system. He is currently Manager of Access
Planning with responsibility for planning connections to
Ivan Dudurych received his EE Degree (with Honour) the transmission grid on the island of Ireland. Formerly
from Lviv Polytechnic University in 1980 and his PhD he has represented EirGrid on the North Sea regional
from Kyiv Polytechnic University in 1990. He has more development group of ENTSO-E and in the North Seas
than 30-years’ international experience in Power Electric Countries’ Offshore Grid Initiative, and has served as
Systems both in industry and academia. At present he is a Ireland’s member of Study Committee 1 of CIGRÉ.
Consultant Engineer in EirGrid. His main specialities are
operational security and grid integration of renewables, Martin Eager graduated in 2005 with an honours
smarter power system operation, protection, modelling bachelor’s degree in electrical engineering from
and control. He has developed a concept, and overseen University College Dublin, Ireland. He presently works
in the Innovation team of the Irish Transmission System
implementation of the world’s first Wind Security
Operator (TSO) EirGrid. He’s primarily involved in
Assessment Tool in the electricity control centres in
power system dynamics and transmission system control
Ireland and Northern Ireland. He has published more than
centre tools. His current focus is on the integration of
80 technical and scientific papers.
large amounts of renewable energy on the all-island
Michael Burke received an honours Bachelor Degree Irish power system. Prior to joining EirGrid he worked
in Electrical and Electronic Engineering from University in manufacturing, software and power system consulting
College Cork in 2003 and the MSc in Electrical Power industries. His previous employment experience includes
Systems from the University of Bath in 2009. He has 13 DIgSILENT Pacific, Australia and Schneider Electric,
years’ experience in the electric power systems industry; Ireland.
primarily in power generation, electricity markets and
Kieran Kelly is originally from Kilkenny, Ireland and
power system operation. At EirGrid Group, he was
graduated from University College Dublin with an
responsible for the development of electricity control
honours bachelor’s degree in Mechanical Engineering in
centre tools to facilitate integrated operation of the power
2012. He currently works in the Operations, Planning and
system of Ireland and Northern Ireland with a high
Innovation department in EirGrid, the Irish Transmission
renewables penetration. He is presently a Senior Power
System Operator (TSO). His areas of expertise include
Systems Engineer at Saudi Aramco.
Asset Management, Project Management, Outage
Louis Fisher graduated in 1981 with an honours bachelor’s Management and the development of new tools for the
degree in electrical engineering from University College electricity control centres of Ireland and Northern Ireland.
Dublin, Ireland. Louis is with the Operations Planning His employment experience also includes working in the
and Innovation Directorate of the EirGrid, and has over tidal energy industry in Ireland as well the manufacturing
35 years of experience in planning and operating the industry in the United States.

Cigre Science & Engineering • N°7 February 2017

101
Reduction of greenhouse gases in GIS pilot
project in UK
E. Laruelle*, A. Ficheux, Y. Kieffel, GE’s Grid Solutions, France
M. Waldron, National Grid, United-Kingdom
Abstract
GE’s Grid Solutions, through its Clean Grid program, has
made a commitment to reduce emission of Green House
Gases (GHG) from its products. The target of this paper is
to present the two main axes of this program.
The first axis is the optimization of existing technologies
by reducing the impact of SF6. The paper presents the
benefits of this eco-design approach carried out on the
most recent generation of GIS.
The second axis was to search for an SF6 alternative.
GE’s Grid Solutions, has developed an SF6-free solution
named g3 (green gas for grid), which is based on 3M
NovecTM 4710 and CO2. g3 reduces the global warming
potential by 98% compared to SF6 with only minor design
modifications from the equivalent SF6 design. The paper
presents the different steps of the development, from
the beginning in 2010 until the implementation in a first
product: the 420kV Gas-Insulated Line.
1. Introduction
In the IEO (International Energy Outlook) 2015
projections [1], energy demand grows by nearly one-third
between 2013 and 2040.
At the same time, the world is waking up to the reality of
climate change. The Paris Agreement [2], signed by 195
nations worldwide at the COP21 conference in December
2015, brings all nations into a common consensus on
their current and future responsibilities regarding the use
of carbon and addressing climate change. Together with
interoperability and flexibility, reducing the environmental
impact of our electrical transmissions systems will be
*elodie.laruelle@ge.com
KEYWORDS
Alternative Gases, Eco-design, g3,Gas Insulated Line, SF6
Cigre Science & Engineering • N°7 February 2017
102
one of the key challenges that we will have to face in the
coming years. A sustainable development approach must
be applied, by responding to “the needs of the present
without compromising the ability of future generations to
meet their own needs”, as defined in the BRUNDTLAND
report in 1987.
For more than ten years, GE’s Grid Solutions has embarked
on the path of eco-design – integrating the environment
into the design phase alongside other conventional design
concerns such as customers’ expectations, technical
performances, cost control, etc. Eco-design is a concrete
solution to sustainable development. Indeed, curative
approaches have already shown their limits paving the way
for an integrated, preventive approach. The eco-design
approach begins at the time of the definition of product
characteristics, because the design phase determines 80%
of future impacts and offers the greatest degree of liberty.
Through the deployment of its eco-design approach,
the objective of Grid Solutions is to reduce the global
environmental impact of the GIS, over its whole life cycle,
from the extraction of the raw materials until the end-of-
life, while considering all environmental indicators.
SF6 is, as a matter of course, one of these environmental
aspects, having a global warming potential about
23500 times greater than CO2. Life-Cycle Assessment
performed on a gas-insulated substation (GIS) shows
that SF6 emissions that can occur during the service life
of the switchgear can represent between 60% and 80%
of the total Global Warming impact of the substation
over its whole life cycle. For the 17 others indicators the
main environmental aspects are materials used and power
losses during the use phase.
 Fig. 1: Life Cycle Assessment results for T155 420kV
(impact of SF6 emissions during the service life appears in purple)
Having identified the importance of SF6 emissions, Grid
Solutions, through the Clean Grid program, has made a
commitment to reduce these emissions from its products.
This paper presents the two main axes of this program: the
reduction of SF6 impact from the latest generation of SF6
filled GIS, and the search for an SF6 alternative.
2. Reduction of SF6 impact from
the latest generation of GIS
The first axis is the optimization of existing technologies
by reduction of SF6 impact from the latest generation
of GIS. While GIS users have a greater responsibility
for reducing SF6 emissions during the use, maintenance
and end-of-life phases, GIS manufacturers have the
heavy responsibility of limiting the use and emissions
of SF6 during the design, as well as manufacturing and
installation phases.
This is why reducing SF6 impact has been directly
integrated in the eco-design approach developed in
the R&D department [3]-[4]. Indeed, two of the six
key environmental performance indicators (KEPI) are
dedicated to SF6 impact: SF6 mass and seal length (see
Fig.2). These KEPI have been created in order to set up
targets to reduce environmental impacts and assess our
progress, as defined in the “eco-design conscious process”
described in IEC 62430 [5].
Fig.2: Six key environmental performance indicators (KEPI) for GIS used in
eco-design approach
Cigre Science & Engineering • N°7 February 2017
103
This chapter presents the benefits of this approach when
applied to the latest generation of GIS regarding SF6
impact: sealing improvement, reduction of SF6 quantity,
reduction of seal lengths, optimization of assembly in
factory, introduction of gas monitoring.
2.1. Sealing improvement
The choice of the gasket and in particular its material
is fundamental to ensure sufficient tightness at each
interface, to ensure the reliability of the GIS during all
its lifetime and not to exceed the maximum leakage rate
of 0.5% per compartment per year as specified in IEC
62271-203 [6].
Early designs of Grid Solutions’ GIS in the 1960’s used
simple O-ring type gaskets sealed with grease. In the
late 1970’s this technology was improved with a second
sealing system added to improve tightness and limit
corrosion. This solution has proven to be efficient over
the long-term and has been implemented on many GIS
substations. It is still used in specific cases like the sealing
of accessories (filling valves, rupture disks etc.).
After more research and testing, an innovative type of seal
for flange-to-flange connection emerged in the 1980’s
which has a particular shape delivering “three-seals-in-
one”. Two outer lip seals protect the huge inner sealing
surface from atmospheric corrosion and from the effects
of SF6 decomposition products. The inner sealing surface
is well protected over the lifetime of the GIS.
Fig.3: Three-in-one gasket
The choice of gasket materials is also critical and must be
investigated in depth and validated via performance tests
representing the worst conditions of use of the product.
All causes of tightness losses have been thoroughly
investigated by Grid Solutions and the performances of
gaskets selected from third-party manufacturers have
passed severe design tests. The EPDM gaskets used on
Grid Solutions’ GIS today do not induce corrosion and
ensure long-term tightness [7].

2.2. Reduction of SF6 quantity

As seen above, SF6 mass is one of the six KEPI that Grid
Solutions uses during the eco-design process. To reduce
SF6 mass, the following strategies have been applied to
the latest GIS generations:
• Reduction of component volumes
• Use of modern 3D software to optimise GIS shape,
dimensions and switching performance
• Optimisation of filling pressure
As illustrated on Fig. 4, Grid Solutions has significantly
reduced the SF6 mass contained in the latest 252kV GIS
compared to the former one.
Fig.4: SF6 mass reduction for latest 252kV GIS
2.3. Reduction of seal length
Considering that the greater the length of the seals, the
greater the risk of leakage, reducing seal lengths is part
of the eco-design approach to limit SF6 impact. The
following strategies are applied:
• Reducing the diameter of enclosures
• Limiting the number of interfaces
• Machining openings only when necessary
Fig. 5 shows the reduction in seal lengths obtained in the new
252kV GIS generation thanks to the strategies listed above.
Fig.5: Seal length reduction for latest 252kV GIS
2.4. Optimisation of assembly in factory
The quality of the assembly of GIS components is critical
to ensure SF6 tightness. Traditionally a GIS bay was
packed in several shipping units and assembled on site.
Grid Solutions’ strategy is now to maximise the amount
of assembly carried out in our factories, the limitations
being mainly transport constraints, especially when
applied at high voltage levels like 245 kV and 420 kV.
All interfaces assembled in the factory are tested more
rigorously than is usually possible on site. This further
limits the potential risk of wrong assembly and leakages
on site.
An example is the latest 420kV GIS bay. The compactness
of the design allows the full bay to be assembled and
tested in factory and shipped to the customer’s site in
only one package. This means that all major components
are tested (including optional embedded LVCC) and
sealed at the factory in the cleaned environment of the
GIS workshop. This type of double busbar bay assembly
in the factory is used for the first time for the UK project
of Sellindge, which is the same installation where the
pilot application with g3 is implemented (see details in
next chapters).
Fig.6: Last 420kV GIS bay shipped in one single unit

Cigre Science & Engineering • N°7 February 2017

104
2.5. Introduction of gas monitoring
Another lever to reduce SF6 impact is to equip the GIS
equipment with an efficient control and monitoring
system which can detect leakages with high accuracy and
can also give trends. This helps the maintenance teams to
react quickly to any unexpected leakage.
Traditional gas densimeters activate an alarm when
the pressure level decreases below the first threshold
level. Usually there are two levels of alarm. In current
GIS, typical values are 5% and 10% below the nominal
pressure. With this traditional system, a leakage cannot be
detected before the first alarm activates.
New monitoring systems now allow leakage detection as
soon as the leakage starts and measure the leakage rate
well before reaching the first alarm level. Interventions
can be planned as soon as any leakage is detected and the
emission of SF6 into the atmosphere can be minimised.
Grid Solutions has developed the BWatch3 monitoring
system which monitors the overall GIS and particularly
the gas density with a very high sensitivity. Thanks to gas
sensors that measure gas temperature and pressure for
each compartment, it provides a permanent display of the
status of all compartments. By selecting a compartment,
a display gives its detailed SF6 gas values: gas density,
pressure and temperature are detailed.
BWatch3 is able to quickly identify a leak thanks to trend
calculations: the daily average density of each compartment
is calculated and an advanced alarm is activated if the
second low density threshold is calculated to be reached
within the coming 100 days. This allows operators to
manage SF6 emissions by planning interventions well
before the traditional densimeter alarm goes off.
Fig.7: Specification for SF6-free solution for high-voltage applications
Cigre Science & Engineering • N°7 February 2017
105
This system has been installed on all 420 kV GIS
substation delivered by Grid Solutions to National Grid
in UK since 2004.
3. Search for an SF6 alternative &
first pilot project
In the previous section, strategies performed at R&D
stage for reducing SF6 emissions have been presented.
They have allowed considerable improvement to the latest
generation of GIS. However, the scope to continue to
reduce the global warming impact of GIS within existing
design constraints is limited and the need for a disruptive
innovation was identified.
The second axis of Grid Solutions Clean Grid approach
was then to search for an SF6 alternative. This section
presents the different steps in the development of an SF6
alternative, from the beginning of the research in 2010
until the implementation in a first product: the 420kV
Gas-Insulated Line.
3.1. Search for an SF6 alternative
The objective was to find an SF6 substitute bringing
considerable environmental benefit. Target was to have a
global warming potential (GWP) reduction of more than
95% compared to SF6, and no ozone depletion potential.
In addition to these environmental targets, the SF6
alternative should also meet the very tough specification
for high-voltage switchgear, as well as health and safety
prerequisites, as shown on Fig. 7.
During four year of research, Grid Solutions, together
with the 3M Company, has investigated several families of
gases: fluoroketones, hydrofluoroolefins, fluorooxiranes,
and the promising one: fluoronitriles.
The specific fluoronitrile Heptafluoro-iso-butyronitrile,
(CF3)2CFCN, called NovecTM 4710, was selected from
amongst different candidates based on toxicity, dielectric
strength, GWP and boiling point [8].
Thus, the first interesting alternative to SF6 with similar
performances was found: a gas mixture consisting of CO2
and Heptafluoro-iso-butyronitrile, referred to by the name
g3 (green gas for grid). This new gas mixture reduces the
global warming potential by 98% compared to SF6 with
only minor design modifications from the typical SF6
design. Its GWP is 327 against 23500 for SF6.
3.2. Application in a 420 kV GIL
The announcement of g3 gas mixture as an alternative to
SF6 at the last CIGRE session in 2014 was a major step
in an effort to dramatically reduce the carbon footprint
of high voltage equipment. The next step was to make
products available using the new gas mixture. 420 kV GIL
has been chosen to be one of the first applications of this
g3 mixture, as it represents the most important quantities
of SF6 used at such voltage level. It should be noted that
over the years, Grid Solutions has installed more than 200
km of Gas-Insulated Lines (GIL). About 60% of these
references are at 420 kV voltage level.
3.2.1. Design of the 420 kV GIL insulated with g3
The T155 420 kV GIL using the g3 insulating media has
been designed for a minimum ambient temperature of
-25°C. To achieve this low temperature, the g3 mixture
chosen is 4% by volume of NovecTM 4710 and 96%
by volume of CO2. The filling pressure has then been
increased to 9.6 bar relative for g3 design in order to reach
the dielectric strength of SF6 used at 6.3 bar relative.
Some slights adaptations were needed due to the pressure
increase: adaptation of some enclosures (minimum
thickness is 7mm instead of 6mm for SF6), bursting disks
and density sensors. The insulator remains the same as it
was already design for higher design pressure.
Others adaptations are directly linked to the gas itself. The
g3 GIL is equipped with special type moisture absorbers
adapted to g3 requirements. The BWatch3 software is
adapted to be able to manage the 3 different gases (SF6,
CO2 and NovecTM 4710). A distinction between SF6 and g3
filling valves is made by using different thread and color,
Cigre Science & Engineering • N°7 February 2017
106
in order to avoid confusion with SF6 and any potential
wrong operation on-site.
A part of the development also included various tests
to check long-term behavior of the gas and material
compatibility. For instance, new sealing materials were
tested and validated for both g3 and SF6 gases. Tests
showed that the gasket size and groove design required
no modification and are compatible with existing gas-
insulated busbar designs.
3.2.2. Testing of the 420kV GIL insulated with g3
The new version of the g3 GIL has been fully type-tested
according to IEC standard 62271-203 [6]. The equipment
successfully passed the dielectric type-tests performed
on an arrangement with composite bushing and busbar
elements, i.e. lightning impulse, switching impulse and
power frequency test voltages, as well as temperature rise
tests to meet the same performance levels as GIS with
SF6 [9]. The following table sums up the 420kV GIL g3
performance, which is also the performance required for
the first pilot project in UK detailed further down:
Voltage 420 kV
Short-duration power frequency 650 kV
Switching impulse 1050 kV
Lightning impulse 1425 kV
Continuous current 4000 A
Short-time withstand current 63 kA
Peak withstand current 170 kAp
Table 1. 420kV GIL g34%vol performances
Other tests were carried out to assess the behavior of the
equipment under abnormal conditions such as PD behavior
in the presence of particles or internal fault test [9].
3.2.3. Environmental comparison of SF6 and g3 GIL
products
Life Cycle Assessment (LCA) has been carried out to
compare SF6 and g3 technology [10]. A comparative LCA
compares the global environmental impact of two products
over its whole life cycle (from extraction of raw material
to the end-of-life) and on all environmental indicators.
The target is to check if an environmental improvement on
one indicator (for example global warming) induces any
pollution transfer, i.e. an increase of the environmental
impact on others environmental indicators.
 Fig.8: Sellindge substation location

This LCA shows that the use of g3 in the 420kV GIL allows
to considerably reduce environmental impact compared
to the SF6 technology:
• on Global Warming, with a huge reduction of 96% part of National Grid’s transmission network and is an
• on all other environmental indicators (ozone important network reinforcement project. It is located in
depletion, acidification, metal depletion..), with an
average reduction of 14% on all indicators.
3.3. Focus on the first pilot project in UK
420 kV GIL will be the first g3 product to be energized.
Thus, this chapter focusses on the first pilot project that
is delivered and installed during first semester 2016 to
National Grid utility in the UK.
National Grid has a strong commitment to managing all
of its environmental impact and actively promotes and
supports initiatives, such as g3, to identify ways to reduce
SF6 usage across its network. Furthermore, under the RIIO
(Regulation = Incentive + Innovation + Outputs) regime
in the UK continuous reduction of SF6 losses is financially
incentivised. Taken together these drive a strong interest
in being at the forefront of adopting commercially viable
methods of reducing or eliminating SF6.
Thus, National Grid wanted to undertake a pilot project
with the 420 kV GIL for its new Sellindge substation in
Kent. The new Sellindge substation will be an integral
the South of England adjacent to the existing 2,000 MW
HVDC interconnector which links the British and French
electricity transmission systems.
The new substation consists of five feeders (four overhead
lines and one cable feeder to the interconnector). Two of
these feeders, including 300 metres of gas-insulated busbar
on which g3 will be piloted, are separable from the rest of
the substation using a bus section. In the unlikely event
of a problem with the pilot installation this arrangement
provides a level of risk mitigation which is important when
introducing new technologies onto operational transmission
systems. The fact that the g3 hardware can be relatively
easily converted back to operation with SF6 also contributes
to the overall risk assessment undertaken by National Grid
prior to embarking upon the live pilot at Sellindge.

