A Comprehensive Assessment of Markets For Frequency and Voltage Control Ancillary Services

A Comprehensive Assessm ent of
Markets for Frequency and

Voltage Control Ancillary
Services
A diesis submitted to The University o f M anchester for die degree of
D octor o f Philosophy
in die Faculty o f Engineering and Physical Sciences
2008
Yann Rebours
School o f Electrical and Electronic Engineering

ProQuest Number: 13841598
All rights reserved
INFORMATION TO ALL USERS

The quality of this reproduction is d e p e n d e n t upon the quality of the copy subm itted.
In the unlikely e v e n t that the a u thor did not send a c o m p le te m anuscript

and there are missing pages, these will be noted. Also, if m aterial had to be rem oved,
a n o te will ind ica te the deletion.
uest
ProQuest 13841598
Published by ProQuest LLC(2019). C opyright of the Dissertation is held by the Author.
All rights reserved.

This work is protected against unauthorized copying under Title 17, United States C o d e
M icroform Edition © ProQuest LLC.
ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, Ml 4 8 1 0 6 - 1346
CONTENTS
C ontents 3
L ist of tables 9
L ist of figures 11
A bstract 17
D eclaration 19
Copyright 21
A cknow ledgem ent 23
A bbreviations an d acronym s 25
Symbols 33
C hapter 1 In troduction 43
1.1 Increasing Welfare o f Power System Users 43
1.2 Stakeholders 46
1.3 Marketplaces 49
1.3.1 Basic features of a marketplace 49
1.3.2 Generation markets 51
3
4 CONTENTS
1.3.3 Transmission markets 53

1.3.4 Retail markets 54
1.3.5 Other markets 54
1.4 Fundamentals o f Frequency Control 55

1.4.1 Frequency and active power 55
1.4.2 Dynamic and quasi-steady-state frequency deviations 57
1.4.3 Self-regulation of load 57
1.4.4 Speed control of generators 58
1.4.5 Combined effect of speed control and self-regulation 59
1.4.6 Frequency characteristics 60
1.4.7 Management of an imbalance in an interconnected power system 62
1.5 Fundamentals o f Voltage Control 65

1.5.1 Apparent, active and reactive powers 65
1.5.2 Voltages and reactive power 69
1.5.3 Reactive power from a generating unit 71
1.5.4 Reactive power from a line 73
1.5.5 Management of a voltage drop in a power system area 77
1.6 Technologies Used to provide Ancillary Services 77

1.6.1 Generating units 78
1.6.2 Basic transmission and distribution assets 79
1.6.3 Purpose-built devices 80
1.6.4 Loads 80
1.7 Summary 81
C hapter 2 D elivery of ancillary services 83
2.1 Introduction 83
2.2 Needs o f Users for System Services 84

2.2.1 Reliability 84
2.2.2 Power quality 86
2.2.3 Optimal utilisation of resources 87
2.2.4 Specification of the needs 87
CONTENTS 5
2.3 Specification o f the Quality o f Ancillary Sendees 88

2.3.1 Optimal specifications 89
2.3.2 Elementary specifications 91
2.3.3 Functional specifications 93
2.3.4 Actual specifications 99
2.3.5 Standardised specifications 114
2.4 Quantity o f Ancillary Sendees 118

2.4.1 Optimal definition of the quantity of s}rstem sendees 118
2.4.2 Definitions used within the UCTE 119
2.4.3 Discussion of the UCTE recommandation 121
2.4.4 Innovative methods 127
2.4.5 Actual requirements across countries 132
2.5 Location o f Ancillary Sendees 135

2.5.1 Impact of the location of ancillary sendees 135
2.5.2 Location in actual systems 137
2.6 Conclusion 140
C hapter 3 Cost of ancillary services 143
3.1 Introduction 143
3.2 Main Cost Components o f Ancillary Services 144

3.2.1 Fixed costs 144
3.2.2 Variable costs 146
3.3 M ethodology and Hypotheses to Estimate Cost 149

3.3.1 Time horizon for a generating company 149
3.3.2 The daily optimisation process at EDF Producer 151
3.3.3 Principle of the de-optimisation cost calculation 154
3.3.4 Data considered 155
3.3.5 Hardware and software used 157
3.3.6 Cost calculation 162
3.4 Day-Ahead De-Optimisation Cost for a Producer 165