Table 2. Environmental benefit of the g3 technology applied on Sellindge pilot project

Cigre Science & Engineering • N°7 February 2017

107

Fig.9: Sellindge g3 GIL pilot project
The first column of Table 2 gives the amount of gas
contained in the product, in equivalent carbon. By
choosing the g3 option on the two GIL feeders described
above, 2500kg of SF6 are saved on the newly installed
assets, equivalent to 58750 tons of CO2 equivalent.
The second column of the table gives the gas emissions
that would occur during the operation period, considering
a maximum leakage rate of 0.5%/year during 40 years.
Thanks to the environmental benefits of this new g3
technology an important reduction of greenhouse gas
emissions has been achieved on this pilot project: more
than 11000 tons of CO2 equivalents will be saved during
the service life of the product.
4. Conclusion
SF6 technology benefits of practical experience gained
over many years. Thanks to design improvements brought
by the manufacturers and best practices applied during
the operation period, SF6 emissions are now minimised.
Nevertheless, each new SF6 substation increases further
the utilities’ installed mass of SF6, particularly when
installing assets such as 420 kV GIL. SF6 is targeted by
the Kyoto Protocol and we have seen in recent years an
increase of the regulatory pressure upon its use. This
pressure will increase over the next few years, considering
Cigre Science & Engineering • N°7 February 2017
108
the focus that is now made on climate change and may
even lead to taxes on SF6 use. The availability of this
new green gas alternative signals a major revolution in
high-voltage technology and introduces the potential for
a tremendous reduction in the tons of CO2 equivalents
deployed in this technology worldwide.
These new g3 products represent a major advance in
replacing SF6 as an insulation medium in high voltage
equipment. The next step is to replace SF6 in switching
equipment. The testing required for this is well under
way.
5. Bibliography
[1] “World Energy Outlook 2015”, International Energy Agency (IEA),
2015, http://www.worldenergyoutlook.org
[2] “Adoption of the Paris agreement”, UNFCCC, Conference of the
Parties, Twenty-first session, Paris, December 2015, http://unfccc.
int/resource/docs/2015/cop21/eng/l09r01.pdf
[3] E. Laruelle, A. Ficheux and all, "Benefits of eco-design practices to
latest generation of GIS/GIL”, MATPOST 2011, Lyon, 2011
[4] Y. Kieffel, A. Spinosa, E. Laruelle, K. Stumpf, “Life cycle assessment
and end-of-life management of GIS and AIS”, CIGRE Paper C3-
203, Paris, 2014
[5] IEC 62430 “Environmentally Conscious Design for Electrical and
Electronic Products”, 2009
[6] IEC 62271-203, “High-voltage switchgear and controlgear – Part
203: Gas-insulated metal-enclosed switchgear for rated voltages
above 52 kV”, International Standard, Edition 2.0, IEC, 2011
[7] A. Ficheux, D. Depres, E. Laruelle, Y. Kieffel, P. Prieur, “Limiting
SF6 Gas Emissions by Optimization of Design and Testing of Gaskets
in High Voltage Gas-Insulated Substations”, CIGRE Paper C3-209,
Paris, 2012
[8] Y. Kieffel et. al., “SF6 alternative development
for high voltage switchgears”, CIGRE Paper
D1-305, Paris, 2014
[9] D. Gautschi, A. Ficheux, M. Walter, J. Vuachet, “Application of a
fluoronitrile gas in GIS and GIL as an environmental friendly
alternative to SF6”, CIGRE Paper B3, Paris, 2016
[10] Y. Kieffel, E. Laruelle, G. Gaudart, F. Biquez et al., “g3 – The
alternative to SF6 for HV equipment”, MATPOST 2015, Lyon, 2015
Amplification of harmonic background
distortion in wind power plants with long
high voltage connections
C. F. Jensen*, Energinet.dk, Denmark
Ł. H. Kocewiak, Dong Energy, Denmark
Z. Emin, PSC UK, United Kingdom
Assessment of harmonic disturbance in offshore wind power
plants (WPPs) is becoming an increasingly important task
as the WPPs are increasing in size. It has been customary
in the past to base all compliance and design studies on
positive sequence simulation models. However, the use
of long high voltage (HV) cable connection systems
gives rise to the need for more sophisticated modelling.
It is justified that in case of unsymmetrical cable systems,
such as flat formation single phase HV cable systems a
decoupled sequence model can lead to underestimation of
the harmonic distortion in the system.
The electrical connection to the 400 MW Anholt offshore
WPP is used as a study case example. Synchronized
harmonic voltage measurements were taken at the Point
of Common Coupling (PCC) for Anholt (400 kV) and
at the Point of Connection (PoC) (33 kV). Based on the
measurements, corresponding per phase harmonic voltage
gains are determined. Furthermore, the harmonic voltage
sequence components are determined at the PCC of Anholt.
It is shown that at and near resonances the harmonics do
not follow their natural sequence order, but the voltages
contain some portion of all three components (i.e. positive,
negative and zero sequence components).
1. Introduction
An AC cable system comprising of submarine and land
cable sections is the most common way of integrating a
large offshore WPP to the onshore transmission network.
The length of the combined cable system can, in some
cases, be very long – in Denmark for instance the
longest WPP cable connection is 84 km. Furthermore,
the array cable system interconnecting the individual
wind turbines (WTs) to a collection point can be over
100 km long making the total length of the cable system
close to 200 km. The total length of HV cables for future
* cfj@energinet.dk
KEYWORDS
Cable crossbonding, Cable modelling, Electromagnetic compatibility, Harmonics, Harmonic amplification, Harmonic
propagation, Inter-Sequence coupling, Phase domain,
Cigre Science & Engineering • N°7 February 2017
109
offshore projects is expected to be even longer, perhaps
beyond 150 km. Experience shows that operation of
such a long cable system can have a significant impact
on power quality in terms of harmonic background
amplification due to resonances caused by the offshore
WPP electrical infrastructure as well as the onshore
and offshore transmission system. During low power
production or in some cases under normal production,
the background harmonic distortion in the onshore grid
can potentially be amplified towards the offshore WPP
leading to potential design or compliance issues both at
the PoC and internally in the park array cable system.
Such harmonics amplification at different locations
within the WPP and transmission system electrical
infrastructure can possibly bring concerns in relation to
estimated harmonic voltage distortion levels or limits
specified in grid codes.
The importance of the phenomenon is gaining momentum
due to the growing number of large-scale WPP connection
into the transmission grid. In Denmark, the HV export
system is owned and operated by the transmission system
operator (TSO) and therefore harmonic compliance is
evaluated and specified at the PoC designated as on the
secondary side of the park transformers at the offshore
platform. Hence it is the responsibility of Energinet.dk
as the Danish TSO to ensure power quality at the PoC is
compliant before the connection of the WPP. However, due
to the combined length of the array cables and other passive
components within the WPP electrical infrastructure,
amplification can increase due to the shift in resonance
conditions or introduction of new resonances when the
WPP is switched in. Therefore it is equally important
for both the TSO and the WPP developer to understand
and limit the cause of the background harmonic voltage
amplification due to resonances within the electrical
infrastructure. Due to the availability and access to
measurements data, Anholt WPP was utilised to illustrate
possible difficulties that may be encountered and hence
should be taken into account during harmonic assessments
where the grid connection system is unsymmetrical (i.e.
with unbalanced impedance) in nature and wherefore
simplified positive sequence analysis may be inadequate.
2. System description
The electrical infrastructure of the Anholt WPP is shown
in Figure 1.
Figure 1 The electrical connection of the 400 MW Anholt WPP to the
Danish transmission system.
The export system consists of two 400 MVA transformers
and three 220 kV cable parts; a 59.6 km cross-bonded
land cable laid in flat formation, a 0.5 km beach cable and
a 24.5 km three-core submarine cable. A reactor station
(GNK220) is placed at the transition point between the
land and beach cable. The offshore platform includes
three 140 MVA, 220/33 kV transformers. In the medium
voltage (MV) array cable system, 150 mm2, 240 mm2
and 500mm2 copper cables are used to connect the wind
turbines (WTs) to the offshore substation. The layout of the
array cable system is non-uniform leading to differences
in the total length of the MV cables connected to each
PoC as presented in Figure 1 [1].
The 33 kV array cable system has a combined length of
152 km and the details are presented in Table I.
Array cable
Connected to WTs Arrays
length [km]
PoC#1 37 4 50
PoC#2 37 4 48
PoC#3 37 4 54
Table I - Anholt array grid details
Because the array cable system at each PoC is different in
length and structure, the effect of harmonic amplification
can also be different at the three PoCs and hence they
must be evaluated individually.
2.1. Power quality measuring and analysis system
Power quality measurements are recorded at Trige 400 kV
substation which is the PCC of the Anholt WPP.
Traditional HV and extra high voltage (EHV) instrument
transformers are not capable of correctly representing
all harmonic frequencies in their secondary circuits due
to the internal resonance in the measuring systems; this
is certainly the case for capacitive voltage transformers
and for the electromagnetic wound voltage transformers
beyond a certain frequency [2]. Because of this, correct
evaluation of the harmonic content of HV and EHV
signals can be problematic. Hence, in recent years
specially designed transducers have been developed
which can be incorporated into traditional capacitive
voltage transformers providing a bandwidth of the
measuring unit well beyond the 50th harmonic [3]. The
wound voltage transformer (VT) at the PoC can reflect the
harmonic content in the high voltage signal up until some
cut-off frequency. This frequency is typically around 800-
1000 Hz (around about 20th harmonic order) at 33 kV. In
this study only measured harmonics until the 13th order
are discussed; hence the 33 kV VTs can be used directly.
Time domain voltages are recorded at the PCC and PoC
at a sampling frequency of 20 kHz. The harmonic phase-
to-ground voltage magnitudes and their phase angles are
determined using the Discrete Fourier Transform (DFT)
at each location.
3. Transmission level harmonic
voltage background distortion
For transmission level simulation studies and system
design purposes, it is a common approach to assume that
either all harmonic voltages are purely positive sequence
Capacitance of
array grid [µF]
or that all non-triplen harmonics are positive sequence
and triplen harmonics are zero-sequence. However, field
14
measurements indicate that especially on and near cable
13
systems in unsymmetrical formations (as in case with the
15
flat formation single core HV cables used for the land cable
part of the grid connection system for Anholt WPP), the
Cigre Science & Engineering • N°7 February 2017
110
 Figure 2 Harmonic magnitudes and phase angles obtained at Trige 400 kV.

harmonic voltages can be v unbalanced. An unbalanced
harmonic order will contain a portion of all tree sequence
components rendering the former assumption invalid.
A typical Danish transmission level harmonic spectrum
from the 2nd to the 20th harmonic taken at Trige 400 kV is
shown in Figure 2. The signals used for the analysis are
measured using the specially developed high bandwidth
high voltage sensors described in Section 2.1
Figure 2 shows strong unbalanced behaviour for some of
the harmonics and more balanced behaviour at others. For
instance, the magnitude of the 11th order harmonic voltage
on phase B is almost twice as large as the magnitude of
phase C. Also, the 3rd order harmonic, often assumed to be
a pure zero-sequence harmonic, shows strong unbalanced
behaviour indicating a diverse content of the three
sequences.
With both the harmonic magnitudes and phase angles
correctly determined for each phase, the sequence
components of the harmonic voltages can be determined
using Fortescue transform. This is done for the harmonic
voltages shown in Figure 2 and the results are presented
in Figure 3.
The sequence decomposition of the harmonics shows that
there is a tendency that most of the harmonics follow their
natural sequences under the balanced system assumption.
For instance the 2nd order harmonic is predominantly a
negative sequence harmonic, the 4th is predominantly a
positive sequence harmonic, the 5th is predominantly a
negative sequence harmonic and so on. However, the 11th
order harmonic, that is expected to be negative sequence,
contains a non-negligible positive sequence component.
The 3rd order harmonic contains all three sequence

Figure 3 Harmonic voltages in sequence component form at Trige 400 kV.

Cigre Science & Engineering • N°7 February 2017

111
 Figure 4 (a) Phase A, B, C and (b) positive, negative and zero-sequence harmonic voltage at the 220 kV
offshore substation for an applied positive sequence harmonic voltage at Trige 400 kV
components with a slightly higher content of positive
sequence. Hence, it is safe to conclude that the assumption
that individual harmonic orders contain only one unique
sequence component will lead to an underestimation of
the highest phase-to-ground harmonic voltage which can
be an issue for planning and design purposes. The reason
for the unbalanced behaviour at some harmonic orders is
analysed in Section 4.
4. Phase domain harmonic
modelling
Harmonic propagation studies are most often conducted
before system components are selected because changes
to the system are either not an option or very expensive
after the installation process begins. Hence, the designer
relies on trustworthy simulation results for design- and
compliance studies. Such harmonic design studies are
often carried out in the decoupled sequence domain where
positive sequence and sometimes zero-sequence data is
used to build a simplified harmonic model of the system
under study. Such modelling approach can be inaccurate
on a system with a significant amount of cables laid in
flat formation due to the inherent unbalance introduced
naturally as indicated by the measurements in Section
3. Furthermore the detailed phase-domain modelling
requires appropriate representation of harmonic sources
(e.g. harmonic background distortion measured at Trige
400kV) including information about phase amplitudes
and angles.
A phase-domain model is necessary to predict the correct
behaviour of the harmonic voltages in an unbalanced
system like the Anholt WPP. The land cable part of the
electrical connection is comprised of three single-core
cables cross-bonded and laid in flat formation. Such a flat
formation cable system introduces strong asymmetry at
Cigre Science & Engineering • N°7 February 2017
112
several harmonic frequencies. Even at power frequency
the impedance of the middle laid phase can be 12-15%
lower than the two outer phases [4]. Near resonance, the
phase impedance of one cable core conductor can be
significantly lower or higher than the two others, causing
unsymmetrical harmonic currents to flow in the system
even under balanced harmonic injection. As a consequence
of the unbalanced series-impedance matrix of the flat
formation cable, pure positive sequence injection can,
besides energizing the positive sequence system, also
energize the negative sequence system and the zero
sequence system to some degree, with unsymmetrical
phase voltages as a consequence.
A full detail frequency dependent phase-domain model of
the Anholt grid-connection cable system and array cable
system is constructed in DIgSILENT Power Factory. All
cables are modelled based on a geometrical description
and all cross-bondings are manually implemented. The
frequency dependence of the transformers and reactors is
included using appropriate multiplication factors [5]. The
WTs are modelled using a Thévenin equivalent including
turbine transformers and WT internal filters.
To illustrate the importance of phase domain modelling,
an ideal positive sequence harmonic voltage is applied
behind the system impedance at Trige 400 kV at each
integer frequency from the 2nd to the 20th harmonic
and the three phase-to-ground voltages and the positive,
negative and zero-sequence voltages at the 220 kV
offshore substation are calculated as shown in Figure 4.
The resulting phase impedance as seen from Trige 400
kV (Zp(h)=Up(h)/Ip(h)) where p is equal to A, B or C ) is
presented in Figure 5.
By comparing Figure 4 and Figure 5, it can be clearly
seen that the application of positive sequence voltage
only predicts well the harmonic phase-to-ground voltage
 Figure 5 Phase A, B and C impedances seen from Trige 400 kV
at frequencies far from resonance. This is at conditions
where the phase impedances of the three phases are
balanced as seen from the location of harmonic injection.
At resonance and near resonance (i.e. harmonic range
from the 3th to 4th and between the 9th and 14th in Figure 5),
the three phase impedances differ significantly and hence
inter-sequence coupling will occur. Notice for instance the
negative sequence voltage at the 12th harmonic in Figure
4 is larger than the positive sequence voltage – this is
under positive sequence injection only into the system as
explained. Notice also that at the 13th harmonic, the phase
B voltage is more than double in magnitude compared to
the magnitude of phase A voltage. In sequence domain
this manifests itself as a much lower (almost half) positive
sequence voltage implying that assessment based on
positive sequence will lead to underestimation of the
harmonic voltage. In a decoupled sequence networks
based model this phenomena would be missed altogether.
Therefore, using only the positive sequence component
for compliance analyses at planning/design stage can be
problematic in unbalanced systems.
5. Phase domain harmonic
voltage gain
During harmonic compliance studies it is common
practice to calculate harmonic voltage gains to establish
effect from one node to another. In the decoupled
sequence domain, the positive sequence harmonic
voltage gain from Bus X to Bus Y at the hth harmonic
can be determined as:
where U1X(h) and U1Y(h) are the positive sequence
harmonic voltage at the hth harmonic at Bus X and Y
respectively. The negative and zero sequence harmonic
voltage gains are defined in the same manner but not
often used in practical studies.
Cigre Science & Engineering • N°7 February 2017
113
The harmonic voltage gains of the three sequence
components are well defined due to the definitions
governing the decoupled sequence domain. There is a
linear relationship between the sequence components of
the voltage at the two busses; increasing the harmonic
distortion at Bus A by some factor will lead to an increase
at Bus B by the same factor in a one-source system. This
is one of the main benefits of the decoupled sequence
domain when it comes to harmonic propagation studies.
The same cannot be said in the phase domain. The per
phase harmonic voltage gain would naturally be defined
as:
where UpX(h) and UpY(h) are the harmonic voltage of
phase p (A, B, or C) at the hth harmonic at Bus X and Y.
Many TSOs use a decoupled sequence model for
harmonic studies. However, harmonic emission limits
are issued and measured in relation to the highest phase-
to-ground value. Because mutual coupling between
the three phases differs in an unsymmetrical conductor
configuration (such as that introduced by a cable
system laid in flat formation), the decoupled sequence
assumption is violated. In the decoupled sequence
domain, the three sequences are independent of each
other. However, the harmonic phase voltage gain depends
on the unique flows in the other system conductors. This
makes the calculation of the phase harmonic voltage
gain complicated in an unbalanced system.
To illustrate the error introduced by using per default
the decoupled sequence domain and hence positive
sequence gains, the harmonic phase and positive
sequence voltage gains from Trige 400 kV to the reactor
station GNK220 and the gains from the reactor station
to the off-shore platform are determined (see Figure 1
for system diagram and bus references). The results are
shown in Figure 6.
 Figure 6 Harmonic voltage gains from (a) TRI400 to GNK220 and (b) GNK 220 to AHA220.

Figure 6(a) shows that the per phase and the positive
sequence harmonic voltage gains from TRI400 to GNK
220 (asymmetrical flat cable formation) strongly differ in
magnitude. This is more pronounced at and near resonance
frequencies.
The displayed phase and positive sequence gains in
Figure 6(b) confirm that on symmetrical conductor
configurations (such as the submarine cable laid for the
Anholt WPP), the positive sequence gain can be used
without loss of information. Notice the small deviations
for instance at the 19th and 20th harmonic in Figure 6(b).
The deviations are caused by the 0.5 km flat formation
beach cable used between the reactor station and the
beach cable/submarine cable joint.
To emphasise the point at an individual harmonic level, the
13th order harmonic voltage phase gains from Trige 400
kV to Anholt 33 kV are calculated based on measurements
over two days. The 13th order harmonic voltage as function
of time (10 min avg. values per IEC 61000-4-30) are shown
in Figure 7(a) and the resulting harmonic voltage gains
calculated using Eq. 2 in Figure 7(b).
Figure 7(b) shows that the measured 13th order harmonic
voltage gains differ by a more than a factor of 2 between
the three phases. Unfortunately, due to the inherit
limitation of the MV IVTs, the phase voltage angles cannot
be determined at Anholt 33 kV and hence the positive
sequence gain cannot be calculated and compared to the
phase gains. However, with the displayed differences
between the lowest and highest phase gains, the positive
sequence gain will not correctly represent all three phases
as discussed in the previous section.

Figure 7 (a) Trige 400 kV and Anholt 33 kV harmonic 13th order voltages and (b)
corresponding harmonic voltage gains.

Cigre Science & Engineering • N°7 February 2017

114
 Figure 8 Phase A harmonic voltage gains from Trige 400 kV to Anholt 33 kV on a (a) cable system in touching
trefoil (symmetrical), and (b) a cable system in flat formation (unsymmetrical) as the negative sequence voltage
is varied from 0 pu to 1 pu.

The positive sequence harmonic voltage gain in the
decoupled sequence domain is per definition not dependent
on the negative or zero-sequence component and vice
versa. However, the harmonic voltage phase gain can be
strongly dependent on which sequence components are
energized if the system is unbalanced. This is illustrated in
Figure 8 where two cable systems are energized by a 1 pu
balanced harmonic voltage at all harmonics from the 2nd to
the 20th. Gradually, a harmonic phase voltage unbalance is
introduced by increasing the negative sequence content of
the voltage from 0 pu to 1 pu while the harmonic positive
sequence voltage component is kept constant. One cable
system is laid in touching trefoil formation (symmetrical)
and the other in a flat formation (unsymmetrical). The
harmonic voltage gain from Trige 400 kV to Anholt 33
kV for phase A is then plotted for both cases.
Inter-sequence coupling is absent on the cable system in
touching trefoil system as the harmonic phase impedance
matrix is balanced (in contrast to flat-formation system)
and as a result, the harmonic sequence impedance matrix
is dominated by the diagonal elements (off-diagonal
elements are close to zero). Therefore, the per phase
harmonic voltage gain is not dependent on the magnitude
of negative sequence harmonic voltage applied. For the
unsymmetrical cable system, the per phase harmonic
voltage gain is strongly dependent on the magnitude of
the negative sequence voltage applied (or present) at and
near resonance due to inter-sequence coupling as seen in
Figure 8(b). Additionally, there will be dependence on the
angle of the sequence components, complicating the issue
further, but this is not covered here.
Phase-domain modelling becomes increasingly relevant
as more and more cables are installed at transmission level
and the degree of (harmonic) unbalance thereby increases.
The MV array cables and most often the HV submarine
cables are symmetrical due to construction characteristics
and hence the phenomenon of inter-sequence coupling is
less important. However, even with a relative short land
cable section laid in an unsymmetrical formation in the
system, inter-sequence coupling should not be ignored.
6. Conclusions
the paper discussed the modelling approach for harmonic
propagation and assessment studies on systems consisting
of long EHV cable systems. The 400 MW Anholt offshore
WPP is used as a case study example.
Decomposition of high accuracy measurements taken at
the PCC of Anholt WPP into sequence components shows
that individual harmonic orders in general follow their
natural sequences (that is negative for the 2nd, zero for
the 3rd, positive for the 4th and so on). However, this is
not the case at and close to resonances as each individual
harmonic will contain some portion of the two other
sequence components. This is expected to be the case in
almost all systems due to the inherent unbalanced nature
of the power networks. However, the situation may be
exacerbated in certain situations and especially when
there is a high degree of natural unbalance introduced
around the assessment location. The analysis has shown
that when modelling highly unsymmetrical cable systems
as is the case with EHV and HV cable systems laid in
flat formation, determination of voltage gains to a remote
node or amplification of background distortion at the local
node can vary depending on the complexity of modelling
assumed.