6 CONTENTS
3.4.1 De-optimisation cost over two and a half years 165

3.4.2 Seasonality of the de-optimisation cost 168
3.4.3 Parameters affecting the de-optimisation cost 171
3.4.4 De-optimisation cost and demand for reserves 176
3.5 Marginal Costs o f Frequency Control for a Producer 179

3.5.1 Study of the binding constraints 179
3.5.2 Study of the non-binding constraints 181
3.6 Cost o f Time Control in France 182
3.7 Conclusion 183
C hapter 4 P rocurem ent of S ystem Services 187
4.1 Introduction 187
4.2 Nom inating the Entity Responsible o f Procurem ent 189
4.3 Matching Supply and D em and 190

4.3.1 Long-term matching 190
4.3.2 Short-term matching 191
4.4 Choosing the Relevant Procurem ent Methods 193

4.4.1 Identified procurement methods 193
4.4.2 Procurement methods in practice 196
4.5 Defining the Structures o f Offers and Payments 197

4.5.1 Identified structures of offers and payments 197
4.5.2 Structures of offers and payments in practice 200 ■
4.5.3 Price sign and symmetry 202
4.6 Organizing the Market Clearing Procedure 202

4.6.1 Structural arrangement 203
4.6.2 Types of auction 204
4.6.3 The scoring problem 209
4.6.4 Coordination of the different markets 210
4.6.5 The settlement rule 213
4.6.6 The timing of markets 216
CONTENTS 7
4.7 Avoiding Price Caps 220
4.8 Providing Appropriate Incentives 222

4.8.1 Stakeholdersthat should have incentives 222
4.8.2 Allocation of system sendeescosts 223
4.8.3 Transmission o f data 227
4.8.4 Monitoring 228
4.8.5 Penalties and rewards 231
4.9 Assessing the Procurem ent M ethod 232

4.9.1 An effective procurement process 232
4.9.2 Low running cost 232
4.9.3 Economic efficiency 233
4.10 Summary 238
C hapter 5 Conclusions an d F u tu re R esearch 241
5.1 Rationale for the Thesis 241
5.2 Contributions to Knowledge 242
5.3 Short-Term Evolutions Desirable in France 247
5.4 Suggestions for Future W ork 248
References 251
References by categories 279
Publications 287
The au th o r 289
Appendices 291
A .l BasicsofStatisti.es 291
A. 1.1 Univariate data 291
A. 1.2 Bivariate data 294
8 CONTENTS
A .I.3 Time-dependent data 295
A .I.4 Descriptive analysis 296
.2 T ech n ical D e sc rip tio n o f O T E S S 299
A.2.1 Basic functionalities 299
A.2.2 Hardware architecture 300
A.2.3 Software architecture 301
A.2.4 Parameters for data analysis 304
A.2.5 Screenshots 307
A.2.6 Code example 310
Index 315
Approximate number o f words in the main text (Chapters 1 to 5): 58 000

LIST OF TABLES
Table 1.1: Characteristics of a 50-Hz transmission line impedance for one phase. Based on
Kundur (1994) and EDF internal documents 74
Table 1.2: Possible generation variation as a function of unit type. Based on UCTE (2004b)78
Table 2.1: General versus precise specifications of AS qualities 90
Table 2.2: Capabilities and controllers related to functional ancillary sendees 93
Table 2.3: Basic information on s}^stems studied 100
Table 2.4: Vocabulary used to name frequency control reserves in various systems 102
Table 2.5: Frequency deviation for which the entire primary frequency reseive is deployed
as a function of the droop and the primary frequency control reserve 104
Table 2.6: Technical comparison of primary frequency control parameters in various

systems 106
Table 2.7: Impact of the K-factor on secondary frequency control 108
Table 2.8: Technical comparison of secondary frequency control parameters in various

systems 110
Table 2.9: Technical comparison of voltage control parameters in various systems 113
Table 2.10: Recommendations for secondary reseive in some systems within the UCTE 121
Table 2.11: Repartition of a population for a normal distribution 124

10 LIST OF TABLES
Table 2.12: Example of a Groves-Clarke tax 130
Table 3.1: Daily management of frequency control reserves provided by EDF Producer 153
Table 3.2: Periods considered in the present study 156
Table 4.1: Parameters influencing the choice of AS-procurement method 195
Table 4.2: Procurement methods chosen across various systems as of October 2006 196
Table 4.3: Structures of payment chosen across various systems as of October 2006 201
Table 4.4: Example 1 of a Vickrey-Clarke-Groves auction 207
Table 4.5: Example 2 of a Vickrey-Clarke-Groves auction 207
Table 4.6: Synopsis of the usual auction methods 208
Table 4.7: Parameters of the demand in the considered examples of market coordination212
Table 4.8: Parameters of the offers in the considered examples of market coordination 212
Table 4.9: Result of the clearing processes in the examples of market coordination 212
Table 4.10: Settiement rules chosen across various systems as of October 2006 216
Table 4.11: Frequencies of market clearing across various systems as of October 2006 217
Table 4.12: Frequencies of reviews of the needs across various systems as of October 2006218
Table 4.13: Price caps across various systems as of October 2006 222
Table A.2.1: Structure of the table “unit” 302
Table A.2.2: Structure of the table “demand” 302
Table A.2.3: Structure of the table “supply” 303
Table A.2.4: Structure of the table “global_dispatch_statement” 303
■
LIST OF FIGURES
Figure 1.1: Distinction between ancillary and system sendees. Based on Eurelectric (2000)44
Figure 1.2: Basic layout of a power system 49
Figure 1.3: Market equilibrium 51
Figure 1.4: Principle of a generating unit 56
Figure 1.5: Dynamic and quasi-steady-state frequency deviations. Based on UCTE (2004b)57
Figure 1.6: Simplified regulation scheme of a generating unit 59
Figure 1.7: Example of instantaneous and average frequency characteristics of a power

system 62
Figure 1.8: Configuration of a zone ^ 63
Figure 1.9: Configuration of the zone 0 with the loss of a generating unit 64
Figure 1.10: A dipole (receipting convention) 65
F ig u rel.il: Temporal representation of voltage, current and instantaneous power

(inductive dipole) 66
Figure 1.12: Temporal representation of the three instantaneous powers (inductive dipole) 67
Figure 1.13: Representation of the powers in the complex plane (inductive dipole) 68
Figure 1.14: Electrical representation of an R-X dipole 69
11
12 LIST OF FIGURES
Figure 1.15: Phasor representation associated with Figure 1.14 70
Figure 1.16: Typical P-Q diagram from the stator side. Based on Testud (1991) 73
Figure 1.17:7i-representation of a line 74
Figure 1.18: Reactive power consumption of a 400 kV / 2 000 MVA line as a function of
the phase current 76
Figure 2.1: (a) Centralised and (b) decentralised dependent controllers 92
Figure 2.2: The three functional frequency controls considering a generating unit 95
Figure 2.3: Frequency and French regulation signal on 31 March 2005 [Tesseron (2008)] 96
Figure 2.4: The three functional voltage controls considering a generating unit 99
Figure 2.5: Representation of responses of an AS provider for two different inputs 115
Figure 2.6: Classification of AS based on frequency-domain characterisation 116
Figure 2.7: Proposed standardised definition of quality of afrequency or voltage control AS117
Figure 2.8: Representation of the actual response of the group versus its estimate 117
Figure 2.9: Utilisation of a profile to define a category of AS quality 118
Figure 2.10: Deployment of secondary and tertiary controls 123
Figure 2.11: Probability density o f the necessary frequency control power 124
Figure 2.12: Empirical and normal cumulative distribution functions of the activated tertiary
control power in 2006 in France 126
Figure 2.13: Cost curves of system services 129
Figure 2.14: Allocation of SS costs between four groups of users 131
Figure 2.15: Value of SS for a group o f user 131
Figure 2.16: Representation of the optimal quantity q* of SS 132

LIST OF FIGURES 13
Figure 2.17: Frequency control reserve indicators in 2004-5 across systems surveyed 135
Figure 2.18: Direct control by the reserve receiving TSO [UCTE (2005)] 139
Figure 2.19: Control through the reseive connecting TSO [UCTE (2005)] 139
Figure 3.1: Representation of two over-sized elements because of voltage control 145
Figure 3.2: Schematic dispatches to understand de-optimisation and opportunity costs 148
Figure 3.3: Overview of the optimisation process at EDF Producer [based on Ernu (2007)] 151
Figure 3.4: The two datasets considered to calculate the de-optimisation cost 155
Figure 3.5: APO G EE algorithm 160
Figure 3.6: Principle of OTESS 161
Figure 3.7: Frequency distributions of the gap between A POGEE dispatch and APOGEE
demand for all the studied time steps between 1 and 48 164
Figure 3.8: Evolution of the relative de-optimisation cost from 01/01/2005 to 30/08/2007166
Figure 3.9: Frequency distribution of the relative de-optimisation cost for all the studied
days 167
Figure 3.10: Frequency distribution of the variation of de-optimisation cost over two days
for all the studied days 168
Figure 3.11: Autocorrelation function of the de-optimisation cost up to 1-year period 169
Figure 3.12: Partial autocorrelation function of the de-optimisation cost up to 1-year periodl70
Figure 3.13: Partial autocorrelation function of the de-optimisation cost around a 3-month
period 170
Figure 3.14: Partial autocorrelation function of the de-optimisation cost around a 1-month
period 170
Figure 3.15: Partial autocorrelation function of the de-optimisation cost around a 1-week
period 171
14 LIST OF FIGURES
Figure 3.16: Normalized de-optimisation cost as a function of the average normalized