Cigre Science & Engineering • N°7 February 2017

115
The concept of per phase harmonic voltage gain is
introduced and synchronized measurements at the PCC
and PoC of Anholt WPP are used to determine the per
phase harmonic voltage gain for the 13th order harmonic.
The sequence harmonic voltages are compared to the
phase-to-ground harmonic voltages. It is shown that
the highest harmonic phase-to-ground voltage can be
significantly higher than the positive sequence voltage,
and hence underestimation of the harmonic content
can occur when considering only decoupled sequence
modelling. Although not the case in the utilised case study,
the shortcomings of using an oversimplified assumption
(use of models where inter-sequence coupling is ignored)
can be two fold; likely distortion may be underestimated
and hence may bring uncertainties in the design with
possible non-compliance issues at a stage where the cost
may be severe to the WPP developer or TSO.
7. References
[1] . H. Kocewiak, B. L. ÿhlenschlÊger Kramer, O. Holmstr¯m, K. H.
Jensen og L. Shuai, ªActive Filtering Application in Large O shore
Wind Farms´, International Workshop on Large-Scale Integration of
Wind Power into Power Systems as well as Transmission Networks
for O shore Wind Farms, Berlin, 2014.
[2] CigrÈ Technical Brochure 596, ªGuideling for Power Quality
Monitoring - Measurement Locations, Processing and Presentation
of Data,´ CIGR…/CIRED C4.112, 2014
[3] F. Ghassemi, P. Gale, T. Cumming og C. Coutts, ªHarmonic Voltage
Measurements Using CVTs´, IEEE Transaction on Power Delivery
Vol. 20 No. 1, 2005.
[4] C. F. Jensen, ªOnline Location Of Faults on AC Cable in Underground
Transmission Systems´, Springer, 2014.
[5] CigrÈ WG CC-02, ªGuide for assesing the network harmonic
impedance´, Electra 167, July 1996.
Cigre Science & Engineering • N°7 February 2017
116
8. Bibliographies
Dr Christian Flytkjær Jensen is an Electrical Engineer
with a Ph.D. in power system engineering. He possesses
expert knowledge in all aspects of power quality- and
electromagnetic transient analysis. He holds the position
of team lead and senior engineer in the Grid Analysis
group at Energinet.dk where he leads a team of highly
specialized power system engineers. He leads and set
company policy in all power quality aspects, develops
national standards as well as undertaking the role as
specialist engineer for solution development of tasks
within the areas of power quality and electromagnetic
analysis. Christian is involved in several Cigré working
and is the author of a number of peer revered international
papers (Cigré, IEEE etc.)
Dr Zia Emin is a power systems engineer with many
years of experience in power quality and switching
studies. He is the Technical Director on power system
planning and analysis with PSC, UK. Prior to this he
was with Parsons Brinckerhoff and National Grid where
he worked on switching transients and more extensively
on power quality issues. He has extensive knowledge in
power system modelling and substantial experience in
harmonic performance specification for HVDC converter
stations, renewable generation connections and the
connection of traction supply points. He is a Fellow of the
IET, a Senior Member of IEEE, a Distinguished Member
of CIGRE and a Chartered Engineer in UK. He is member
of GEL/210/12 Committee at the BSI and of WG8 in IEC
SC77A. He is currently the Chairman of CIGRE Study
Committee C4 System Technical Performance and is the
recipient of CIGRE Technical Committee Award in 2013.
Capacity mechanisms:
Results from a world wide survey

Hanspeter Höschle*† and Gerard Doorman‡

on behalf of the CIGRE working group C5.17
“Capacity Mechanisms: Needs, Solutions and State of Affairs”
†EnergyVille (KU Leuven & Vito), Belgium
‡Stattnet & NTNU, Norway

Abstract: The purpose of the survey is to get an up-to-date

Capacity mechanisms are discussed and implemented
in markets worldwide. Different types of capacity
mechanisms exist and the implementations vary with
the available generation technologies, demand patterns,
market design, external factors like weather conditions
or interconnectors with other market areas. In this
paper, we provide an up-to-date overview of capacity
mechanisms all over the world. We try to link system
characteristics with real-life implementations. For this
purpose, a survey was conducted within the CIGRE
working group on “Capacity Mechanisms: Needs,
Solutions and State of Affairs”. This paper summarizes
the contributions from 31 jurisdictions and outlines
the findings along system reliability, power markets
& expected profitability, impacts of renewables and
impacts of interconnectors. The answers indicate that
the large variety of individual designs and combinations
of capacity mechanisms is based on country-specific
characteristics. General power system characteristics
that distinguish between the redundancy or the need for
a capacity mechanism cannot be identified.
Summary
Capacity mechanisms are defined as an arrangement that
is part of the market design, meant to provide revenues
to capacity providers (generation and demand resources)
in addition to the revenues from selling energy in the
day-ahead, intraday and possibly other physical markets.
As part of its work, the CIGRE working group C5.17,
“Capacity Mechanisms: Needs, Solutions and State of
Affairs” [1], conducted a survey.
overview of capacity mechanisms all over the world. This
includes the status and description of currently installed
mechanisms as well as planned implementations in the
future. A set of contributions is received from 31 countries
respectively regions or operating zones (grouped under
jurisdictions) from all over the world. Major covered
topics of the survey are characteristics of the power
system and the power market, possible concerns about
system reliability presently and in the near future, and the
inclusion of a capacity mechanism in the present market
design, as well as the possible future implementations
of such mechanisms. In the present paper, we report the
results from this survey, which provide an insight in the
use and perceived future needs of capacity mechanisms
worldwide. The paper presents an excerpt of the full
evaluation which is part of the technical brochure [1]
covering the work of the CIGRE working group C5.17.
Thanks to the 31 received contributions, the survey
gives a good sample of currently existing and developed
capacity mechanisms around the world. The power
systems surveyed present a full range of characteristics.
These systems have a large spread in the size of the
systems, ranging from a few 1000 MW to more than
160 GW annual peak load in countries with different
fuel supply mixes – ranging from single-fuel to multi-
fuel (coal, nuclear, hydro), different market structures
and consumption patterns ranging from peak capacity-
constrained to energy-constrained systems. By capturing
these characteristics, the survey can empirically check
the existence of a link with an implementation of a
capacity mechanism.

*hanspeter.hoschle@energyville.be

KEYWORDS
Capacity mechanism, Market adequacy, State of affairs, Survey results,
System adequacy, CIGRE working group C5.17

Cigre Science & Engineering • N°7 February 2017

117
The evaluation shows that the integration of capacity
mechanisms in the market design is independent from the
market / system size. Capacity mechanisms are identified
in both very large (PJM) and very small (Ireland) systems.
Capacity mechanisms occur more often in system
with large seasonal differences and where respondents
indicate doubts about the generation adequacy in the
long run. Countries that use capacity mechanisms
reside on all continents. Three of the countries that do
not have a capacity mechanism today, have decided to
introduce one or are currently implementing one. Two
other countries have serious discussions about capacity
mechanisms. An additional four countries are changing
their designs. This shows that market design is dynamic
and continuously evolving.
In general, the large variety and combination of capacity
mechanisms in place indicates that country-specific
characteristics lead to individual designs. Based on the
survey however, it was not possible to identify objective
power system characteristics that distinguish between
the need or not for a capacity mechanism.
1. Introduction
The underlying work of this paper was conducted in
the framework of the CIGRE working group C5.17
“Capacity Mechanisms: Needs, Solutions and State
of Affairs” [1]. The scope of the working group is to
draft the need for capacity mechanisms, describe and
compare relevant market models, describe present
solutions in various countries and to document current
plans and developments. As part of the work, a survey
is conducted to assess the current state of affairs.
The survey covers countries respectively regions or
operating zones (referred to as jurisdictions) from all
over the world. The purpose of the survey is to get
an up-to-date overview of capacity mechanisms. This
includes the status and description of currently installed
mechanisms as well as planned implementations in
the future. Moreover, the presence of relationships
between the implementation and power system and
market characteristics is assessed.
The objective of the survey is threefold.
• An overview of the current implementation is given. and available interconnection capacity. The last section
The results reveal the types of mechanisms that are at
the time of the survey in place and also indicate the
Cigre Science & Engineering • N°7 February 2017
118
market framework in which the capacity mechanism
is embedded.
• A detailed lookout on ongoing discussions and
planned implementation indicates the development
of capacity mechanism in market frameworks
worldwide. It shows that market designs are dynamic
and continuously evolving.
• An analysis is done that tries to assess if certain
observable characteristics exist that trigger
the discussion and implementation of capacity
mechanisms in a given market area. The points
of interests are characteristics of a power system,
expectations about system reliability, renewable
energy sources (RES) shares and interconnection
capacity of a jurisdiction.
As approach to reach the targets of the survey,
questionnaires were sent out in 2014 to all jurisdictions
around the world through the CIGRE network. A total
of 31 jurisdictions responded providing the status of
current implementations, given the available data basis
of 2012 / 2013, and an outlook up to 6 to 10 years
ahead. The answers have been evaluated and assessed
with a coherent methodology. Statistical overviews,
graphical representations and discussions of the findings
are reported in the technical brochure of the working
group. The purpose of this paper is to give a condensed
summary of the survey results. The full discussion of
the survey results can be found in technical brochure of
the CIGRE working group [1]. This paper highlights the
most significant findings of the evaluation. Next to the
state of implementation or ongoing discussions in the
individual jurisdictions, the focus is on the relationship
of capacity mechanisms with the power system size,
expectations about system reliability, RES shares, and
available interconnection capacity.
The paper is organized as follows. Section 2 outlines
the data baseline and gives an overview on the power
systems’ characteristics. Section 3 lists an overview on
the capacity mechanisms worldwide, both present and
planned. Sections 4, 5 and 6 describe the findings on
the link of a capacity mechanism implementation with
system reliability, embedded power markets, RES shares
concludes the work of the survey and summarizes the
presented results in the paper.
 Figure 1: Map showing origin of submissions

2. Represented power systems The data baseline is a set of contributions received

and data baseline
The purpose of this survey is to get an up-to-date
overview of capacity mechanisms implemented all
over the world. This includes the status and description
of currently implemented mechanisms, as well as
planned implementations in the near future. With
the help of data about the individual power systems
triggers can be identified that lead to the discussion
towards the implementation of certain mechanisms.
To further facilitate this discussion, a questionnaire is
used to gather information of multiple power systems.
Besides general country information, the questions are
structured by means of five categories, namely power
system characteristics, power market characteristics,
system reliability and capacity mechanisms (current and
planned).
from 31 countries respectively regions or operating
zones (jurisdictions) from all over the world. A list of
submissions can be found in the appendix. Grouped
by continents, the contributions originate from North
America (5), South America (3), Europe (15), Asia
(5), Africa (1) and Australia (2). In terms of embedded
market systems, the contributions originate from both
regulated (6) and liberalized (25) power market systems.
With respect to capacity mechanisms, the jurisdictions
represent both systems with present capacity mechanisms
(16) and without capacity mechanisms (15). In 8 of
the jurisdictions, implementations or redesigns of a
capacity mechanism are currently taking place1. In what
follows, a more detailed evaluation of the power system
characteristics is done.
Figure 2 gives a visual overview of the jurisdictions
in terms of annual peak demand (in GW) and total

Figure 2: System size of contributions (based on data from 2012/2013)(Abbreviation see Table III)

_______________________
1- This information was collected in 2014 with respect to capacity mechanisms in place (CM (present)) as well as planned or discussed capacity mechanisms (CM
(planned i.e. up to 6 to 10 years ahead) which can either be plans for the introduction or the revise of an existing mechanism. In the following, present will always
refer to a period 2012/2013 for observed data or 2014 for present market systems in place, and planned refers to “from 2014 on”.

Cigre Science & Engineering • N°7 February 2017

119
 Figure 3: Map of present capacity mechanisms in jurisdictions
annual consumption (in TWh) given for the years
2012 respectively 2013. Both winter ( ) and summer
( ) peak systems are represented. A direct link to the
existence of a capacity mechanism cannot be identified.
Systems with and without capacity mechanisms exist
both with summer or winter peak. Also the size of a
system cannot be seen as a direct indicator for or against
the need of a capacity mechanism.
Similarly, the ratios between day/night and winter/
summer are compared (see technical brochure for
numeric values [1]). The ratios represent a rough
estimate of the daily respectively seasonal demand
volatility. These indicators are valid for categorizing the
power system but similar to the peak demand and annual
consumption, a direct link cannot be drawn. However, for
the winter/summer ratio, the bigger the deviation from 1,
the more seasonal a system is, which might lead to less
operating hours of peaking units and an (emerging) need
for a capacity mechanism (France, Finland, Sweden).
On the other side, this is not the case for example for
Norway. Therefore, the extended assessment in the
technical brochure of the working group also includes
the share of installed capacities respectively generated
energy per fuel type. This is in particular interesting for
the evaluation of systems that are dominated by one
resource, i.e. nuclear-, fossil fuel-, and hydro- or RES-
based systems. Moreover, no direct link can be derived
from system specification in terms of generation mix
and the existence of capacity mechanism. For example,
hydro-based systems exist both with (Brazil, Colombia)
and without capacity mechanism (Norway).
3. Capacity mechanisms
worldwide
Present Implementations
The evaluation of present capacity mechanisms is
limited to the jurisdictions with a history of capacity
mechanisms for several years. In total, 16 out of 31
market areas have implemented a capacity mechanism
Cigre Science & Engineering • N°7 February 2017
120
Figure 4: Map of planned capacity mechanisms in jurisdictions
as part of their market design. Strategic reserves or
operating reserves are implemented in 7 markets
(AU NEM, FI, PL, SE, US PJM, US NYISO)2. They
are intended to cope with seasonal extremes (“cold
reserve”) or as “safety net” if demand cannot be met
at the market. In case of US PJM and US NYISO, the
reserves are part of a combination with other market
based capacity mechanisms. Capacity payments are in
place in 5 markets (ES, GR, IE, IT, CL). In AU WA,
CL, US NYISO and US PJM capacity payments are also
part of a combined capacity mechanism. US ISONE has
implemented a capacity auction. Capacity auctions are
also part of the market design in US NYISO and US
PJM. A capacity mechanism based on reliability options
is implemented in Colombia. A visual overview is given
in the map in Figure 3.
Planned Implementations
In 10 of the 31 jurisdictions discussion is ongoing about
the redesign, introduction or abolition of a mechanism.
BE, FR and GB are in the process of introducing a
capacity mechanism coming from a market design
based on an energy-only market. BE introduces strategic
reserves to cope with seasonal shortages in the winter.
GB and FR chose for a market-based approach, namely
capacity auction respectively ex-post capacity obligation.
Discussions in DE and JP have not yet reached consensus
about a possible introduction, not to speak of a type of
mechanism. CL, IE and IT are in the process of changing
their existing capacity mechanism design. In all the 3
jurisdictions the transition is made towards a reliability
options design. SE has fixed a time line to phase-out the
strategic reserves until 2020 again. This is illustrated in
Figure 4.
4. System Reliability
In all answers, the responsibility for the long-term
generation adequacy and short-term security of supply is
borne by the government, the system operator or a panel
_______________________
2- Abbreviations for jurisdictions are listed in Table III in the appendix.
 Table I: Overview of system reliability and present / planned capacity mechanism

Country AU AU BE BR CL CN CO CZ DE ES FI FR GB GCCIA GR IE
NEM WA
System Reliability

Present ++ ++ + - - ++ ++ ++ ++ ++ ++ ++ ++ - ++ ++
3 years ahead ++ ++ - - - ++ - ++ ++ + +/- ++ ++ - ++

6 years ahead ? ++ - ? ? ? - ++ ++ - - + ++ ++ +/-

Capacity Mechanisms

Present ()        
Planned/Discussed    
None   
Main reason

Generation • • • •/- • • - - • - - •/- - - - -

Network - - - • • - - • - - - - - - -

Other •/- - - - - - - - - - - - - • -

Country IL IN IS IT JP NL NO PL SAPP SE US US US US US
CAISO ERCOT ISONE NYISO PJM
System Reliability

Present ++ + + + +/- ++ + ++ - ++ + + ++ ++ ++

3 years ahead ++ ++ ++ + ++ ++ - - ++ - - + ++ ++

6 years ahead ++ ++ ++ + +/- ++ + ++ + - - + ? ++

Capacity Mechanisms

Present       
Planned/Discussed

None        
Main reason

Generation - - - - - •/- - •/- • • • - • - -

Network - - • - - - • - - - - - - -

Other - - - - - - - - - - - - - -

that have at least one of the two parties as member. This
allocation clearly shows that the long-term responsibility
also in liberalized market design is not only left to the
outcome of the market decisions. The system operator
has hereby the role of an approved expert and counsellor
on future capacity demand expectations while the final
responsibility and authority to set rules is with the
government. In most cases, the system operators conduct
a certain monitoring and assessment. However, the
answers indicate that there are hardly procedures that are
found common in all jurisdictions. Common predefined
measures that are repeatedly used are for example the
Loss of Load Expectation (LOLE), accompanied by e.g.
simulations of severe weather conditions such as dry years
or cold spells. Mostly, such simulations and calculations
lead to annual or multi-annual adequacy / security of
supply reports that form baseline for policy making.
If applicable for the jurisdictions, interconnections are
taken into account for the assessment of security of
supply. Especially in case of system simulations and
scenario calculations, the impact of the interconnections
is considered. At the same time, next to the generation
adequacy assessment, the sizing of interconnections and
internal network planning is part of the security of supply
studies.
The received answers for system reliability is given below
in Table I. The table shows the information received for
present and expected system reliability together with the
existence of capacity mechanisms. In all jurisdictions
with concerns about the system reliability, the concern
is connected with the generation adequacy. Only in three
jurisdictions the network adequacy causes concerns about
the system reliability. On the long-term perspective, the
concerns increase with the unpredictability of the system
development.
Jurisdictions with a capacity mechanism in place
tend to have a positive expectation about their system
reliability in the future. Jurisdictions that are discussing
or implementing a capacity mechanism at the moment
have been woken up by concerns about the generation
adequacy in the near future.

Cigre Science & Engineering • N°7 February 2017

121
 Table II: Overview of existing markets

Country AU AU BE BR CL CN CO CZ DE ES FI FR GB GCCIA GR IE
NEM WA
Liberalized             
Power pool        
Bilateral contracts             
Day-ahead markets         
Real time/Balanc-
ing        
Ancillary services              
Capacity
Mechanisms ()        
Country IL IN IS IT JP NL NO PL SAPP SE US US US US US
CAISO ERCOT ISONE NYISO PJM
Liberalized            
Power pool 
Bilateral contracts            
Day-ahead markets             
Real time/Balanc-
ing           
Ancillary services           
Capacity
Mechanisms       

5. Power market and expected design. Combinations of capacity mechanisms with

profitability
Part of a comprehensive description of a power system
is the design of the power market. Main distinction
in power markets is made between liberalized and
regulated markets. In addition, the market design can be
split up based on the mechanisms and market platforms
within the market zone. The received answers are for the
greater part from liberalized markets (25 out of 31).
Out of the 31 answers, 16 markets have already had
implemented a capacity mechanism before 2014. All
capacity mechanisms are part of a liberalized market
power pools (5), with day-ahead market designs (8) or
both (3) exist (Table II). The regulated markets do not
show a need for an explicit capacity mechanism.
Figure 5 illustrates the expected development of
profitability for different generation types, namely base,
mid and peak units. Note, these values represent a general
impression and may vary for certain plants or installed
units. It shows that there is a general trend towards less
profitability of peaking units which are mostly situated
in the low state and are likely to stay there. For the mid
and base units the situation is improved and for most
jurisdictions the units are located in the reasonable
profitability state.

(a) Base (b) Mid (c) Peak

Figure 5: Expected development of profitability for generation types (current → up to 6 to 10 years)

Cigre Science & Engineering • N°7 February 2017

122
 (a) RES share (b) Interconnection capacity
Figure 6: Relationship of the presence of capacity mechanisms and RES share or interconnection capacity

6. Impact of renewables & 7. Conclusions

interconnection capacity The evaluation based on the questionnaires gives a
good overview of currently existing and developed
Next to the description of the individual system
capacity mechanisms around the world. Thanks to the
characteristics, the link of specific technologies and the
31 received contributions, a worldwide overview could
implementation of capacity mechanisms is analyzed.
be established. This paper presents the results from the
In particular, if there is an observable link between the
survey, which gives an insight over the use and perceived
share of RES on the total generation and the presence of
future needs for capacity mechanisms worldwide. It
a capacity mechanism, and, if there is a link between the
includes an excerpt of the full evaluation which is part of
level of interconnection with neighboring markets and a
the technical brochure covering the work of the CIGRE
capacity mechanism. Figure 6a and Figure 6b are used
working group C5.17 “Capacity Mechanisms: Needs,
to illustrate the relationships based on the received data.
Solutions and State of Affairs” [1].
A first assumption is that capacity mechanisms are more
The surveys were sent out in 2014. A total of 31
often present in jurisdictions with a high share (right part
jurisdictions responded giving the status of current
of Figure 6a) of intermittent RES (wind, solar) with low
implementations, given an available data basis of
marginal costs. However, the data does not confirm this
2012/2013, and an outlook up to 6 to 10 years ahead.
assumption. Both, systems with a high share and capacity
The contributions originate from both regulated (6) and
mechanism like Spain and a high share without capacity
liberalized (25) power market systems. With respect
mechanism like Germany can be observed. Obviously,
to capacity mechanisms, the jurisdictions represent
the data only shows a snapshot of the data from 2012 and
both systems with present capacity mechanisms (16)
2013. A time line showing the growth of share of RES and
and without capacity mechanisms (15). In 8 of the
the start of the capacity mechanism could reveal a clearer
jurisdictions, an implementation or a redesign of capacity
relationship. Also at the other end of the spectrum, i.e.
mechanism is currently taking place.
system with low shares of intermittent RES, both systems
with capacity mechanism (Brazil, US PJM) and without The integration of capacity mechanisms in the market
capacity mechanism (Norway) can be observed. design is independent from the market / system size.
Capacity mechanisms in both large and small systems
A second assumption is that isolated and less
are identified. Capacity mechanisms occur more often
interconnected system (left part of Figure 6b) are
in system with large seasonal differences and in system
more dependent on domestic capacity and might
with doubts about the generation adequacy in the long
therefore earlier decide to implement a mechanism to
run. The large variety and combination of capacity
remunerate domestic capacity. Similar to intermittent
mechanisms in place indicates that country specific
RES, a coherent link between interconnection capacity
characteristics lead to individual designs.
and capacity mechanism cannot be observed. Strongly
interconnected jurisdictions like Finland and Sweden still The ongoing discussion in multiple European countries
opt for a capacity mechanism to cover seasonal peaks. about new implementations of capacity mechanisms and
France, GB and Belgium with a low interconnection the reported redesign of existing capacity mechanisms
capacity for European jurisdictions are in the process emphasize the dynamic nature of power markets. While
to implement a capacity mechanism. However, a direct a direct link between implementation of a capacity
link to the interconnection is not stated as reason for mechanism and a certain power system criterion could
implementation. not be observed, the assessed system criteria contribute

Cigre Science & Engineering • N°7 February 2017

123
to some extent to the decision for or against the 25 CN CSG China Southern Grid, CSG
implementation of a capacity mechanism. But, it also 26 GCCIA GCCIA (Kuwait, Saudi Arabia Eastern
shows that the decisions are mostly taken due to very Operating area, Bahrain, Qatar, Abu

Asia
Dhabi, Oman)
country-specific constellations of the described criteria.
27 IN India

References 28 JP Japan
29 SAPP South African Power Pool
[1] G. Doorman, J. Barquin, L. Barroso, C. Batlle, A. Cruickshank, C.

rica
Af-
Dervieux, R. Flanagan, J. Gilmore,
30 AU NEM Australia National Electricity Market
J. Greenhalg, H. Höschle, P. Mastropietro, A. Keech, M. Krupa, J.

Austra-
Riesz, B. LaRose, S. Schwenen, 31 AU WA Western Australia

lia
G. Thorpe, K. De Vos, L. de Vries, and J. Wright, Capacity
mechanisms: needs, solutions and state of affairs. Technical Brochure
647, Paris, CIGRE, 2016. Acknowledgement
Appendix Thanks to all country experts for providing answers
in the worldwide survey. Special thanks to the CIGRE
Table III: List of submissions
C5.17 working group members for feedback on the
# Code Jurisdiction survey design and evaluation of the answers.

1 BR Brazil Biographies
America
South

2 CL
3 CO
4 US CAISO
5 US ERCOT
America
Chile
Hanspeter Höschle holds a Diplom degree in
Colombia
industrial engineering from the Karlsruhe Institute of
California Independent System Operator Technology, Germany. Since 2012, he has worked as a
Electric Reliability Council of Texas Ph.D. researcher at the KU Leuven, Belgium, on power

6 US ISONE ISO New England Reliability system economics and the modeling of electricity
North

Coordinator
markets, in particular capacity mechanisms. He holds
7 US NYISO New York ISO
a Ph.D. fellowship of the Research Foundation
8 US PJM PJM Interconnection
- Flanders (FWO) and the Flemish Institute for
9 CZ Czech Republic
Technological Research (VITO). He was an active
10 BE Belgium
member in the CIGRE C5.17 working group and
11 ES Spain
organized the worldwide survey. He already discussed
12 FI Finland
various aspects of capacity mechanisms both at the
13 FR France
CIGRE symposium 2015 and the Paris Session in 2016.
14 DE Germany
15 GB Great Britain Gerard Doorman has a in Electric Power Engineering
16 GR Greece
from the Norwegian University of Science and
Europe

17 IE Ireland & Northern Ireland (SEMO)

Technology (NTNU) from 2000. He has experience
18 IS Island
from consulting and research. From 2006 to 2013
19 IL Israel
he was full time professor at NTNU. Presently he is
Special Advisor at the Norwegian TSO Statnett, and
20 IT Italy
also holds a professor position. His research interests
21 NL The Netherlands
are within power market design and system operation.
22 NO Norway
He was convenor of the CIGRE C5.17 working group
23 PL Poland
on capacity mechanisms.
24 SE Sweden

Cigre Science & Engineering • N°7 February 2017

124
An adaptive protection infrastructure for
modern distribution grids with distributed
generation
Vasileios A. Papaspiliotopoulos, George N. Korres* and Nikos D. Hatziargyriou
National Technical University of Athens, Greece
Abstract
The presence of renewable energy resources and
conventional distributed generation (DG) in modern
distribution networks challenges the traditional protection
schemes, since the radial single-fed concept is no longer
adopted. In this paper, a new adaptive protection system,
which takes advantage of multiple relay setting groups
(SGs) and optimisation techniques, is introduced. Moreover,
a hardware-in-the-loop (HIL) configuration appropriate
for testing adaptive protection systems is presented. A
distribution test system with large DG penetration is used
to demonstrate the efficacy of the proposed solution, and
simulation results are thoroughly discussed.
1. Introduction
These days, the global demand for more “green” energy
resources has led to large integration of distributed
generation (DG) into distribution systems, and the concept
of “smart grid” has introduced new sophisticated network
configurations. However, this evolution of distribution
systems has posed new challenges to long-established
protection schemes, including feeder protection blinding,
sympathetic tripping of adjacent feeder relays, mis-
coordination between reclosers and downstream fuses,
failed auto-reclosing, and unintentional islanding.
Therefore, the development of advanced protection
solutions is of great importance, since distribution systems
reliability and service continuity are inextricably linked
with the performance of protection schemes.
So far, non-directional overcurrent protective devices were
successfully employed in distribution networks operating
with single-source radial topology. Nevertheless, the
growing DG penetration in distribution grids, combined
with the variable grid operating mode, has caused
* gkorres@cs.ntua.gr
KEYWORDS
Terms - Adaptive protection, directional overcurrent relays, distributed generation, distribution systems, hardware-in-
the-loop simulation.
Cigre Science & Engineering • N°7 February 2017
125
significant change in short circuit levels and bidirectional
fault current flows. To address this paradigm shift in
distribution systems operation, new protection strategies
are mainly orientated to the implementation of adaptive
protection schemes, utilising directional overcurrent
relays (DOCRs) with communication capabilities and
multiple setting groups (SGs).
Several methods have been used by protection engineers
to derive proper relay setting values, satisfying also the
so-called coordination constraints between primary
and backup protective devices. Worldwide, distribution
utilities were using expert rules to determine appropriate
pickup current and time dial values for the inverse-time
characteristics of feeder relays, achieving also sufficient
coordination with downstream reclosers and fuses. On
the other hand, various optimisation methods have been
proposed in the literature for the problem of optimal
setting and coordination of the relays, which has been
stated and successfully solved as linear programming
(LP), nonlinear programming (NLP), and mixed-integer
nonlinear programming (MINLP) problem.
In this paper, the impact of DG presence on conventional
distribution protection schemes is thoroughly discussed.
Then, a hardware-in-the-loop (HIL) infrastructure is
proposed, which can be used as a complete testbed for
adaptive protection schemes. Moreover, the integration
of optimisation techniques in the adaptive procedure
is proposed. This novel concept can lead to great
enhancement of adaptive protection schemes developed
so far, since it combines the flexibility of automatic relay
SG adjustment with the determination of optimal relay
setting parameters.
The remainder of the paper is structured as follows. In
Section 2, protection blinding, sympathetic tripping,
 Fig. 1. Change in flowing fault currents due to DG presence (protection blinding case).

and failed reclosing phenomena are discussed in detail. which frequently follows an out-of-zone fault, or due to
Section 3 presents and describes the architecture of the DG backfeed to a fault on an adjacent feeder [1], [2]. This
proposed adaptive protection infrastructure, introducing phenomenon is also called false tripping, and belongs to
also the concept of optimal setting group determination. the wide class of nuisance tripping problems due to various
Proof of concept and results obtained by applying the root causes (e.g. motor starting, magnetising inrush current
proposed approach to a distribution test system with large during transformer energisation, cold-load pickup, voltage
DG penetration are analysed in Section 4. Finally, Section sag or swell etc.). In this paper, we focus on the sympathetic
5 highlights the conclusions of this work. tripping scenario caused by the contribution of DG for
faults beyond its feeder protection zone.
2. Impact of distributed generation As has been discussed in recent literature, DG units can
on feeder protection cause false tripping and undesirable disconnection of an
adjacent healthy feeder. The basic principle of sympathetic
2.1. Protection Blinding tripping can be explained in Fig. 2, where a fault occurs
at Feeder 2 and the DG unit connected to Feeder 1 feeds
Traditional distribution systems had only one source
the short-circuit through the substation bus. If the DG
contributing to a fault incident, that is the main substation,
contribution is particularly large, relay R1 may operate
and protective relays were set to see a certain distance down
before relay R2 takes action and clears the fault. This
the radial feeder. This distance is usually referred to as the
is possible when non-directional overcurrent relays are
reach of the protective device, which is actually determined
used for feeder protection, which cannot discriminate
by the minimum fault current that the device can detect.
the change of fault current direction (reverse fault). The
However, the connection of DG units to modern distribution
utilisation of non-directional overcurrent relays is a
networks introduces additional fault current sources, which
long-established protection practice for most distribution
may increase the total short-circuit level within the network,
utilities owing to their single-fed radial networks, unlike
while altering the magnitude and direction of fault currents
modern multi-source distribution systems with bi-
sensed by installed protective devices.
directional current flow.
When the DG plant is located between the utility substation
and the fault location, the total fault current is increased
due to the partial contribution of DG, as illustrated in
Fig. 1. On the other hand, the fault current seen by feeder
relay R1 is actually decreased for the same fault, owing to
fault current division between the sources, which may not
exceed the pickup current setting of R1. This undesirable
protection performance is widely known as protection
blinding [1], [2]. The blinding effect results in delayed
protection operation or even total desensitisation in case
of weak upstream system and large DG penetration. This
phenomenon is also called protection under-reach, since
the actual reach of the feeder relay is decreased due to DG
fault current contribution, as also shown in Fig.1. Fig. 2. DG contribution to fault incident on adjacent feeder
(sympathetic tripping case)

2.2. Sympathetic Tripping 2.3. Failed Reclosing

Sympathetic tripping refers to the undesirable operation of Protection problems in distribution networks may be
the feeder relay due to unbalanced or high-load condition, caused by the automatic reconnection of the utility in

Cigre Science & Engineering • N°7 February 2017

126
 Fig. 3. First shot of reclosing sequence and DG disconnection.

case of DG presence. A reclosing attempt is deemed to be
successful when there is sufficient time margin between
shots for the fault arc to dissipate and clear, which means
that each DG unit must detect the presence of a fault and
disconnect early in the reclose interval, as illustrated in
Fig. 3. Otherwise, the DG unit will continue to feed the
fault hindering the arc extinction, and the fault that would
have been temporary becomes permanent.
A potential reclosing failure to clear the fault means that
some customers will suffer a sustained outage, whereas
they should be subjected to only momentary interruptions.
Thus, the reliability of the power delivery system is slightly
degraded. Furthermore, the active power unbalance
during the dead time of the reclosing sequence can lead to
frequency variation in the islanded part of the distribution
grid, and generators may drift away from the synchronism
with respect to the main system [1], [2]. In this scenario,
a reclosing attempt would couple two asynchronously
operating systems. Moreover, conventional reclosers are
designed to reconnect the circuit only if the substation
side is energised and the opposite side is de-energised. In
case of DG integration, there would be active sources on
both sides of the recloser, hampering proper operation of
the reclosers.
In addition to the aforementioned protection challenges,
the constantly increasing presence of DG units in
temporarily. The breaker re-closes its poles after a
sufficient time delay, to allow transient faults to extinguish
and re-establish cold non-ionised air at the fault point. In
case of a permanent fault, the recloser should operate with
its slow curve, using a time-delayed overcurrent element
51P/N and allowing the load-side fuse to clear the fault.
Fuse-saving schemes are based on the premise that both
recloser and downstream fuse sense the same current.
However, the DG presence in distribution networks
increases the total fault current, and thus the branch fuse
will always see more current than the upstream recloser
(Fig. 4). As a result, the fast curve will be shifted to the
right on the time-current characteristic (TCC) diagram,
and the fuse may blow before the fast operation of the
recloser, as depicted in Fig. 5.
Fig. 4. Fuse-saving scheme in DG-penetrated distribution system.

distribution systems deteriorates the coordinated operation

of feeder reclosers with downstream lateral fuses. Fuses
are the most common overcurrent protective means in
distribution networks, characterised by low cost and
high reliability, while reclosers are overcurrent protective
devices most often used to “give every fault a chance to be
temporary” [3]. The vast majority of distribution utilities
have successfully employed fuse-saving schemes so far,
aiming at the highest level of service availability for their
end-customers.
In a fuse-saving scheme, the recloser operates with
its fast curve (an instantaneous overcurrent element
50P/N is usually used), and trips the feeder breaker for
any downstream phase or ground fault. The breaker
is tripped before the downstream fuse begins to melt, Fig. 5. Loss of coordination between recloser and lateral
and the instantaneous protection element is deactivated fuse due to DG presence.

Cigre Science & Engineering • N°7 February 2017

127
 Fig. 6. Set-up of the proposed HIL infrastructure.
3. Proposed adaptive protection
infrastructure
In recent years, several protection solutions for the
previous challenges have been suggested; however
adaptive protection concept has dominated over its
antagonists. Adaptive protection has been defined in [4]
as “a protection philosophy which adjusts the settings
of various protection functions in order to make them
more attuned to prevailing power system conditions”.
Characteristic implementations of adaptive protection in
distribution systems and microgrids have been presented
in [5]–[7].
Adaptive protection schemes require off-the-shelf digital
or numerical DOCRs with several SGs, which can
be configured either locally or remotely. The element
parameter values of available SGs are calculated and
stored offline in relays. In addition, adaptive protection
schemes require communication between each protection
device and a central control unit. The central controller
monitors constantly the grid configuration by collecting
operational data (e.g. circuit breakers status), and updates
appropriately the SGs of the relays when necessary.
Current adaptive protection technology can be developed
further, making use of powerful optimisation algorithms.
These algorithms can determine optimal relay SGs, which
minimise the operating time of feeder DOCRs, either
acting as primary or backup protection, optimising thus
the overall protection performance. Optimal SGs can be
either pre-calculated offline when the system operating
scenarios are pre-specified or calculated online when the
system operates under numerous configurations and the
number of available relay SGs is not sufficient.
Cigre Science & Engineering • N°7 February 2017
128
3.1. Description of HIL Laboratory Infrastructure
An innovative HIL infrastructure has been developed
in the Electric Energy Systems Laboratory (EESL) of
the National Technical University of Athens (NTUA),
Greece, for testing and evaluating the performance of
adaptive protection schemes. HIL simulation is a very
promising validation technology, constituting a closed-
loop procedure where a hardware device is tested through
bidirectional interaction with a simulator. HIL process
provides the capability of identifying hidden defects of
the tested equipment before applied to the real system,
and reveals features not visible in pure simulations.
The set-up of the proposed infrastructure, illustrated in
Fig. 6, consists of a Real Time Digital Simulator (RTDS),
digital multifunction protection relays with multiple SGs,
as well as a programmable logic controller (PLC), which
plays the role of the centralised controller. The examined
distribution network is simulated on the RTDS, while the
digital relays provide supervision and protection of the
distribution feeders. The PLC is firstly responsible for
gathering the circuit breaker (CB) statuses, and secondly
for transition of the relays to the proper SG whenever a
major change in the network operation takes place (e.g.
DG on/off, grid reconfiguration, tie breaker closed/open).
The optimal determination of feeder relay SGs in the
proposed adaptive protection system is based on the
generic DOCR setting and coordination problem for each
possible network configuration. The objective function
aims at minimising the aggregate operating time of both
primary and backup DOCRs installed at the distribution
network, subject to operation and coordination constraints
which are imposed by distribution network operators
 Fig. 7. Single-line diagram of the simulated distribution system.
(DNOs). As a result, the entire performance of the
distribution system protection can be greatly enhanced.
Various approaches have been proposed in recent
literature [8]–[10]. The NLP approach was considered
for the suggested system and solved by state-of-the-art
optimisation packages. An extensive presentation of the
problem formulation and description of the employed
optimisation tools can be found in [11].
4. Test results
To evaluate the performance of the proposed adaptive
protection scheme, a 5-bus distribution network with
variable operating mode was simulated, representing a
simplified two-feeder portion of the Hellenic Distribution
System (Fig. 7). The phase overcurrent elements (51P)
of feeder relays were enabled and tested during the
evaluation procedure, operating with the IEC C1 standard
inverse characteristic. Detailed system and protection
data are given in [11].
The evaluation procedure was actually composed of three
stages. In the first stage, the adaptive logic was inactive,
while the likelihood of protection blinding and sympathetic
tripping occurrence in DG-penetrated distribution
(a)
Fig. 8. Protection blinding (a) and sympathetic tripping (b) of R3 shown on the TCC diagram.
Cigre Science & Engineering • N°7 February 2017
129
systems was verified. The main feeders were protected
only by the non-directional relays R3 and R7, according
to the traditional protection practice. The obtained HIL
test results revealed the inadequacy of existing protection
settings, which were determined considering single-fed
radial configuration for the distribution system.
More specifically, when the simulated network was
operating without DG presence and a three-phase fault
was conducted at Bus 1.2., relay R3 operated properly after
0.554 s, seeing the total fault current (2.12 kA) supplied
solely by the upstream external grid. Then, the same short-
circuit test was conducted with DG1 connected. The total
fault current was increased due to DG presence, however
the partial contribution from the external grid was reduced
to 932 A, resulting in a greatly delayed tripping time of
relay R3 (2.23 s). In addition, a symmetrical fault was
applied at Bus 2.1, causing faster operation of R3 than R7.
Stated another way, R3 was tripped sympathetically to R7
for an out-of-zone fault, disrupting the service continuity
of the healthy line L1. The previous description is made
clear in Fig. 8, where the TCCs of R3 and R7 relays are
plotted for each test case, showing also the sensed fault
currents and the corresponding tripping times.
(b)
In the second stage of the evaluation procedure, the adaptive
protection logic was activated. The connection status (on/
off) of the installed DGs was being changed on the RTDS,
while the PLC was updating appropriately the binary
signals to relays. Each variation of relay input values was
followed by successful transition of the active SG to the
appropriate one, adjusting properly the feeder protection.
In the third and final stage, the proposed infrastructure
was put into effect in combination with the optimal
determination of relays’ SG, whereas several HIL tests
were carried out to demonstrate its efficacy, as well as
the treatment of protection blinding and sympathetic
tripping problems. It was considered that DOCRs were
installed at both ends of the line segments, and their
nominal current was 1 A. Various operating modes for the
simulated distribution network were taken into account,
and protection performance was assessed for several
short-circuit scenarios.
Simulation results for two configuration cases, covered by
two different SGs, are given in Tables I, II. Specifically, all
primary-backup protection pairs with the corresponding
current transformer ratios (CTRs), the close-in fault
currents seen by primary and backup DOCRs, the optimal
and settings of each DOCR, and the achieved tripping
times for each protection pair are presented. The lower
and upper bounds for and settings were taken equal to
1.5÷5 A and 0.1÷1.0, respectively, the desired tripping
times were restricted between 0.1 s and 1.5 s, and the
required coordination time interval was assumed equal to
0.3 s as per common practice. It is evident that the optimal
protection settings ensure not only fast tripping times, but
also complete coordination between each primary-backup
relay pair . Moreover, it should be noted that the employed
optimisation solvers spent less than 1 s to converge and
find the optimum solution, demonstrating the robustness
of the developed NLP model.