weighted marginal cost of reserves for all the studied days (r = 0.94) 172
Figure 3,17: Normalized de-optimisation cost as a function of the average normalized

marginal cost o f reserves for all the studied days 172
Figure 3.18: Normalized de-optimisation cost as a function of the maximum normalized

primary reserve demand for all the studied days (r = 0.39) 174

primary reserve demand from 01/08/2006 to 31/07/2007 (r^ = 0.71) 174

primary reserve demand from 01/01/2005 to 31/12/2005 (rxy = 0.13) 174
Figure 3.21: Normalized de-optimisation cost as a function of the mean normalized primary
reserve demand for all the studied days (r - 0.38) 175
Figure 3.22: Normalized de-optimisation cost as a function of the mean normalized

secondary reserve demand for all the studied days (rX7 = -0.18) 175
Figure 3.23: Normalized de-optimisation cost as a function of the minimum reserve share
provided by thermal units for all the studied days (rXT= 0.44) 176
Figure 3.24: Relative de-optimisation cost as a function of the new demand for reserve for
24/05/2007 178
Figure 3.25: Simplified relative de-optimisation cost as a function of the new demand for
reserve for 24/05/2007 179
Figure 3.26: Percentage of time when marginal costs of energy are higher than marginal
costs of reserves as a function of the time step for all the studied days 180
Figure 3.27: Percentage of time when marginal costs of primary reserve are higher than
marginal costs of secondary reserve as a function of the time step for all the studied
days 180
Figure 3.28: Percentage of time when marginal costs of reserves are null as a function of the
time of the day for all the studied time steps 181
LIST OF FIGURES 15
Figure 3.29: Frequency distribution of the relative de-optimisation cost due to time control
with f = 49.99 Hz for all the studied days from 01 /01 /2005 183
Figure 3.30: Frequency distribution of the relative de-optimisation cost due totime control
f — 50.01 Hz for all the studied days from 01/01/2005 183
Figure 4.1: Impact of the demand responsiveness on market clearing 192
Figure 4.2: NYISO’s demand curve for secondary frequency control. Data based on
NYISO (2008) 193
Figure 4.3: Same utilisation payments for two different utilisations 199
Figure 4.4: Representation of the generic frequency control ancillary sendee trapezium used
in Australia. Based on NEMMCO (2001) 201
Figure 4.5: Reactive power capability curve in Great Britain [National Grid (2008c)] 202
Figure 4.6: Illustration of a gaming possibility with a real-time reactive power market 217
Figure 4.7: Supply curves for different due dates 220
Figure 4.8: Effect of an offer cap (a) and a purchase cap (b) 221
Figure 4.9: Four intensities of an incentive scheme [Keller and Franken (2006)] 232
Figure 4.10: Ancillary sendees cost indicators across systems s u n n e d in 2004-5 235
Figure A. 1.1: Example of a frequency distribution 294
Figure A .I.2: Schematic representation of the smoothing process 297
Figure A.2.3: Hardware architecture of OTESS 301
Figure A.2.4: Entity relation modelling of the main tables of OTESS 302
Figure A.2.5: Modification and import of datasets with OTESS (operations 2 and 6 in
section A.2.2) 308
Figure A.2.6: Management of data in the database (operation 7 in section A.2.2) 308
16 LIST OF FIGURES
Figure A.2.7: Creation of new data in the database (operation 7 in section A.2.2) 309
Figure A.2.8: Basic calculation on the data (operation 7 in section A.2.2) 309
Figure A.2.9: Display of a graph with OTESS (operation 7 in section A.2.2) 310
Figure A.2.10: Analysis of data with OTESS (operation 7 in section A.2.2) 310
ABSTRACT
All users o f an electrical pow er system expect that the frequency and voltages
are maintained within acceptable boundaries at all times. Some participants,
mainly generating units, provide the necessary frequency and voltage control
sendees, called ancillary services. Since these participants are entitled to receive
a payment for the sendees provided, markets for ancillary sendees have been
developed along with the liberalisation o f electricity markets. However, current
arrangements vary widely from a power system to another.
This thesis provides a comprehensive assessment o f markets for