Table I. Simulation results for the 5-bus test system with DG1 connected (SG1 active)

Primary DOCRs Backup DOCRs Δt = tb - tp

Relay ID (CTR) Fault Current (A) PU (A) TD Trip. Time tp (s) Relay ID Fault Current (A) Trip. Time tb (s) (s)
R1 (100/1) 11547 3.55 0.10 0.194
R2 (300/1) 4212 1.50 0.10 0.306 R4 4212 0.606 0.300
R3 (300/1) 8614 2.72 0.10 0.290 R1 1148 0.590 0.300
R4 (300/1) 10642 4.48 0.10 0.331
R5 (200/1) 13630 1.50 0.10 0.177 R3 3470 0.477 0.300
R1 1114 0.606 0.307
R7 (300/1) 12012 4.07 0.10 0.299
R4 3848 0.659 0.360
R9 (200/1) 4213 1.50 0.10 0.258 R7 4213 0.558 0.300

Table II. Simulation results for the 5-bus test system with DG1 & DG2 connected (SG2 active)

Primary DOCRs Backup DOCRs

Δt = tb - tp
Relay ID (CTR) Fault Current (A) PU (A) TD Trip. Time (s) Relay ID Fault Current A) Trip. Time (s) (s)

R1 (100/1) 11547 3.42 0.10 0.192

R4 4212 0.544 0.318
R2 (300/1) 9145 1.50 0.10 0.226
R8 1866 0.544 0.318
R1 1134 0.577 0.300
R3 (300/1) 10203 2.89 0.10 0.277
R8 1735 0.577 0.300
R4 (300/1) 10642 3.94 0.10 0.312
R5 (200/1) 13793 1.50 0.10 0.176 R3 3692 0.476 0.300
R1 1114 0.586 0.300
R7 (300/1) 12012 3.66 0.10 0.286
R4 3848 0.586 0.300
R8 (300/1) 2512 1.74 0.10 0.439
R9 (200/1) 6469 1.50 0.10 0.221 R7 4137 0.521 0.300

Cigre Science & Engineering • N°7 February 2017

130
5. Conclusion
In this paper, an adaptive protection system with
optimally calculated relay setting values was presented.
This innovative concept constitutes a versatile protection
solution for modern distribution systems which face
large DG penetration. First of all, it was stressed that
the proposed scheme can completely resolve the
new protection problems arisen due to DG presence.
Furthermore, optimal determination of setting groups
for the protection relays was achieved by means of
NLP techniques. The efficacy, as well as the high rate
of convergence, of the developed optimisation model
was verified by a series of simulations. As a next step,
the proposed protection scheme can be further evolved
by using optimisation algorithms to recalculate the
protection setting parameters in real time, thus leading
to significant improvement in the operation of power
distribution networks.
6. Aknowledgement
This work was supported in part by the European
Community’s Seventh Framework Programme
(FP7/2007-2013) under grant agreement number 308755
– the SuSTAINABLE project, funded under the EC call
"ENERGY.2012.7.1.1".
7. References
[1] K. Kauhaniemi and L. Kumpulainen, “Impact of distributed
generation on the protection of distribution networks,” in Proc.
2004 Developments in Power System Protection Conf., pp. 315–
318 (vol. 1).
[2] B. Hussain, S. M. Sharkh, S. Hussain, and M. A. Abusara,
“Integration of distributed generation into the grid: Protection
challenges and solutions,” in Proc. 2010 Developments in Power
System Protection Conf., pp. 1–5.
[3] Cooper Power Systems, Electrical Distribution-System Protection,
3rd ed., 1990.
[4] A. Phadke and J. Thorp, Computer Relaying for Power Systems,
2nd ed., John Wiley & Sons, Inc., 2009.
[5] H. Laaksonen, D. Ishchenko, and A. Oudalov, “Adaptive protection
and microgrid control design for Hailuoto island,” IEEE Trans.
Smart Grid, vol. 5, no. 3, pp. 1486–1493, May 2014.
[6] I. Abdulhadi, F. Coffele, A. Dyśko, C. Booth, G. Burt, G. Lloyd,
and B. Kirby, “Performance verification and scheme validation of
adaptive protection schemes,” in Proc. 2010 CIGRE Session, pp.
1–9.
[7] F. Coffele, C. Booth, and A. Dyśko, “An adaptive overcurrent
Cigre Science & Engineering • N°7 February 2017
131
protection scheme for distribution networks,” IEEE Trans. Power
Del., vol. 30, no. 2, pp. 561–568, Apr. 2015.
[8] D. Birla, R. P. Maheshwari, and H. O. Gupta, “A new non-linear
directional overcurrent relay coordination technique and banes
and boons of near-end faults based approach,” IEEE Trans. Power
Del., vol. 21, no. 3, pp. 1176–1182, Jul. 2006.
[9] D. Birla, R. P. Maheshwari, and H. O. Gupta, “An approach to
tackle the threat of sympathy trips in directional overcurrent relay
coordination,” IEEE Trans. Power Del., vol. 22, no. 2, pp. 851–858,
Apr. 2007.
[10] P. Bedekar and S. Bhide, “Optimum coordination of directional
overcurrent relays using the hybrid GA-NLP approach,” IEEE
Trans. Power Del., vol. 26, no. 1, pp. 109–119, Jan. 2011.
[11] V. Papaspiliotopoulos, G. Korres, V. Kleftakis, and N.
Hatziargyriou, “Hardware-In-the-Loop Design and Optimal
Setting of Adaptive Protection Schemes for Distribution Systems
with Distributed Generation,” IEEE Trans. Power Del., in press.
8. Biographies
Vasileios A. Papaspiliotopoulos received the diploma
in electrical and computer engineering from the National
Technical University of Athens (NTUA), Athens,
Greece, in 2012, where he is currently pursuing the Ph.D.
degree in adaptive protection systems. His research
field includes power system protection, optimisation
techniques, and industrial automation. He is a member
of the Technical Chamber of Greece.
George N. Korres received the diploma and Ph.D.
degrees in electrical and computer engineering from
the National Technical University of Athens (NTUA),
Athens, Greece, in 1984 and 1988, respectively. He
has been involved in the development and installation
of the Hellenic emergency-management system by
ESCA in the early 1990s. Currently, he is Professor
with the School of Electrical and Computer Engineering
at NTUA. His current fields of interest include power
system state estimation, power system protection,
substation automation, and supervisory control and
data-acquisition systems. Prof. Korres is a member of
the Technical Chamber of Greece and a Greek member
of the International Council on Large Electric Systems
(CIGRE) Study Committee D2 “Information systems
and telecommunication.”
Nikos D. Hatziargyriou is Chairman and CEO of the
Hellenic Distribution Network Operator (HEDNO).
Since 1984 he is with the Power Division of the Electrical
and Computer Engineering Department of the National
Technical University of Athens (NTUA), and since 1995
he is professor in Power Systems. From February 2007
to September 2012, he was Deputy CEO of the Public
Power Corporation (PPC) of Greece, responsible for
Transmission and Distribution Networks, island DNO
and the Center of Testing, Research and Prototyping.
He is Fellow Member of IEEE, past Chair of the Power
System Dynamic Performance Committee, Distinguished
Cigre Science & Engineering • N°7 February 2017
132
member of CIGRE and past Chair of CIGRE SC C6
“Distribution Systems and Distributed Generation”. He
is chair of the Advisory Council of the EU Technology
Platform on SmartGrids. He is member of the Energy
Committee of the Athens Academy of Science. He has
participated in more than 60 R&D Projects funded by
the EC and the industry.
Solid-gas insulation in HVDC gas-insulated
system: Measurement, modeling and
experimental validation for reliable operation
R. Gremaud1*, C. B. Doiron1, M. Baur1, P. Simka1, V. Teppati1, B. Källstrand2, K. Johansson2, M. Hering3, J.
Speck3, S. Großmann3, U. Riechert4, U. Straumann4

ABB Switzerland, Corporate Research, Switzerland

1

2
ABB AB, Corporate Research, Sweden
3
Institute of Electrical Power Systems and High Voltage Engineering, TU Dresden, Germany
4
ABB Switzerland, High Voltage Products, Switzerland

Abstract interface moves from near the conductor to the enclosure

area between switch-on time until DC steady-state is
Robust dimensioning of DC gas-insulated systems reached (see Ref. [2]).
requires the knowledge of the electrical field distribution
at all times. Starting from a capacitive distribution at
voltage switch-on, the field distribution evolves towards
a resistive distribution at a pace dictated mainly by the
conductivity of the solid epoxy insulation.
This article presents an experimental validation scheme
for modeled DC electrical field distribution. It is based
on (i) measurements of electrical currents in the solid and
the gas, (ii) measurement of surface potential distribution
of the surface of epoxy insulators, and (iii) determination
of flashover voltage of insulators under HVDC stress.
To mimic the effect of ohmic losses in the conductor, a
temperature gradient is applied across the insulator. DC
Field distributions are calculated with a finite element
model taking into account the relevant charge transport
processes in the gas as well as in the solid.

1. Introduction
Interfaces are critical locations in gas-insulated systems:
Failure occurs in the gas, but is often mediated by charges
accumulated on insulators: surface flashovers generally
follow a short path of high electric field. In practice,
this often means that the field component normal to
the insulating surface is critical for surface charge
accumulation, but flashover is mainly triggered by a high Figure 1(a), Components of a HVDC gas-insulated system: (1) conductor, (2)
enclosure, (3) insulating gas, (4) solid insulator, (5) capacitive screen, (6) insert,
tangential field component [1]. The temporal evolution of (7) current collector, (8) dielectric coating on metallic surfaces. Electrical effects
the DC electric field in the system is mainly determined for the design of a HVDC insulator in gas-solid insulation: electronic and/or
ionic conduction, polarization, space charge accumulation. In gas insulation: Ion
by conduction in the solid insulation. The radial potential generation by (a-d) and ion drift (e); On interfaces/surfaces (f-j). (b), Design op-
distribution changes with time from capacitive to timization scheme for HVDC insulators. In red: optimization constraints on the
electrical volume conductivity σ of the insulator material. Other constraints are in
resistive-controlled. This causes a field redistribution, as blue, and potential modification of components in orange color. En and Et: electrical
the location of highest electric field on the insulator-gas field components normal, respectively tangential to the epoxy insulator-gas surface.

*robin.gremaud@ch.abb.com

KEYWORDS
Charge emission, epoxy insulation, gas insulated system, high voltage direct current (HVDC), surface potential.

Cigre Science & Engineering • N°7 February 2017

133
An optimal design of an epoxy insulator with respect to the
DC resistive field distribution follows many –sometimes
contradictory – constraints, as represented in the scheme
of the Figure 1(b). Note that not all design measures
are necessary to achieve a robust design: for the actual
DC design, a sufficient reduction of the dielectric stress
compared to the AC design was obtained by geometrical
optimization and insertion of a current collector. More
than 10 DC insulators in realistic arrangement were
dielectrically tested by superimposed lightning and
switching impulses (unipolar and bipolar) under high-
load steady-state DC stress (both polarities) [3].
The magnitude and sign of charge accumulation on the
insulator depends on the difference between the normal
components of electrical current from the gas and from
the solid. The resulting electrical field at the solid-gas
interface therefore depends on voltage level, nominal
current and induced thermal gradient in the insulation,
system geometry, and nature and magnitude of gas ion
sources [4].
Due to the complexity of the – often non-linear [5] –
physical effects at hand in the gas and solid insulation,
it is necessary to validate the multiphysics finite element
modeling (FEM) approach used to design DC insulation
by a direct monitoring of the electric field [6] or surface
potential at the solid-gas interface during the transition
from capacitive to resistive field [7, 8].
In this contribution, we present surface potential
measurements of an epoxy insulator in a real-scale
DC gas-insulator system under high current load. The
temporal evolution of the surface potential is modelled
by our standard finite-element modeling of gas and
solid insulation. The model is calibrated by independent
measurements of the leakage current of solid insulation and
emission current in the gas of technical metallic surfaces.
Finally, the applicability of the same modeling approach
is demonstrated on breakdown voltage experiments. This
present experimental validation is essential to derive
reliable testing procedures, in particular to confirm the
DC stress duration needed for withstand verification tests.
2 Theory and electrical modeling
Considered are electrical conduction mechanisms in the
Cigre Science & Engineering • N°7 February 2017
134
solid and in the gas; the two parts of the model being
connected at the solid-gas interface (Figure 1(a)). Surface
and bulk of the solid are assumed to have the same
conduction properties.
2.1 Solid: DC conductivity and polarization
Considering an insulator contacted on one side to a ground
electrode, and on the other side to a high voltage (HV)
electrode at potential U, the total current density through
any cross-sectional area of the insulator is then:
(1)
where is the DC conductivity of epoxy and
(2)
the displacement current. The first term represents the
“fast” capacitive displacement current. At the time scale
of experiments (seconds to week), the starting instant
polarization ε∞ is well-approximated by the 50 Hz
permittivity value . The second term P accounts for all
“slow” polarization processes such as structural processes
– e.g. dipole reorientations-, interface polarization or
inter-phase polarization. Measurements of the leakage
current as a function of time on a plate insulator geometry
are used to determine the DC conductivity and the slow
polarization of the epoxy composite insulation material.
The procedure for such measurement is to rise the
voltage from 0 to the test voltage U with constant rate
dU/dt and then keep the voltage constant. The leakage
current between HV and ground electrode is monitored
continuously. As soon as the applied voltage (U) is
constant, the current density becomes
(3)
An empirical description for the amplitude of the slow
time-dependent polarization current may be given by the
empirical Curie-von Schweidler relation:
(4)
If the polarization obeys linear relaxation dynamics,
the depolarization current (i.e., the current flowing after
switching-off the voltage) becomes
The DC conductivity can therefore be determined for long
measurement times as . grows exponentially with increasing electric field, but
2.2 Gas: ion generation and drift
Unlike solids, the current-voltage characteristics of gases
is linear only for very low applied fields. Hence, the
gas conductivity should rather be described by a carrier
density based model, where is a purely
local quantity, with carrier density ni, mobility µi and
charge qi for each carrier (ion) type i.
Field lines in the gas ending onto the solid insulation
surface define a capture volume Vcapt for ions. This
capture volume generally changes with time during
surface charging. Ions generated in the capture volume
drift along field lines and, if not recombined with counter
ions, contribute to surface charging of the solid. Because
of the comparatively large mobilities, HV fields in the
gas-insulated systems (>> 100 V/m) imply high ionic
drift velocities, and consequently separation of ions of
opposite polarity is fast enough for recombination in the
bulk gas to play a negligible role. A local ionic current can
therefore be defined if one takes into consideration the
density of gas ion (generation), and ion drift.
2.2.1 Charge injection and multiplication in SF6
In SF6 insulation, accumulated surface charges are
associated to positive or negative SF6 ions which can be
produced by three different mechanisms:
1. Ionization by cosmic and earth radiation forming a free
electron and a positive SF6 ion. If the electron is attached
by another SF6 molecule an ion pair (IP) is formed.
This process is usually referred to as natural ionization
and is characterized by a production rate between 20
and 50 IP/(s·cm3), dependent on geographic location
and gas pressure. To reflect the fact that volume ion
production conserves charge neutrality, our model
assumes a symmetric ion generation scheme, and a
constant natural production rate of ion pairs (IP) .
2. Electron emission from a metallic cathode by
Cigre Science & Engineering • N°7 February 2017
135
thermionic or field emission. If the electron is attached
by a SF6 molecule a negative SF6 ion is formed.
The natural ionization current saturates at a rather low
electric field values, <1 V/mm. The emission current
starts usually at very high electric fields (~1 MV/mm).
For both cases free electrons can again be released from
SF6- by collisional detachment. The above mentioned
mechanisms are called “charge injection” from now on.
3. If the applied electric field is sufficiently high to make
the electron attachment rate to SF6 molecules lower
than the ionization rate, then collisional ionization can
lead to an exponential growth of free electrons (and
positive SF6 ions). This mechanism is called “charge
multiplication” from now on.
In the model, the additional1 combined effect of charge
injection and multiplication near surfaces is taken into
account by a current source term on conductor and
enclosure with boundary condition .
2.2.2 Ion drift
To describe the non-linear nature of the conduction
behavior in the gas, the best approach is to solve the
drift-diffusion-reaction equations for the gaseous ions in
parallel with the Poisson equation for the electric field [9].
Negative and positive charge carriers (expressed as charge
densities ρ – and ρ +) in an electric potential ϕ move
because of the electric field and diffusion.
The equation of continuity reads:
(5)
with the recombination coefficient R and the elementary
charge e.
The ionic current densities can be expressed by the drift-
diffusion equation2 :
(6)
In this equation, is the ion mobility, which for
SF6 can be assumed to be the same for both polarities,
_______________
1- Additional to the natural ionization contribution.
2- Neglecting any additional background current distribution, e.g. caused by
convection.
 Figure 2 Left: Polarization (IPol) at 20 kV source voltage and Depolarization (IDepol) currents for a cast alumina-filled epoxy
sample. The grey dashed line indicates the level of DC current. Temperature: 50°C. I0: base current at 0 V source voltage.
Right: cross-section of cast sample geometry used for the (de-)polarization measurement.
[10] and D the diffusion coefficient. The diffusion
coefficient is related to the mobility via Einstein’s relation
the Boltzmann constant, T0 = 300 K is the
gas temperature.
A direct solution of these equations is numerically
challenging for complex geometries, so we instead use
a model inspired from Ref. [11]. In this model, only the
ionic current from the gas impinging the insulator surfaces
is solved for, and not the ion concentration in the whole
gas volume. In practice, this is done by solving the steady-
state drift-diffusion equations(6) together with the source
term for natural ionization (7):
constant (7)
for each ion species in the gaseous domain at every time
step, making use of the fact that the time needed for newly
created ions to drift to a surface is much shorter than all
the other relevant timescales in our problem.
2.3 Charging of gas-solid interface, surface potential
Gas ions located within the capture volume accumulate
on the insulator surface where they build a surface charge.
Similarly, space charge from the solid insulation also
tends to accumulate on the other side of the interface.
Accumulated charges typically reduce the normal
component of the field at the insulator interface at the side
from which they reach the interface. Eventually, they can
also contribute to conduction in the solid.
The charging of the gas-solid interface is modeled by
introducing the following boundary current source on the
insulator interface:
(8)
where ng is the normal vector pointing from the surface
of the insulator in the gas and the current
density from the gas phase.
So-called “open” boundary conditions are applied to the
other interfaces (electrodes and grounded tank), i.e. using
in regions where the charges flow out
Cigre Science & Engineering • N°7 February 2017
136
of the domain, and otherwise. The development
of surface charge on the solid-gas interface (“insulator
surface”) is thus determined by the continuity equation:
(9)
where is the current in the solid. Electric field
and surface potential ϕ are then calculated
using the Poisson equation:
(10)
Whether gas or solid charging current dominates on the
interface depends locally on the ratio of normal currents.
If natural ionization is the only source of gas ions, this
ratio is where L is the
length of the field line ending on the insulation surface
and En the normal field component in the solid [2, 11].
Therefore, the solid tends to dictate the field distribution
for high voltages, unless the insulator geometry creates
large capture volume or charge injection and multiplication
dominates. This ratio determines also the sign of the
surface charge and of the remaining surface voltage when
the main voltage is switched off.
3. Experimental measurments of
electrical currents in insulation
3.1 Solid insulation: DC conduction
Figure 2 shows current curves for polarization-
depolarization experiments of alumina-filled epoxy
composite cast into an aluminum electrode system,
including guard-ring. This cast block geometry offers
several advantage respectively to the previously used
contacting system of conductive rubber on epoxy plates
[2]: namely i) the same metal/insulation contacting as
in the real insulator application ensures realistic charge
injection behavior at contact, ii) reduced quantity of void
at contact compared lead to lower partial discharge noise,
and ii) less influence of short-time temperature fluctuation
(increased thermal mass).
 Figure 3 Left: Simplified sketch of the emission current measurement system. The
inner conductor (emitter) radius is rin = 7 cm, the outer conductor (collector) radius is
rout = 17.8 cm, the length of the collector is l = 2 m, the length of the guards is 30 cm.
Right: Realized vessel, connected to the vacuum system composed by three turbo-
pumps. The system is capable of reaching vacuum with pressure < 0.08 Pa and a SF6
pressure of 0.45 MPa.
Figure 4 Left: Emission current at low electric fields, and
right, at higher fields, for negative polarity and smooth
(rms roughness Ra = 3±0.2 mm) emitting electrode both
for vacuum and 0.45 MPa SF6

As seen in Figure 2, in this case an estimation (within
20% error) of the DC current is
already achieved after one day (8.6·104 s) of measurement
time, while a steady state necessitates 11 days (106s) of
measurement. By repeated polarization/depolarization
measurement at all operating temperatures, the Arrhenius
behavior of the DC conductivity
is determined. The conductivity value varies over three
decades from 20°C to the maximum operating temperature
of 105°C [4]. As a consequence, an inhomogeneous
temperature distribution results in space charges in the
solid and field redistribution under DC compared to AC
stress.
3.2 Gas insulation
Although the physical mechanisms of charge injection
and multiplication are well known, the absolute number
of charges and the onset-fields can vary in orders of
magnitude. The biggest unknown in the system – which
can also not be described analytically – is the electrode
surface roughness. Therefore the amount of injected
charges for a technical relevant geometry was investigated
experimentally (see Ref. [12] for a description of the
experimental setup). In order to separate gas dominated
effects from electrode-dominated effects the experiments
were conducted using both vacuum and pressurized SF6.
Small electric fields were applied (< 2 kV/mm) in order
to detect the onset electric fields of charge injection and
multiplication.
Figure 3 shows a simplified sketch of the measurement
system: High voltage (positive and negative polarity, up
to 100 kV) is applied to the “injecting” electrode while
another electrode (collector) collects these charges. The
collector is connected to a current measurement system.
The size of the electrodes, i.e. the volume in between was
chosen in order to reach approximately 1-2 pA due to
natural ionization.
(11)
where V is the volume of the gas. With a value of =
37.5 IP/(s·cm3) [13], the current due to natural ionization
in the given volume (~ 0.168 m3) is ~2 pA. Although the
natural ionization current is dependent on the volume, it is
plotted as current density (A/m2) to represent an “injected
charge”. The surface of the inner conductor is ~1 m2,
therefore the absolute values of current and current density
approximately coincide. The onset for charge injection
can be located in a range between 0.3 and 0.7 kV/mm.
This value is much lower than values from literature [14,
15]. Emission currents at higher electric fields are shown
in the right panel of Figure 4, in both SF6 and vacuum, for
negative polarity.

Figure 4 Left: Emission current at low electric fields, and right, at higher fields, for negative polarity and smooth (rms roughness Ra = 3±0.2 μm) emitting electrode
both for vacuum and 0.45 MPa SF6.

Cigre Science & Engineering • N°7 February 2017

137
 Figure 5 a) Test set-up for the long-term measurement of surface potential on insulator-gas interfaces. System nominal voltage:
UDC = 350 kV DC. Max. load current = 4.6 kA. Potential of non-contact electrostatic probe: 0 < Up < 350 kV DC. b) Angle-
adjustable surface potential probe for surface scanning of DC insulators. c) Projection of insulator and measured azimuthal
angle ϑ (ϑ= 0°: vertical upward direction).

4. Surface potential measurement
Surface potential measurement are performed with
a modified non-contact electrostatic voltmeter from
TREK, Inc, using a miniature electrostatic field chopper
probe [16]. A positioning system for the probe allows
scanning of two insulator surfaces within a 320 kV
DC gas-insulated system (see Figure 5). Tested are DC
insulators under 0.45 MPa SF6 at relative humidity RH
≤ 8%. Measurements are performed at full load current,
resulting in an average 25°C thermal gradient across the
insulation3. To speed up the transition to DC state, the
minimum enclosure temperature is lifted up to 50°C.
After system thermalization, a voltage UDC = 100 kV is
applied on the central conductor and the surface potential
of the gas-insulator interface is measured periodically at
various azimuthal angles ϑ and radii r until steady state is
reached.
Figure 6a) shows the temporal evolution of the measured
and modeled surface potential change at an average radius
r = 120 mm. For modeling of the potential evolution,
DC conductivity values of the alumina-filled epoxy
insulator material as function of temperature and field are
determined independently by polarization experiments on
plates and cast blocks4. The only adjusted parameter in
the model is the injection/multiplication current (see
Figure 6b)). Effective values used are within the
same range as measured in section 3.2 and Ref. [14] for
comparable surface roughness (Ra, see Ref. [17]).
___________________________
3- Due to natural convection, the thermal distribution depend on the azimuthal angle. A separate measurement of the temperature distribution as function of load current
is performed prior to voltage switch-on by a network of thermocouples distributed on the insulator surface.
4- A direct scale-up of the fast polarization current obtained from thin samples (plates/cast blocks) to this real-size component is difficult and thus neglected here.
5 An exception is the upward direction (0°). In this location the natural convection is considerably higher than anywhere else, and the associated gas velocity field might
interfere with the field-driven ion drift.
The time-scale of the transition to the resistive state
(which depends mainly on the conductivity of the solid
insulation) is well-predicted by the model5. For early
times (t < 300 h), the measured potential increases in
most of the cases more rapidly than predicted, which is
to be expected as fast polarization currents are neglected.
The amplitude of the change in surface potential ∆U is
however controlled by charges in the gas: more charge
injection/multiplication reduce the final potential change,
as can be seen in Figure 6c). In this particular case,
considering natural ionization only, the expected increase
in surface potential would be ∆U=13 kV on the insulator
surface, corresponding to a steady-state injection current
= 0.35 pA/m2 into the insulator surface. In contrast, the
maximum experimentally observed ∆U(ϑ=−30°)= 8 kV,
corresponding to an additional injection current
10 pA/m2 from the metallic enclosure, corresponding to
= 2.2 pA/m2 onto the insulator surface.
These relatively low additional ionic currents reduce the
change in radial distribution between the capacitive and
the steady state DC electric field (see Figure 6c).
5. Flashover breakdown voltage
After proper experimental calibration of the DC simulation
model at nominal field values, its applicability should be
tested at near-breakdown conditions. For this purpose,
a setup for the determination of the flashover voltage
of cylindrical epoxy specimens subjected to a thermal

Cigre Science & Engineering • N°7 February 2017

138
 Figure 6 a) Measurement and modeling of surface potential change ΔU as a function of time and azimuthal angle (see b)) at an average radius of 120
mm on an alumina-filled epoxy partition insulator. Initial values of the potential are distributed on the y-axis for better readability. Natural ionization
rate = 50 IP/cm3/s, Epoxy DC conductivity from independent (de-)polarization experiments. b) Effective injection/multiplication
current as a function of absolute azimuthal angle used for the modeling, and compared to experimental emission currents from the literature. c)
Corresponding modeled surface potential and electric field (gas side) at the insulator surface (107 s = 115 d) for various injection currents (ϑ=−90°).
Voltage on conductor UDC = 100 kV, max. load current (4.6 kA)

gradient at 0.1 MPa SF6 pressure was built and operated scaling the flashover voltages Vbd,DC after long-term stress
at TU Dresden [18]. with VDC for (6, 48, 92) hours.

5.1 Capacitive state: cold and warm 5.2 Capacitive-resistive transition

Performing voltage rising tests (VRT), the dielectric
strength under electrostatic field conditions comparable to
AC stress was examined. The tests were executed in cold
condition without a temperature gradient and in warm
condition with the temperature distribution seen in Figure
7a). The breakdown strength is about 15 % lower in warm
than in cold condition, due to the locally seen lower gas
density in the vicinity of the heated electrode [19]. The
corresponding breakdown voltage Vbd,VRT,warm is used for
For suitable DC pre-stress times, the experiments were
carried out at a voltage VDC of 45 % of the breakdown
voltage of the AC-like voltage rising test:
(12)
Under cold conditions, the conductivity of epoxy is low
and the time of the field transition is in the order of weeks.
Hence, no decrease of the breakdown voltage occurred
after 48 h or 92 h (see Figure 8).

Figure 7 a) Equilibrium temperature distribution b) Calculation of capacitive-resistive transition after DC energizing at voltage VDC with
temperature distribution neglecting charge multiplication. (calculation 1 in Figure 8)

Cigre Science & Engineering • N°7 February 2017

139
 Figure 8 Left: symbols: experimental flashover voltage Vbd,DC as a function of the DC stress duration with voltage VDC. Lines: calculation 1: simulation
for no-charge emission from gas (solid conduction only, low homogeneous gas conductivity (10-21 S/m) ). Calculation 2: including additional charge
injection on insulator surface (see text). Right: Test execution to investigate time-dependence of flashover voltage during field transition to resistive
state. VRT: voltage rising test.

Under warm conditions, all long-term tests show a
decrease of the breakdown voltage in comparison to the
values of the voltage rising test and show the effects of the
capacitive-resistive transition. After 48h, the insulation
strength is on average about 10 % lower than in the VRT.
Due to the temperature-dependent conductivity of the
insulator material, the highest field strength is increasing
and its location is shifted to the cooler part of the insulator
into the gap near the high voltage electrode (Figure 7b).
5.3 Comparison with simulation
The simulation of the capacitive-resistive transition
reproduces the execution of the experimental test (see
Figure 8, right): Using i) Curie-von Schweidler approach
to simulate the polarization processes, and ii) the
independently measured DC conductivity in the epoxy
resin, the electric field is calculated for a DC voltage
application of various pre-stress durations. Subsequently,
a fast voltage rise is implemented in the FEM tool,
until reaching the breakdown field strength on the HV
electrode. Considering conduction in the gas, simulation
was carried out for two cases:
1. Calculation 1: residual homogenous conductivity of
10-21 S/m in the whole gas volume. This approximation
is justified, as the contribution from natural ionization
is negligible for this voltage level and geometry.
2. Calculation 2: as calc. 1, with additional field-dependent
injection current = of the model for the prediction of breakdown voltage as a
10 pA/m2) on the insulator interface subjected to high
field next to the HV electrode.
Comparing the calculated values with the measurements
(Figure 8a)), agreement is found for both cases for times
of DC stress up to 100 h. For longer times, calculation 1
predicts that flashover should happen after twelve days
under pure DC stress at VDC. However, two experiment
have been carried out at VDC for 14 days, respectively
27 days without breakdown. Subsequent VRT applied
show a moderate reduction of the minimum flashover
breakdown strength of 0.8 · Vbd,VRT,warm. This behavior
can be explained with calculation 2 by the presence of
low injection ionic currents from the neighboring HV
electrode. The magnitude of these currents is comparable
to those reported in Ref. [14].
6. Conclusion/outlook
This contribution presents a joint experimental/theoretical
approach for the determination of electric potential at the
gas-solid interface in HVDC gas-insulated systems. The
associated model enables prediction of the spatial and
temporal evolution of the electric field.
The input experimental parameters needed for
modeling are the transient and steady-state current-
voltage characteristics of the insulation media, i.e. the
temperature-, field- and time-dependent residual DC
currents in gas and solid insulation. These are determined
by independent measurements on model experimental
setups: for the solid, epoxy plates or cast blocks, for the
gas, a coaxial electrode geometry.
The time scale of the capacitive-resistive surface potential
transition is controlled by the DC conductivity of the
solid epoxy insulation. The timescale predicted by the
independently-calibrated model is in good agreement with
the measured temporal evolution of the surface potential.
This result is essential to derive reliable testing procedure
of DC gas-insulated system. Furthermore, the applicability
function of DC stress duration is demonstrated.
Further investigations should aim at completing the
characterization of charge injection and multiplication
in the gas, with e.g. the effects of paint, various
surface roughness, and gas humidity content. Effects
on conduction in the solid, such as long-term thermal
and electrical aging, as well as the scalability of “fast”
polarization currents to real-scale insulation components
deserve also further study.

Cigre Science & Engineering • N°7 February 2017

140
7. Bibliography
1. Pedersen, A., T. Christen, A. Blaszczyk, and H. Boehme. Streamer
inception and propagation models for designing air insulated
power devices. in Conference on Electrical Insulation and Dielectric
Phenomena. 2009.
2. Gremaud, R., F. Molitor, C.B. Doiron, T. Christen, U. Riechert, U.
Straumann, B. Källstrand, K. Johansson, and O. Hjortstam. Solid
Insulation in DC Gas-Insulated Systems. in Cigré Session 45. 2014.
Paris, France.
3. Riechert, U., U. Straumann, F. Blumenroth, and E. Sperling. Dielectric
Testing of Gas/Solid Insulation Systems for HVDC GIS/GIL. in SCD1
Colloquium Trends in Technology, Materials, Testing and Diagnostics
Applied to Electric Power Systems. 2015. Rio de Janeiro, Brazil.
4. Gremaud, R., F. Molitor, C. Doiron, A. Krivda, T. Christen, K.
Johansson, N. Lavesson, U. Riechert, and U. Straumann. Solid-
gas interfaces in DC gas insulated systems. in 4. ETG-Fachtagung
Grenzflächen in elektrischen Isoliersystemen. 2013. Dresden,
Germany.
5. Lavesson, N. and C.B. Doiron. Hybrid Resistive-Capacitive and Ion
Drift Model for Solid Gas Dielectrics. in COMSOL conference 2015.
2015. Grenoble, France.
6. Johansson, K., L. Walfridsson, U. Gafvert, B. Källstrand, and S.
Hörnfeldt. Probe for measurements of the DC electric field in air
around high voltage apparatus. in Electrical Insulation and Dielectric
Phenomena (CEIDP), 2010 Annual Report Conference on. 2010.
7. Gremaud, R., M. Schueller, C.B. Doiron, U. Riechert, U. Straumann,
and C.M. Franck. Experimental validation of electric field modeling
in DC gas-insulated system. in SCD1 Colloquium Trends in
Technology, Materials, Testing and Diagnostics Applied to Electric
Power Systems. 2015. Rio de Janeiro, Brazil.
8. Ming, L., M. Leijon, A. Hjortsberg, and T. Bengtsson. Surface
charging of particle-contaminated spacer in SF6 under DC stress. in
Xth international conference on gas discharges and their applications.
1992. Swansea, UK.
9. Straumann, U., M. Schueller, and C.M. Franck, Theoretical
investigation of HVDC disc spacer charging in SF6 gas insulated
systems. Dielectrics and Electrical Insulation, IEEE Transactions on,
2012. 19(6): p. 2196.
10. Morrow, R., A Survey of the Electron and Ion Transport Properties of
SF6. Plasma Science, IEEE Transactions on, 1986. 14(3): p. 234-239.
11. Kindersberger, J. and C. Lederle, Surface charge decay on insulators
in air and sulfurhexafluorid - part I: simulation. Dielectrics and
Electrical Insulation, IEEE Transactions on, 2008. 15(4): p. 941-948.
12. Simka, P., V. Teppati, and M. Vukas. Charge injection and
multiplication in SF6 and vacuum under DC Electric Fields. in
ICPADM 2015. 2015. Sydney, Australia.
13. Koch, M., U. Straumann, and C.M. Franck. Determination of waiting
times between successive breakdown experiments. in Conference on
Electrical Insulation and Dielectric Phenomena (CEIDP). 2012.
14. Zavattoni, L., Conduction phenomena through gas and insulating
solids in HVDC Gas Insulated Substations, and consequences on
electric field distribution. 2015, Université de Grenoble: Grenoble.
15. Kuffel, J., E. Kuffel, and W.S. Zaengl, High voltage engineering
fundamentals. 2nd Edition ed. 2000: Newnes.
Cigre Science & Engineering • N°7 February 2017
141
16. TREK. TREK Model 541A Non-Contacting Electrostatic Voltmeter
for EOS/ESD. 2013; Available from: http://www.trekinc.com/
pdf/541A_Sales.pdf.
17. Schueller, M., R. Gremaud, C.B. Doiron, and C.M. Franck, Micro
discharges in HVDC gas insulated systems. IEEE Trans. Diel.
Electrical Insulation, 2015. 22: p. 2879.
18. Hering, M., J. Speck, K. Backhaus, S. Großmann, and U. Riechert.
Capacitive-resistive transition in gas insulated DC systems under
the influence of particles on the insulator surface. in The 19th
International Symposium on High Voltage Engineering. 2015. Pilsen,
Czech Republic.
19. Hering, M., J. Speck, S. Großmann, and U. Riechert. Investigation
of the Temperature Influence on the Breakdown Voltage in Gas
Insulated Systems under DC Voltage Stress. in The 18th International
Symposium on High Voltage Engineering (ISH). 2013. Seoul, Korea.
8. Biographies
Robin Gremaud received his diploma in physics from
the University of Fribourg, Switzerland, in 2003, and his
Ph.D. degree in condensed matter physics from the VU
University Amsterdam, Netherlands, in 2008. After a
postdoctoral position during 2008-2010 at the Research
Institute for Materials Science EMPA, Switzerland, he
joined the Swiss ABB corporate research center in 2010
and is currently Principal Scientist in the “Materials,
Design and Testing” group. His current research addresses
the physics of solid and gas insulation under HVDC stress.
Charles B. Doiron received his B.Sc and his M.Sc.
degrees in physics from the University of Sherbrooke,
Canada, in 2002 and 2005, respectively, and his Ph.D.
degree in theoretical physics from the University of Basel,
Switzerland, in 2009. He joined the Swiss corporate
research center of ABB in 2009, where he currently
leads the Theoretical Physics group. His current research
interest include gaseous insulation, alternatives to SF6,
gas discharges, as well as high-temperature properties of
gases.
Matthias Baur received his diploma in physics in 2007
and later his Ph.D. degree in physics in 2012 from ETH
Zurich, Switzerland. After a short postdoctoral position in
2013 at the University of Queensland, Australia, he joined
the Swiss ABB corporate research center in 2014, where
he is now working as a Scientist in the Materials, Design
and Testing group. His current research activities focus
on the design of power capacitors and the physics of high
voltage solid insulation materials.
Philipp Simka received the M.Sc. degree from the
Department of Electrical Engineering and Information
Technology, ETH Zurich, Switzerland in 2005 and Ph.D
from the Institute for Power Systems and High Voltage
Technology, ETH Zurich, Switzerland in 2011. Since then
he is with the Swiss corporate research center of ABB. His
research interests include gaseous insulation under AC
and DC, high voltage test techniques and high frequency
phenomena associated with high voltage equipment.
Valeria Teppati received her M.Sc. degree in
Electrical Engineering and Ph.D. degree in Electronic
Instrumentation from Politecnico di Torino, Italy, in
1999 and 2003 respectively. From 2003 to 2014 she was
Assistant Professor at Politecnico di Torino, her main
research interests being microwave measurements of
active and multiport devices. From 2012 to 2014 she
was Visiting Professor at ETH Zürich developing a 100
GHz active load-pull measurement system. In 2014 she
joined the Swiss ABB Corporate Research Center in the
gas circuit breakers group. Her current research activities
include gas insulation, alternatives to SF6 and gas circuit
breaker CFD simulations.
Birgitta Källstrand received her BSc degree in
meteorology from Uppsala University, Sweden, in 1992
and her PhD degree in meteorology from the same
university in 1998. She joined the Swedish corporate
research center of ABB in 2000. Main focus of her research
during the years with ABB has been on HVDC cable
insulation and processing, and electrical measurement
techniques for solid and gaseous insulation materials. She
is now team manager for the High Voltage physics team at
ABB Corporate Research, in Västerås, Sweden.
Kenneth Johansson received a technical college graduate
in electronics in 1986. He started to work at ASEA
Research and Innovation (later ABB Corporate Research)
in Västerås in 1986 as a laboratory engineer with measuring
techniques for dielectric materials. He develops scientific
measuring equipment for dielectric and high voltage
studies. Main focus areas have been measurements of very
low currents, electric field measurements in gas, fluid and
solid insulation materials and high frequency behaviors of
power products. He holds the position Senior Engineer.
Maria Hering finished her studies in electrical engineering
at the Technische Universität Dresden (TUD) in 2011
with the diploma degree (Dipl.-Ing.) and received the
Ph.D. degree from the TUD on the topic of gas-insulated
DC systems in 2016. Maria Hering is Young Member of
CIGRÉ and contributes to different Working Groups of
Cigre Science & Engineering • N°7 February 2017
142
the Scientific Committee D1. She is also a member of
the German Power Engineering Society, department for
Materials, Electrical Insulations and Diagnostics. Maria
Hering received the Young Researcher Award of the
International Symposium on High Voltage Engineering in
2013 (Seoul) and 2015 (Pilsen).
Joachim Speck was born in Dresden, Germany, in 1951.
He received the Ph.D. degree in electrical engineering from
the Technische Universität Dresden, Germany, in 1978.
He has been with the Institute for High Voltage and High
Current Engineering, Technische Universität Dresden
since 1974. His research interests include field calculation,
statistical evaluation, SF6 insulation and breakdown of
solid dielectrics.
Steffen Großmann was born in Dresden, Germany,
in 1954. He received the Ph.D. degree in electrical
engineering from the Technische Universität Dresden,
Germany, in 1988. From 1976 to 2003 he was with
Starkstrom-Anlagenbau Dresden and RIBE Electrical
Fittings Radebeul and Schwabach. Since 2003 he is
Professor for High Voltage and High Current Engineering,
Technische Universität Dresden. His research interests
include high voltage insulation systems, ampacity and
heating of electrical power devices and electrical contacts
and connections.
Uwe Riechert finished the studies in electrical engineering
at the Dresden Technical University (TUD) in 1994 and
received the Ph.D. degree at the TUD on the topic of
polymeric insulated HVDC cables in 2001. Since 1999 he
is with ABB Switzerland. He conducted several product
development projects. In 2013, he became a principal
manager. Since 2011 Dr.-Ing. Uwe Riechert is working as
project manager for HVDC substations. Uwe Riechert is
a member of DKE, CES and different CIGRE and IEC
working groups.
Ueli Straumann received his diploma in theoretical
physics from the University of Zurich, Switzerland, in
2001 and his Ph.D. degree in electrical engineering from
ETH Zurich, Zurich, Switzerland, in 2007. Since then, he
has been a Lecturer of High Voltage Engineering, ETH
Zurich. After being Senior Assistant at the High Voltage
Laboratory of ETH Zurich between 2007 and 2012, he
became Senior and Principal Engineer at the GIS R&D of
ABB Switzerland Ltd.
Classification of customers based on
temporal load profile patterns
I. Benítez*, Instituto Tecnológico de la Energía, Spain
A. Quijano, Instituto Tecnológico de la Energía, Pain
I. Delgado, Instituto Tecnológico de la Energía, Spain
J.L. Díez, Universitat Politècnica de València, Spain
Summary
The deployment of Advanced Metering Infrastructure
(AMI) is providing to utilities large amounts of energy
consumption data from their customers, in form of
daily load profiles with energy consumed per hour or
a smaller period. These data can yield valuable results
when analyzed, in order to extract useful knowledge
about the typical patterns of consumption of energy from
the customers. The proper mechanisms and tools have to
be developed and implemented for this objective.
Big Data and Big Data Analytics systems will
contribute to analyze this information and help to
extract knowledge from the data, summarized in form
of patterns or other mining knowledge, that will aid
experts in decision support. In the present work a
classification of customers based on their temporal
load profiles is proposed. This classification procedure
could be implemented in the current Big Data
Analytics software systems, providing an added value
to their statistical analysis options. Previous works in
the literature present algorithms that allow to classify
load profiles from customers by processing batch
datasets and obtaining static patterns of load profiles.
The proposed technique allows to analyze patterns not
only in shape but also in their evolution or trend of
energy consumption at each hour of the day through
time. Specific quantitative indicators that characterize
the patterns (and the consumers associated to them) are
described and tested for this purpose.
1. Introduction
The current deployment of Information and Communication
Technologies (ICT) to manage the electrical transmission
*ignacio.benitez@ite.es
KEYWORDS
Load profiles; Dynamic clustering; Pattern recognition; Classification.
Cigre Science & Engineering • N°7 February 2017
143
and distribution grid, and the integration of the Advanced
Metering Infrastructure (AMI) and Smart Metering from
a growing number of customer premises, has brought to
light the necessity of the development of adequate systems
to gather, process and store huge amounts of data. For
instance, one single smart meter can collect 96 quarter-
hourly measures of active energy per day, 365 days per
year, which yields an amount of 35040 values of energy
per year, only for one client.
Utilities are solving this challenge by the implementation
of Big Data systems and Big Data Analytics software
tools, which are specifically designed to manage big
amounts of data [1]. Big Data systems are built on new
architecture paradigms such as cloud computing, and
database systems such as NOSQL (Not Only Structured
Query Language), which support the management of
unstructured data. Big Data Analytics systems rely on
the Big Data systems to gather, process and analyze the
data, which can be obtained from distributed, scalable
resources, or from the cloud, either in real time (i.e.
“Stream” data) or off-line, even though stream mining
systems are still in an initial phase of development [2].
Big Data Analytics systems, as indicated in [3], include
three main functionalities:
• Data visualization, in form of charts, maps and new
visualization designs that help to understand the
results of the Big Data analysis.
• Statistical analysis, in form of summaries that describe
collections of data, or inferential statistical analysis,
that can be used to draw inferences about the process.
• Data mining, described as the computational process
of discovering knowledge in large data sets. Usual
algorithms for the processing of large data sets include
 classification, clustering, regression, statistical
learning, association analysis and link mining.
The present work describes a classification system of
customers based on their temporal load profiles, i.e.,
considering the daily load profiles through a sequence of
consecutive days. In the data set analyzed, this period is
of one year, but any other time frame could be considered
(the previous week, for instance, or the past season). The
development presented allows first to obtain patterns that
summarize the energy consumption habits of a group of
customers, and then allows to classify new customers to
one of the previous patterns obtained by similarity in the
energy consumption load profile. With this approach, the
customers are classified not only by how they consume
energy during the day, but also how they consume energy
through a sequence of days. Each consumer is assigned
to one of the resulting temporal patterns. The analysis
allows the definition of quantitative indicators that
characterize the patterns (and the consumers associated
to them) from two perspectives:
• Load profiles (shape): indicators such as maximum this work is a two-step process. First, all the shapes
daily consumption or peak/valley relation can be
defined and computed.
• Trends (time): specific indicators are proposed to surfaces and the vertices is predefined, based on the
measure the trend of the pattern through time (e.g.
increasing, decreasing, flat), and the amount of
variations between sequential days, as some sort of
dynamics measure.
Obtaining these indicators allows to globally describe the
consumption of a number of consumers in a given period
of time at a glance: the visualization of the patterns, such
as the example of Figure 1, will allow an experienced
user to clearly identify the common patterns of energy
consumption, with the higher number of consumers
assigned to them, and to quickly identify unusual patterns
with unexpected behaviours or with fewer number of
consumers assigned to them, and to think about the
possible reasons behind this behaviour. Besides, the
computation of the previously commented indicators
will allow a quantitative characterization of the patterns,
which can be used as an input for a classification system.
This system will classify and identify consumers given
some specific criteria for the classification, which must
be defined in form of rules or threshold values for the
indicators and the number of consumers at each pattern.
Cigre Science & Engineering • N°7 February 2017
144
2. Segmentation of customers
based on temporal load profiles
A dynamic clustering technique is applied on the
data set of customers’ load profiles. This algorithm
is called END-KMH (Equal N-Dimensional K-means
Hausdorff-based), and has been designed for the
clustering of time series data with n dimensions or
characteristics, such as load profiles time series (24 in
this case).
The energy consumption profiles from the customers
are seen as 3D shapes, defined by the 24 hours load
profile, where each hour is considered as a feature
or dimension of the data, with a length equal to the
number of days considered in the analysis (one year in
the present work). The time series clustering algorithm
needs to define a measure of similarity between each
client’s profile and all the resulting centroids at each
iteration. The Euclidean distance is usually used
for this in most works. The proposed solution in
are decomposed in a number of linear surfaces, by
applying least squares regression. The number of
expert’s knowledge of the typical behavior from
residential users regarding energy consumption. Then,
the resulting surfaces are compared by computing the
Hausdorff distance [4] between them, and a global
similarity value is obtained, given by the average value
of all the Hausdorff distances between the different
surfaces. A more detailed description of this algorithm
can be found in [5], where the same algorithm has
been applied with the objective to identify uncommon
patterns for reliability and maintenance purposes.
For the present work, a data set has been selected to
serve as an example of the capabilities of the analysis
proposed. This data set is the “SmartMeter Energy
Consumption Data in London Households”, provided
by the Greater London Authority through the London
Data Store web site. This data set gathers half-hour
energy consumption measures from a sample of 5567
customers between years 2012 and 2014. For this
analysis, the hourly energy consumption values have
been processed from the first 2000 households through
year 2013 (therefore 365 days x 2000 households =
 Figure 1. Example of six temporal clusters (and patterns) obtained from the first 2000 households of the Smart Meter Energy
Consumption dataset for year 2013, provided by the Greater London Authority.