frequency and voltage control ancillary sendees along three axes: (a) defining
tire needs for frequency and voltages, as well as specifying tire ancillary sendees
that can fulfil these needs; (b) assessing tire cost o f ancillary sendees for a
producer; and (c) discussing the market design o f an efficient procurem ent of
ancillary sendees.
Such a comprehensive assessment exhibits several advantages: (a)

stakeholders can quickly grasp tire issues related to ancillary services; (b)
participants benefit from a standardised m ethod to assess their system; (c)
solutions are proposed to improve current arrangements; and (d) theoretical
limitations that need future w ork are identified.
This work, titled A Comprehensive Assessment of Markets for Frequency and

Voltage Control Ancillary Services, was submitted in 2008 by Yann Rebours to
The University o f M anchester for tire degree o f D octor o f Philosophy. ■
17
DECLARATION
No portion of the work referred to in this thesis has been submitted in support of an
application for another degree or qualification of this or any other university or other
institute of learning. ■
19
20 DECLARATION
COPYRIGHT
The author of the thesis (including any appendices and/or schedules to this thesis) owns
any copyright in it (die “Copyright”) and s/he has given The University of Manchester the
right to use such Copyright for any administrative, promotional, educational and/or
teaching purposes.
Copies of tliis thesis, either in full or in extracts, may be made only in accordance
with the regulations of die John Rylands University Library of Manchester. Details of these
regulations may be obtained from the Librarian. This page must form part of any such
copies made.
The ownership of any patents, designs, trade marks and any and all other intellectual
property rights except for the Copyright (the “Intellectual Property Rights”) and any
reproductions of copyright works, for example graphs and tables (“Reproductions”), which
may be described in this thesis, may not be owned by the author and may be owned by
third parties. Such Intellectual Property Rights and Reproductions cannot and must not be
made available for use without the prior written permission of the owner(s) of the relevant
Intellectual Property Rights and/or Reproductions.
Further information on the conditions under which disclosure, publication and

exploitation of this thesis, the Copyright and any Intellectual Property Rights and/or
Reproductions described in it may take place is available from the Head of the School of
Electrical and Electronic Engineering. ■
21
22 COPYRIGHT
ACKNOWLEDGEMENT
First, I would like to sincerely thank Pr Daniel Kirschen, who was really an amazing
supervisor during these three years. He took the time to discuss and read, had great ideas,
made numerous useful comments, showed an open-mindedness and was not bothered by
my French accent.
I would also like to thank the people at Electricite de France (EDF) that make this
PhD happens, and in particular the persons that got closely involved with this work:
Etienne Monnot, Bruno Prestat, Sebastien Rossignol and Marc Trotignon. Moreover, I
address particular thanks to Duane Robinson who helped me to find this thesis. Lastly, I
would like to thank EDF for its funding over the three years. However, the Hews expressed
in this work are not necessarily those of EDF.
This thesis, though meaning some lonely hard time, matured in an environment
both nice and studious. I thus would like to address many thanks to all the fellows from the
University of Manchester that helped me enjoy this thesis, and in particular: Chandra for his
knowledge of the Curry Mile; Jerry for his pertinent philosophical thoughts; Miguel for his
daily quotes from The Simpsons; Ricardo for his enthusiasm to visit the U.K.; Sky for the
badminton games; and Vera for the walks in the Peak District. Many thanks also to all my
colleagues from EDF, and especially the colourful R12 group, who constantly showed
support, suggested ideas and had good cheer. I would like to thank in particular: Sebastien,
my first Jedi master; Etienne, who made me run and was constantly of good counsel; Bruno
and Mehana, for their support over these three years; Marc, who spent hours reading and
discussing the ideas laid in this thesis and others; Stefan, who has always good ideas,
especially at Miami Beach or in a squash court; Frederic, who has unfailingly interesting
discussions that pop, from Into the Wild to the amount of wind blowing in Corsica; Jean-
23
24 ACKNOWLEDGEMENT
Pierre, with whom we tried to solve the world’s problems (we are still working on it); and
Jerome for his explanations about APO G EE clearer than those about squash.
As it is not possible to find all the information in books, papers or on the Internet, a
part of the information given in this thesis come from discussions with various people, in
particular: Richard Benejean and Jean-Michel Tesseron for historical procedures on
frequency control; Jean-Louis Bousquet for transmission lines; Alain Chollois for contracts
between landlords and wind producers; Renaud Crinon and Jean-Paul Echivard for the
regulation of a nuclear power plant; Arnaud Fauchille for time control; Christian Launay
and Romuald Texier-Pauton for ED F’s operational process to dispatch generating units;
Virginie Pignon for various subjects in economics; Alain Tanguy for high-voltage
transformers; Luc Tran for the behaviour of French hydro units; and Francois Bouffard and
Julian Barquin for their numerous useful comments on the final draft of this thesis. The
surveys of the various systems have been possible only with the kind participations of
Jurgen Apfelbeck, Christer Back, Noel Janssens, Ton Kokkelink, Thomas Meister, Luis
Rouco, Ilya Usov and Raymond Vice.
Lastly, I would like to express my thanks to my family and my friends who provide
me constant support over these three years and who even got the curiosity to read some of
this work. I wish that you keep drying your hair every morning thinking about frequency
and voltage controls.
This work is dedicated to Solenne.