730000 load profiles being analyzed). Figure 1 depicts
the patterns obtained after applying the dynamic
clustering algorithm previously described. The number
of clusters to be found is a parameter that must be
defined. A number of 6 clusters to be found has been
chosen, based on previous experiences with the data.
Observing the rasulting patterns in Figure 1, some
quick conclusions can be derived. It can be observed,
for instance, that there are four clusters which gather the
majority of households (clusters 1 to 4), and there are
two clusters with probably unusual energy consumption
patterns and a more reduced number of customers or
households belonging to them. These are clusters 5 and
6, with 54 and 11 customers respectively. Moreover,
observing clusters 1 to 4, another conclusion could
also be obtained: that the shapes of the four clusters are
very similar, and their main difference is in the average
energy consumption per day, ranging from a maximum
peak of around 0.3 kWh in households from cluster 3,
to a maximum peak of energy consumption of around
2 kWh in households from cluster 4. The seasonality
effect can also be observed in the six clusters obtained,
with an elevated rate of energy consumption in Winter
(around January and February months).
Time series clustering allows, therefore, obtaining
patterns that evolve through time, in a time frame
defined by the expert. This allows an interpretation of
the results that depicts the full dynamic behavior of all
the objects, allowing a much more complete (and also
complex) global view of the resulting temporal patterns.
Following, the resulting patterns are characterized
by means of specific quantitative indices that allow
to describe the resulting patterns in their evolution in
shape and through time.

Cigre Science & Engineering • N°7 February 2017

145
3. Definition of indices to daily values at the same hour of the day, to a lineal
model with two coefficients, being the slope and
characterize temporal load profiles an independent variable. The regression model is
In order to characterize the resulting patterns of temporal depicted in black color in the example of Figure 2 (b).
load profiles, the following qauntitative indices have
Daily evolution: RMS through the year. This index gives
been described:
information regarding the amplitude of the daily energy
• Shape analysis: daily average energy consumption. consumption values through the year. It is computed
This index is computed as the average value of all the as the Root Mean Sqaure (RMS) of all the energy
average values of energy consumption at each hour of consumption values. The result can be observed in blue
the day through the year. These are the values that can colour, in Figure 2 (b).
be seen depiected in the example of Figure 2 (a).
• Shape analysis: Relation between peak and valley Other indices could be defined, for instance a new
energy consumption. This index is computed as the index to describe the seasonality variation. This will be
difference between the maximum value of hourly a subject for further developments. Table I displays the
energy consumption from the whole year data and the resulting indices from the patterns previously obtained.
minimum value of hourly energy consumption. As can be seen, based on the analysis of the resulting
• Daily evolution: consumption trend through the year. patterns, the six clusters, and the customers that belong
A simple linear regression model is obtained, by to them, are characterized quantitatively in terms of how
applying the least squares algorithm, to fit the data they consume the energy during the day, and also their
of energy consumption, arranged as sequences of hour by hour dynamic evolution through the year.

Shape Daily evolution

Cluster No. of customers Trend through year RMS through year
Daily average (kWh) Peak – valley (kWh) (slope) (kWh)
1 823 0.414 0.718 -0.00015 0.439
2 585 0.214 0.369 -0.00007 0.226
3 237 0.123 0.258 -0.00006 0.130
4 290 0.897 1.789 -0.00057 0.960
5 54 0.064 0.283 -0.00009 0.069
6 11 0.022 0.480 0.00014 0.049
Table I. Clusters and quantitative indices computed from the patterns.

Figure 2. Graphical representation of the quantitative indices defined to describe the temporal patterns obtained,
in terms of shape analysis (a) and the evolution of the load profile through the days (b). Cluster 1.

Cigre Science & Engineering • N°7 February 2017

146
The next step of the analysis is to classify new customers
on the clusters previously obtained, based on similarity to
one of the six patterns detected. This step is described next.
4. Classification of new customers
Whenever time series load profiles from new customers
are available, these can be compared with the six
patterns previously obtained, and the new customer
can be assigned to the most similar one, based on a
measure of mathematical similarity. This classification
will have a certain degree of error, that can be evaluated
by computing the same quantitative indices for the
custome’s load profile, and comparing the values with the
assigned pattern ones. Global error indices, such as the
MAE (Mean Absolute Error) can be computed for each
pattern based on the classification of new customers.
As a validation example, a test has been developed
with the six patterns obtained from the first 2000
households of the Greater London Authority data set.
A set of new 1000 households from the same database
has been classified to one of the six patterns. A modified
Euclidean distance has been used to compare each
customer’s time series load profile with each of the six
patterns. The computation of this similarity measure is
shown in Equation (1), where n is the number of features
or characteristics of the data (24 hours in this case), and
B is the norm. Since the Euclidean distance is used, the
identity matrix has been applied as the norm.
The 1000 new households are classified to the pattern
with the highest similarity, as computed by the distance
No. of new cus- Shape
Cluster tomers classified MAE Daily average MAE Peak – valley
1 399 0.099
2 309 0.055
3 125 0.030
4 132 0.210
5 32 0.026
6 3 0.021
Table II. Results of classification of new customers and classification errors per index.
Cigre Science & Engineering • N°7 February 2017
in Equation (1). Following, the quantitative indices are
computed for all the customers time series load profiles,
and a global error measure in the classification is
computed for each pattern, as the mean of the difference
between the indices from the customers assigned to the
pattern and the pattern itself. This error is computed as
absolute (i.e., the MAE) in all the indices except the
slope or trend. Table II shows the results obtained.
Comparing these results with the values of the indices
in Table I it can be observed that the errors obtained in
the classification are low in most of the indices, for all
the six clusters. The only exception could be the peak –
valley relation, which is high in all the cases, probably
due to the disparity in energy consumption time series
profiles from the customers, compared to the averaged
patterns of consumption.
5. Conclusions
The classification system presented can automatically
process the temporal patterns from a given time frame,
and summarize a big number of load profiles from
customers in summarized groups of consumers with a
specific behaviour in shape and time. Applications of this
system can be diverse: since the selection of customers
for demand side management purposes, to the design of
dynamic or temporal maps of energy consumption that
take into account the computed trends in increasing or
decreasing energy consumption.
The definition of quantitative indices computed from the
resulting patterns allow to characterize these patterns,
and the customers associated to them, in the shape of
their load profiles and the variation of these load profiles
Daily evolution
Error in trend through year MAE RMS through year
2.996 -2.29 x 10 -05
0.205
2.221 2.88 x 10-05 0.135
1.722 2.26 x 10 -05
0.094
4.038 9.79 x 10-05 0.292
1.413 -2.64 x 10 -05
0.084
0.333 1.52 x 10-04 0.043
147
through the days. From the temporal patterns obtained,
new customers can be classified to them, according to
mathematical similarity. The classification errors can
be computed from the defined quantitative indices,
allowing to validate the process.
6. Bibliography
[1] C. P. Chen and C.Y. Zhang “Data-intensive applications, challenges,
techniques and technologies: A survey on Big Data” (Information
Sciences , 2014, vol. 275, pages 314-347)
[2] K. Kambatla, G. Kollias, V. Kumar and A. Grama “Trends in Big
Data analytics” (Journal of Parallel and Distributed Computing ,
2014, vol. 74, pages 2561-2573)
[3] H. Hu, Y. Wen, T.S. Chua and X. Li “Toward Scalable Systems for
Big Data Analytics: A Technology Tutorial” (Access, IEEE, 2014, vol.
2, pages 652-687)
[4] F. Hausdorff “Grundzüge der Mengenlehre” (Veit and Company,
1914)
[5] I. Benítez, A. Quijano, J.L. Díez and I. Delgado “Clustering of Time
Series Load Profiles for Grid Reliability” (CIGRE International
Conference on Condition Monitoring, Diagnosis and Maintenance
- CMDM 2015)
7. Biographies
Ignacio Benítez
PhD in Automation and Industrial Computer Science
from the Universitat Politècnica de València in 2016.
Automation and Industrial Electronics engineer from
the Universitat Politècnica de València. In October
2008 he joined the Instituto Tecnológico de la Energía
as R+D+i technician. He is currently part of the Smart
Grids department, where he develops projects related
to sustainable mobility, analyzing the impact of the
integration of the EV in the low and medium voltage
grid. Previously, he was responsible for the Coordination
and Development of actions undertaken in the area
of Advanced Control Techniques, addressed to the
application of control methodologies and knowledge
aimed at achieving goals related to energy efficiency
and integration of renewable energy resources. He has
proven experience in robotics, artificial intelligence,
machine learning, control systems engineering, fuzzy
logic, prediction models, neural networks, data mining
and pattern recognition techniques (clustering).
Alfredo Quijano
Alfredo Quijano was born in 1960 in Valencia, Spain.
Cigre Science & Engineering • N°7 February 2017
148
He received the Electrical Engineer degree and the Ph.D.
degree from Universitat Politècnica de València, in 1986
and 1992, respectively. He is the CEO of the Instituto
Tecnológico de la Energía – Universitat Politècnica de
València and Vicepresident of ITE. He is also a teacher
and researcher at the Universitat Politècnica de València
in the Electrical Engineering Department. His current
research activity is focused on applied research for the
Energy area and the electrical technology including
renewable energies, high voltage, metrology, new
materials and applications and research results transfer
to companies.
Ignacio Delgado
Mr. Ignacio Delgado is Industrial Engineer with over six
years of industry experience. From May 2006 to August
2016, he was part of the ITE team, developing his
work as head of the department of Smart Grids of ITE,
working at management and development of national
and European projects on smart grids, Renewable Energy
Integration network, Distributed Generation, Demand
Management and Automation and Remote management
of the network.
In recent years his research has focused on projects on
the active management of demand in the domestic sector,
analyzing and studying the functional requirements to
integrate systems on networks to provide all of greater
intelligence to manage and behavior as the real-time
user consumption. He has recently moved to Glasgow,
to work in the Centre of Excellence of Scottish Power
Energy Retail.
José Luis Díez
José Luis Díez received a M.Sc degree in Industrial
Engineering in 1995, and the Ph.D. degree in Control
Engineering in 2003, both from the Universitat
Politècnica de Valencia, Spain. He is currently an
Associate Professor and he has been teaching since 1995
at the Systems Engineering and Control Department,
Universitat Politècnica de Valencia, in a wide range of
subjects in the area such as automation, linear systems
control theory, digital signal processing, and intelligent
systems. His research interests include complex systems
modelling and identification (biomedical, biological,
energy and social systems), data mining, clustering
techniques, intelligent control, and control education.
Using the flow-based method for network
investments’ evaluation – a new concept
Nikoleta Kandalepa, Dr. Jasper van Casteren, Dr. Susana Almeida de Graaff*, Danny Klaar
TenneT TSO B.V.
Arnhem, The Netherlands
Abstract
Nowadays transmission network planning approaches
are not aligned with the operational Flow-Based (FB)
capacity calculation method. This bears the risk that the
evaluation of an investment with nowadays planning
methods might have as result a different solution than the
calculation with the FB market coupling algorithm, which
may result in under-investments or over-investments. An
innovative and understandable concept was developed
to bridge the gap between network planning and system
operations, leading towards a more integrated network
planning solution. The flow-based for investments
concept uses the principles of the operational flow-based
method for investment evaluation, assessing the impact
of investment proposals on the zone-to-zone available
capacity and considering interdependencies between all
cross-border zones. Network investments are in this way
evaluated in a transparent and complete way throughout
the whole common grid.
Introduction
The Flow-Based (FB) method is the current capacity
calculation and allocation method in central western
Europe (CWE: Germany, France, the Netherlands and
Belgium) and was successfully run for the first time in the
day-ahead market process at May 21st, 2015 [1]. The FB
method considers the inter-dependencies between cross-
border interconnections and the physical limits imposed
by transmission networks. Through the application of the
laws of physics together with an economic social welfare
optimization function, this method calculates zone-to-
zone transmission capacities, allowing a secure and
efficient use of the transmission network.
*susana.de.graaff@tennet.eu
KEYWORDS
Flow-Based method, network investments evaluation, network sensitivities, optimization, zone-to-zone transmission
capacity
Cigre Science & Engineering • N°7 February 2017
149
Apart from the daily capacity calculation, the FB method
may also be used to evaluate network investments such
as new network infrastructures and generating facilities.
Nowadays network planning approaches are not aligned
with the FB capacity calculation and allocation method,
but with Grid Transfer Capacity (GTC) calculations [2,3].
The evaluation of an investment with today’s planning
methods will for that reason result in different estimated
transmission capacity than the one calculated by the FB
market coupling algorithm. This bears the risk for under-
investments or over-investments.
In order to bridge the gap between planning and operation of
the transmission network, TenneT TSO B.V. is developing
a concept where the principles of the operational FB
method are used for planning (long-term, short-term and
operational) and investment evaluation. This concept,
an innovative mix of existing and new insights, aims to
complement today’s planning methods and to analyse
investments in a transparent and integral way. The effect
of an investment throughout the whole common grid is
monitored and the impact on the zone-to-zone available
transmission capacity is assessed. Any type of investment
can be evaluated and linked to the estimated zone-to-zone
capacity, representing what becomes available every day
to the day-ahead electricity market.
This paper describes the FB method in section 2 and
presents the proposed FB for investments concept in
section 3. First, the concept is presented from a high level
perspective and then each subsection discusses every
step of the process in more detail. Section 4 presents and
discusses simulation results for the FB for investments
concept and Section 5 draws conclusions and suggests
future developments.
 Figure 1: Simplified flow-based process representation
2. Flow-Based Method
The FB method is applied two days before the execution
day to calculate the day-ahead zone-to-zone transmission
capacities. The resulting FB domain represents the
maximum available capacity for market transactions in the
day ahead market [4].
Every day, each participating TSO creates its individual
network model (IGM) based on the best available forecast
for the execution day regarding vertical load, grid topology,
renewable infeed and available production. Two days ahead
of the execution day, these models are provided to the FB
computation, together with files containing the generation
shift keys (GSKs) and the combinations of critical branches
(CBs) and critical outages (COs). The Generation Shift
Key (GSK) defines how a change in the net position (NP)
of a bidding zone is mapped to the generating units of that
zone [5].
Critical Branches (CBs) are a subset of all network
elements that are considered significantly impacted by
zone-to-zone exchanges. These branches are taken into
account in the capacity calculation and, consequently, in
the market allocation process. The CBCO set is built upon
the critical branches, taking into account both N and N-1
situation by adding the combinations of a critical branch
with a critical outage (CO). Therefore a single CBCO may
be a single CB or the combination of a CB with a CO,
when this combination is considered relevant to the zone-
to-zone market capacity. The CBCO set thus represents
the monitored network constraints from a zone-to-zone
capacity perspective.
Figure 1 shows a simplified representation of the flow-based
Figure 2: Illustration of a two-dimensional Flow-Based domain [6]
Cigre Science & Engineering • N°7 February 2017
150
calculation process. The IGMs of the TSOs are merged into
one network file, which contains the common grid model
(CGM) with the best forecast for one hour of the execution
day. Based on a DC load flow computation, the FB method
computes the power transfer distribution factors (PTDFs)
and the remaining available margin (RAM) for the CBCO
set. These constitute the so-called FB parameters.
Each RAM equals the remaining flow capacity of a single
CBCO. The zone-to-zone PTDFs (as defined in section
3.1.3) equal the change in RAM caused by a variation in
zone-to-zone power exchange for each combination of two
specific zones.
The RAM of a CBCO is the maximum capacity of the
branch (PmaxCB) minus the reference flow of the CBCO
(FrefCBCO):
Equation (1) describes the RAM as used in the FB for
investments concept.
During day-ahead capacity calculations, non-costly
remedial actions (RA) may be applied in a coordinated
way, such as changing the tap position of phase shifting
transformers (PST) or topological actions.
Figure 2 shows an example of a FB domain that is defined
by the FB parameters. The FB domain determines the
market constraining CBCOs and the maximum allowed
zone-to-zone exchange. The FB domain is impacted by
both changes in RAM and PTDF of each CBCO and
these are normally both altered by investment proposals.
Therefore, the evaluation of an investment proposal needs
always to address both the impact on RAMs and zone-to-
zone PTDFs for all CBCOs.
3. Flow-Based for Investments
concept
The FB for investments concept uses the FB method
to evaluate investment proposals. An overview of the
proposed evaluation concept is given in figure 3.
Figure 3: FB for investments concept
The first step in figure 3 is to import the network
models that contain the different scenarios for analysis.
Operational network models and operational GSKs, as
for day-ahead FB capacity calculation within CWE, have
been used during the development of the method and to
prove the concept, but future network planning models
representing several types of scenarios would have to
be used for investment analyses. These initial network
models constitute the base cases. They are analysed one
by one, independently. The second step is to add the
proposed investment to the base case, creating a new
network model, the so-called investment case. Afterwards
the computation of the network sensitivities follows for
both models (i.e. base case and investment case): the
nodal power transfer distribution factors (PTDFs), the line
Cigre Science & Engineering • N°7 February 2017
151
outage distribution factors (LODFs) and the phase shifter
distribution factors (PSDFs).
Subsequently, the CBCO set is selected based on a set
of criteria that are presented in section 3.2. The CBCO
selection is made simultaneously for the base case and the
investment case, in order to create a common list that can
be used for comparing results. Together with the network
model, the corresponding GSK values are imported as
well. By using the network sensitivities and the GSK
values, the flow-based parameters and the capacity
indicators are calculated for every selected CBCO for
both network models.
The phase shifting transformers are kept at their neutral
tap positions up to this step. Optionally, using the network
sensitivities, a security-constrained optimization of the
PST tap positions is performed, maximizing the available
transmission capacity in the various zone-to-zone
directions. The PST optimization is performed separately
for the base and investment case and for each specific
zone-to-zone direction.
Finally, the results for the investment case are compared
to the base case results both with the PST taps at their
neutral positions, and with the optimized tap positions.
The steps of the process in figure 3 are further discussed
in the following sections.
3.1 Network Sensitivities
The FB for investments, in the same way as FB method,
uses a linear approach to detect changes in branch flows
in a fast way, by calculating distribution factors. These
distribution factors are computed individually for both the
base case and the investment case.
3.1.1 Nodal PTDFs
The nodal Power Transfer Distribution Factors (PTDFs)
are computed directly from the network topology. The
nodal PTDFCB,n equals the change in active power flow
in a critical branch, ΔPCB, by a change in active power
injection at node n, ΔPn, and an opposite change at the
reference node, so that:
Where is the initial (reference) power flow in the CB,
and PCB is the changed power flow.
The matrix of nodal PTDFs is computed by using the
susceptance matrix (Bb) and the connectivity matrix (C) [7]:
The PTDF matrix has a row for each branch in the network
and a column for each node. The calculation of the PTDF
matrix does not require the inversion of the susceptance
matrix, but is more efficiently calculated by solving the
linear equations.
3.1.2 Zonal PTDFs
The nodal PTDF can be aggregated into zonal PTDF by
adding the effect of the MW power shifts of the individual
generators in one zone. The “Generator Shift Key” (GSK)
is the participation factor of a single generator in the MW
change of the net position of a whole zone. A zone is
therefore defined by the set of GSKz for which
=1, and the zonal PTDFCB,z is calculated as:
The flow-based market coupling in the CWE region has
four specific zones: Belgium, Germany, Netherlands and
France, but any definition of zones or GSK can be used.
The zonal PTDF of a CB shows how a change in the net
position of this zone influences the power flow of
this CB (PCB):
Any change in the net position of only one zone
is balanced by an opposite change at the slack node. A
different slack node will thus produce different zonal
PTDFs.
Zonal PTDFs for all zones are computed for every CB.
For the four CWE countries in the FB market coupling,
four zonal PTDFs per CB are thus computed.
3.1.3 Zone-to-zone PTDFs
By subtracting the zonal PTDFs from 2 specific zones,
the corresponding zone-to-zone PTDFs are calculated.
The zone-to-zone PTDF of a CB (PTDFCB,z2z) shows how
a change in the bilateral exchange between these zones
influences the power flow of this CB:
A balanced exchange between two zones does
Cigre Science & Engineering • N°7 February 2017
152
not lead to a change at the slack node. The dependency on
the slack node is cancelled out in the zone-to-zone PTDF.
Therefore the PTDFCB,z2z does not depend on the choice
of the slack node.
3.1.4 LODFs
The line outage distribution factors (LODFs) [8] are
calculated directly from the nodal PTDFs:
where PTDFX,tCO and PTDFX,fCO are the nodal PTDFs for
the “to” and “from”-nodes of the CO, and “X” represents
the CB or the CO.
The LODF is the change in MW power flow of a CB due
to the outage of another branch. The LODFCB,CO is used to
calculate the power flow of the CBCO from the
pre-outage power flows of the critical branch ( ) and
the critical outage ( ):
The LODF matrix has a row for each critical branch and a
column for each critical outage.
3.1.5 PSDFs
The impact of the phase shifting transformers is linearized
by using the phase shifter distribution factors (PSDFs).
The PSDF expresses the change in CB flow for a change
(in radians) of the tap angle of a PST:
where i is the PST index.
The PSDFs are directly calculated from the nodal PTDF
matrix through:
The PSDF matrix has a row for each branch in the network
and a column for each selected phase shifting transformer.
The PST tap positions are normally discrete, and the
PSDF describes the impact of any tap position.
For DC load flow calculations, the PSDFs are mutually
independent and the PTDFs are independent from the
PSDFs. This is the basic assumption for linear PST tap
optimization that is described in section 3.4.
3.1.6 CBCOs Sensitivities
The nodal, zonal and zone-to-zone PTDFs and the PSDFs
have been shown for the N-0 situations of the CBs. In
order to compute them under N-1 conditions, for every
CBCO, the LODFs are used. The zone-to-zone PTDF of a
CBCO depends on the zone-to-zone PTDF of the CB, the
LODF of the CB for the CO and the zone-to-zone PTDF
of the CO as follows:
Similar equation is used for the PSDFs of the CBCOs:
3.2 CBCO Selection
Evaluating transmission capacity, whether in a zone or
in between zones, for any scenario, means finding the
most constraining transmission lines for both N and
N-1 situation. However, the number of possible N-1
combinations is very high, therefore CBCO combinations
have to be selected. The second reason is that the change
in loading of many CBCOs does not have a significant
relation with zone-to-zone power exchanges. Only
CBCOs that are relevant in impacting the zone-to-zone
transmission capacity should be taken into account. The
CBCOs are filtered by a minimum zone-to-zone PTDF
for any border-direction.
After computing nodal PTDFs, LODFs and PSDFs, the
combinations of critical branches and critical outages
are selected. Specific filters are then applied to select
the relevant critical branches. Topological analysis is
performed to identify pocket connectors and bridges.
Bridges will isolate an area when removed and cannot be
used as a critical outage. The flow in a pocket connector
depends only on the power balance in a zonal area and has
no relation with zone-to-zone power exchange.
Critical branches are further selected based on voltage
level, geographical location, or other characteristics.
CBCOs are filtered by their LODF or by the absolute
MW impact of the outage on the CB. The CBCO selection
process can be adjusted according to the analysis needs.
In this way, a common CBCO list for both network models
Cigre Science & Engineering • N°7 February 2017
153
(base and investment) is created which contains N and N-1
situations for the CBs. The used filters are broad enough
in order to track all potential limiting constraints that are
relevant to for zone-to-zone exchanges, but also specific
enough to produce a workable amount of CBCOs.
3.3 FB Performance Capacity Indicators
Two new flow-based capacity indicators were developed
in order to use FB methods for network investment
evaluation. The main objective of these indicators
is to directly relate individual assets to zone-to-zone
transmission capacity. Not only to be able to calculate
the maximum zone-to-zone capacity, but also to be able
to understand why a certain capacity is reached, what is
the impact on other borders and how it may be increased
most efficiently. The objective is to quantify as well as
to create more transparency and better understand the
overall impact on the network.
Both indicators, "Remaining Available Bilateral
Exchange Margin" (RABEM) and "Asset Related
Bilateral Exchange Capacity" (ARBEC) monitor the
impact of proposed network investments on each CBCO
in the whole common grid model and their impact on the
available market capacity, per border, per direction.
3.3.1 RABEM
According to the FB method, RAM and zone-to-zone
PTDFs are needed in order to determine the flow-based
market domain.
The RAM of a CBCO depends on the network conditions,
such as net positions of zones and net exchanges between
them, and is computed as the maximum capacity of the
branch (PmaxCB) minus the flow of the CBCO ( ), for
every CBCO, as:
Equation (6) relates the flow of the CBCO to a change
in zone-to-zone MW exchange. Combining this with
equation (13) gives:
The maximum zone-to-zone exchange is reached for
PCBCO = PmaxCB and is equal to:

This is the RABEM for positive PTDFCBCO,z2z values.
Both PTDFCBCO,z2z and the line flow may also be negative.
Increasing the bilateral exchange may then relief the line
load until the line becomes loaded in the other direction, as
depicted in figure 4.
Figure 4: Bidirectional RAM for a single CBCO
Taking this into account gives the definition of RABEM:
The RABEM of a CBCO for any zone-to-zone direction
is the capacity margin for increasing the MW exchange
in that direction, for this specific CBCO. The RABEM
depends on the existing net exchanges between the
zones. It is calculated for every CBCO and every zone-
to-zone direction for both base and investment case.
The smallest RABEM equals the total remaining
zone-to-zone transmission capacity margin for each
direction. Therefore, in the comparison process, the
smallest RABEM in a specific zone-to-zone direction
of the base case is compared with the one of the
investment case. If the investment leads to a higher
RABEM, then the investment has a positive effect
on the market because the remaining transmission
capacity margin in this border direction increases.
The individual RABEM values show why a specific
change in capacity occurs and which constraints are the
limiting ones. The RABEM differences are displayed
in bar graphs and other diagrams, showing for which
CBCOs the investment has a positive or negative effect,
in terms of allowing more or less transmission capacity
between zones.
Cigre Science & Engineering • N°7 February 2017
154
3.3.2 ARBEC
The RABEM is a relative measure of remaining capacity
for a certain scenario. An additional indicator is needed
to assess the absolute maximum zone-to-zone capacity.
This is the Asset Restricted Bilateral Exchange Capacity
(ARBEC), which is the total zone-to-zone transmission
capacity for a single CBCO. It is computed by adding
each existing zone-to-zone net exchange to the RABEM
for that zone-to-zone direction:
Figure 5: RABEM vs ARBEC
Figure 5 illustrates the difference between RABEM and
ARBEC. The highlighted area represents the FB domain
(secure area) with bilateral exchange axes and the slanted
lines are the constraining CBCOs which limit the FB
domain. The RABEM of a CBCO is the distance from the
reference net position to the constraint, in the direction of
the bilateral zone-to-zone exchange. The ARBEC is the
sum of the existing net exchange and the RABEM of the
CBCO.
The Simple Tie-line Decomposition (STD) method,
described in [9], is used to calculate the net exchanges
from the network model. The European transmission
network is reduced into zones, where each zone is a
single node and every two neighbouring zones are
connected by a single equivalent branch (representing
tie-lines). Starting from a load flow result, all loop flows
are detected and removed from the network. For this
purpose graph-topological cycle detection was used.
Then, by tracing back the imported power from every question whether PSTs could relief the problem can be
sink to all sources, the transit flows in between sources answered without further calculations.
and sinks are identified. The found transit flows are also
The objective of the optimization is to maximize the
removed and the remaining flows are the zone-to-zone
exchanges, used for ARBEC computations. transmission capacity margin in a certain zone-to-zone
direction by changing the PST taps. This means to
The ARBEC is calculated for every CBCO and every maximize the smallest RABEM value in that direction:
border direction in both base and investment case. The
ARBEC differs from RABEM by a fixed amount. The
differences between the base and the investment case are
therefore the same for ARBEC as for RABEM.
3.4 PST Optimisation
The above described FB capacity indicators are used to
evaluate alternative investment proposals, where all PSTs
are kept at neutral tap positions. The zone-to-zone capacity
can often be increased significantly by optimized PST tap
positions. The PSDFs are mutually independent, and the
impact of different PSTs on the flows in the interconnected
system can therefore be added. The PTDFs are also not
affected by the PST tap changes. The PST optimization is
therefore a linear problem.
Figure 6: Maximization of the smallest RABEM (α)
This linear optimization of the PST taps is to increase the
transmission capacity in a specific zone-to-zone direction. This "maximize of the minimum" is done by maximizing
An optimized set of tap positions is calculated for every a dummy variable (α) and adding the constraint that
zone-to-zone direction and for each case separately. The RABEM for every CBCO must be higher than α.
base and investment cases can then be compared for their The PST optimization is security constrained for N
best possible performance. Additionally, the "capacity and N-1 combinations of the complete network model
gain" of these controllable assets can be monitored and by adding constraints to ensure that all CBCOs are not
evaluated. overloaded by PST tap changes, within the considered
To increase the zone-to-zone transmission capacity range. These "endangered" CBCOs are selected by fast
means to increase the lowest RABEM for all CBCOs. assessment of the minimum and maximum RAM for the
The RABEM values depend on the zone-to-zone PTDF extreme PST tap positions.
and on the RAM. The PTDFs are independent of the tap
The linear programming problem for optimizing the tap
positions and therefore:
position thus becomes:

The RAM only depends on the actual MW line flow, and

the ∆RAM is expressed by (19) and ∆RABEM by (20).

where the first line is the objective function that states that
Equation (20) directly relates the PST tap positions to α is to be maximized and the change in tap positions (τ)
the zone-to-zone transmission capacity. Any constraining is to be minimized with Cτ as penalty. The second line
CBCOs can be directly related to the PSTs and the states that the new RABEM must be higher than α for

Cigre Science & Engineering • N°7 February 2017

155
all CBCOs, the third line states that security may not be most constraining CBCO changes when the investment is
jeopardized. The last line states that the smallest RABEM added (from CBCO nr.1 to CBCO nr.12).
should grow (α ≥ 0) and that the tap positions are to stay
within operational ranges (τmin, τmax).
The combination of α and τ in the objective function is to
avoid extreme tap positions. The optimization stops when
the increase in RABEM for any additional change in tap
positions becomes very low.
The PST optimization is performed for every zone-to-
zone direction and every timestamp separately for both
the base and the investment case.

4. Results
A dataset of 816 hourly network scenarios from 34 days Figure 7: Constraining CBCOs from B to A, non-opt. PSTs

was analysed. Each scenario contains the European The same results, but now including the optimization of
network model, and the capacity between four zones (A, PST tap positions, are shown in figure 8. In comparison
B, C and D) is analysed. These models are operational with figure 7, the RABEM is higher for both base and
congestion forecast network models. investment case. The effect of the PST optimization is to
A new tie-line between zone A and zone B was added to more evenly utilize the individual assets. Normally the
each base case model, to create the investment case for PSTs are used to redistribute the asset loading throughout
every hour. the network. In flow-based for investments, the effect
of the PST optimization is to balance the zone-to-zone
The RABEM and ARBEC indicators were computed for capacity margin, per CBCO, resulting in a higher capacity
each timestamp, for every zone-to-zone direction, for that can be made available to the market.
both base and investment case, and with and without PST
optimization. Figure 8 shows the same CBCO (nr. 16) to be the most
constraining for both base and investment case.
The calculation of RABEM and ARBEC provides
transparency and insights into the impact of the
investment on the zone-to-zone transmission capacity.
Reports containing results for single timestamps as well as
aggregated reports for large datasets have been evaluated.
Figure 7 presents the RABEM results for the 15 CBCOs
with the lowest RABEM values for the direction of
zone B to zone A, for one timestamp, for both base and
investment case, without PST optimization.
The CBCOs were sorted to the lowest RABEM for either
base or investment case. The most constraining CBCOs
Figure 8: Constraining CBCOs B => A, opt. PSTs in B => A
for both cases become visible in that way. In order to
evaluate the added value of the investment, the smallest The observed significant impact on the capacity margin
RABEM in the base case (left arrow) is compared with by PST optimization, means that the controllability of the
the one in the investment case (right arrow). The smallest PSTs is significant for both cases for this specific optimized
RABEM is higher in the investment case, and the new tie- zone-to-zone direction. Moreover, the optimization of
line therefore increases the transmission capacity margin PSTs may change the constraining CBCOs. The most
for this zone-to-zone direction, for this timestamp. The constraining CBCOs in figure 8 (after optimization) are

Cigre Science & Engineering • N°7 February 2017

156
different from the ones in figure 7 (before optimization).
The effect of the investment and the PST optimization
can be further illustrated by the FB domain pictures.
Figures 9,10,11 and 12 display the domains for the same
timestamp as shown in figure 7 and 8. The axes show
additional zone-to-zone exchanges: from A to B on the
horizontal axis and from A to D on the vertical axis.
The axes origin is the combination of net positions of
the considered four zones, representing the operational
conditions in the analysed scenario.
Figure 9: FB domain, base case, Figure 10: FB domain, investment
non-opt. PSTs case, non-opt. PSTs
Figure 11: FB domain, base case, Figure 12 : FB domain, inv. case,
opt. PST for B => A opt. PST for B => A
Each CBCO limits additional zone-to-zone exchange up to
its maximum loading. The security limit of each CBCO is
drawn as a line. The most constraining CBCOs are drawn
in figures 9 to 12 as thick lines, which together determine
the secure area of the flow-based domain (highlighted
area), which is the secure market playing field.
Cigre Science & Engineering • N°7 February 2017
157
Figures 9 and 10 show the impact of the investment
without PST optimization. Figures 11 and 12 show the
effect of the PST optimization. The optimized zone-to-
zone capacity direction is towards the left (from zone B
to zone A) in these figures. For this particular timestamp,
the secure area of the flow-based domain moves in the
optimized direction, up to the full utilization of one
CBCO (100% loading), but always respecting all security
constraints.
In order to understand the impact of the investment,
together with RABEM and ARBEC, the flow-based
domain gives a spatial insight of the zone-to-zone
exchange capacity, including 2 and 3 dimensional
representation. However, the flow-based domain does
not provide quantitative indicators for investment impact
evaluation. For investment ranking, RABEM and ARBEC
are to be used.
The evaluation of an investment, including the assessment
of risks and benefits, requires the analyses of many
scenarios, e.g. 8760 hourly timestamps representing a
whole year. This requires the aggregation of results.
The simulation results for every hour of the 34 examined
days are shown in the figures 13 and 14, where the
horizontal axis shows the smallest RABEM value
(min(RABEM)) in the direction from B to A for the
base case and the vertical axis shows the Δ(RABEM)
= min(RABEM_invest) - min(RABEM_base) for the
same direction. In these figures, each dot represents one
timestamp.
Figure 13 : BC minRABEM vs Δ(RABEM), B => A, non-opt. PSTs
The upper quadrant in figure 13 shows timestamps with
positive Δ(RABEM) values, where the investment has a
higher transmission capacity margin for this zone-to-zone
direction. The lower quadrant shows timestamps in which
the investment has a negative impact compared to the
base case, resulting in a lower capacity margin.
Clustering techniques and colouring of specific time-
stamps (off-peak, on-peak, seasons, weekends, holidays,
etc.) can be used in further analyses. Individual outliers
can be investigated by single-timestamp reports as shown
earlier.
Figure 14 shows the results after optimizing the PSTs
in the direction from zone B to zone A, separately for
the base and the investment case. The dots move to the
right compared to figure 13, as the base case RABEM is
increased by the optimization. However some timestamps
show little impact of the PST optimization and remain
close to zero. The majority of dots also move upwards.
The Δ(RABEM) scatter plots of figure 13 and 14 produce
a clear view on the overall performance. In both figures,
the majority of the dots are located in the upper area,
which means that the investment has added value for most
timestamps.
Figure 14 : BC minRABEM vs Δ(RABEM), B => A,
opt. PSTs from B => A
By plotting the scatter plots for all relevant zone-to-zone
directions, and by adding quantitative measures such as
average displacement, extreme values, cluster mid points,
etc., they can provide a clear insight into the benefits
and risks of an investment proposal, when the amount of
analysed scenarios is sufficiently large.
Cigre Science & Engineering • N°7 February 2017
158
Figure 15 : Δ(RABEM), B => A vs B => C, non-opt. PSTs
The flow-based method has the advantage of combining
the different zone-to-zone transmission capacity evaluation
into one analysis. Additional exchange between zones
A and B may reduce the remaining capacity between
other zones. Figure 15 combines the change in RABEM
for two different zone-to-zone directions and shows the
interdependencies between these two directions. The
horizontal axis represents the Δ(RABEM) from zone B to
A. The vertical axis shows the Δ(RABEM) from zone B
to C. No optimization of PST is applied.
Figure 15 shows a upper right quadrant for the timestamps
for which the investment results in a higher available
capacity margin for both zone-to-zone directions. The
lower left quadrant shows the timestamps where the
transmission capacity is lower for both zone-to-zone
directions. The other quadrants show timestamps where
the transmission capacity is lower for one direction and
higher for the other.
Conclusions
A new concept for the evaluation of investment
alternatives has been developed, which can be used in
different planning timeframes: short-, medium- and long-
term planning. This concept is applicable to real network
models for the whole interconnected European grid, and
offers a transparent insight into the impact of investments
on the transmission network. The concept also bridges the
gap between network planning methods and the market
capacity calculation method.
The proposed capacity performance indicators, RABEM and
ARBEC, quantify the impact of a single network component
on the overall remaining zone-to-zone transmission capacity.
RABEM uses both the loading and the sensitivity to zone-
to-zone power exchanges. Additionally, ARBEC uses the
realized zone-to-zone net exchanges.
By sorting the CBCOs according to RABEM or ARBEC,
the sequence of "next" constraints is identified. The
added value of each network constraint that is solved by
investment proposals is determined by the "next" network
constraint which becomes the lowest RABEM.
The "flow-based for investments" concept has been proven
to be practical and applicable. Fully coherent with the
flow-based method used for day-ahead market coupling
in the CWE region, this concept aims to complement
and enhance nowadays transmission planning methods.
Therefore it is expected that planning results will be more
in line with the actual capacity given to market.
New opportunities for a joint and coordinated network
planning approach for interconnected power systems are
facilitated by this concept.
Both single-timestamp and big-data result reports have
been developed, using either graphical or numerical
methods. These types of reports are found to be
complementary.
Any type of investment (e.g. new line, new tie-line, new
PST, new generator) can be evaluated. The analysis of the
impacts of the phase shifting transformers is included in
the concept. Linear optimization of the PST tap settings
Cigre Science & Engineering • N°7 February 2017
159
allows to evaluate controllability together with the
investment for optimized network performance.
More research is required to further develop the
proposed concept, in particular developing new reporting
functionalities, adequate GSK methods for future
planning scenarios and the integration of social welfare
computation.
6. References
[1] TenneT TSO BV, Determining securely available cross-border
transmission capacity within Flow-based, May 2015
[2] European Network of Transmission System Operators for Electricity,
10-year network development plan 2014.
[3] Kwaliteits- en Capaciteitsdocument 2013, TenneT TSO B.V., 2013.
[4] P. Schavemaker, et al., Flow-based allocation in the Central Western
European region, CIGRE Session 2008, Paris, August 2008.
[5] S. Glismann, et al, Selective Generation Shift Key determination -
An enhanced method for the flow-based market coupling capacity
calculation, CIGRE C2 – 181, 2016.
[6] M. Aguado et al, Flow-based market coupling in the Central Western
European region - on the eve of implementation, CIGRE Session
2012, Paris, August 2012.
[7] K. Van den Bergh, et al., DC power flow in unit commitment models,
KU Leuven, WP EN2014-12, May 2014
[8] J. Guo, et al., Direct Calculation of Line Outage Distribution Factors,
IEEE Transactions on Power Systems, Vol. 24, No. 3, August 2009
[9] P. Hoffmann, et. al., The simple tie-line decomposition method – a
new approach for a causation based cost-sharing key, Cigre Science
& Engineering, N°5 June 2016