ABBREVIATIONS AND
ACRONYMS
AC Alternative Current
ACE Area Control Error
ACF Autocorrelation Function
ADSB Adaptive Deterministic Security Boundaries
AER Australian Energy Regulator (Australia)
AS Ancillary Sendee
AU Australia
AVR Automatic Voltage Regulator
BDEW Btmdesverband der Energie- und Wassemnrtschaft e. V (Federal association for the
energy and water management) (Germany)
BE Belgium
BNA Biindesnetâgentur (regulator of the federal grid) (Germany)
CAISO California Independent System Operator (USA)
CAL California
CER Commission for Energy Regulation
25
26 ABBREVIATIONS AND ACRONYMS
Cl Cost Indicator
Cigre Cornell International des Grands RJseaux Electriques (International Council on Large
Electric Systems)
CNSE Comision Nacional del Sistema Electrico (National Commission of the Electrical
System) (Spain)
C 02 Carbon dioxide
CPS Control Performance Standard (North America)
CRE Commission de Regulation de I'Energie (Commission o f Energy Regulation) (France)
CREG Commission de Regulation de I'Electricite et du Ga% (Commission of Electricity and

Gas Regulation) (Belgium)
CSV Comma-Separated Values
DC Direct Current
DCS Disturbance Control Standard (North America)
DE Germany
DG Distributed Generation
DGEMP Direction Generale de I'Energie et des Matieres Premieres (General Direction of Energy
and Raw Materials) (France)
DIDEM E Direction de la DEmande et des Marches Energetiques (Direction of the Demand and
Energy Markets) (France)
DNO Distribution Network Operator
DO Distribution Owner
DP Dynamic Programming
DSO Distribution System Operator

ABBREVIATIONS AND ACRONYMS 27
DTe Direciie Toegicht Energie (Supervision Department of Energy) (The Netherlands)
EDF Electricite de France SA (Electricity of France) (France)
EENS Expected Energy N ot Served
EnBW EnB W Transportnciê A G (Germany)
E.ON E .O N Ne/g GmbH (Germany)
ERAP Entity Responsible for Ancillary sendees Procurement
ERCIM European Research Consortium for Informatics and Mathematics (Europe)
ES Spain
ETSO European Transmission System Operators (Europe)
FACTS Flexible Alternating Current Transmission System
FERC Federal Energy Regulatory Commission (USA)
FQ First Quartile
FR France
FTP File Transfer Protocol
FTR Financial Transmission Right
GB Great Britain
or Gigabyte
GPS Global Positioning System
HHI Herfindahl-Hirscliman Index
Hi. High frequency response (Great Britain)
HMI Human-Machine Interface

I Intentional
or Integral
IEA International Energy Agency
IEEE Institute of Electrical and Electronics Engineers
IET Institution of Engineering and Technology
IMF International Monetary Fund
INPG Imtitut National Tolytechnique de Grenoble (National Polytechnical Institute of

Grenoble) (France)
INSTN Imtitut National des Sciences et Techniques Nucleaires (National Institute of Nuclear
Science and Techniques) (France)
IQR InterQuartile Range
ISO Independent System Operator
KSH Korn Shell
LAMP Linux, Apache, MySQL and PHP
LEG Laboratoire dTilectrotechnique de Grenoble (Electrotechnical Laboratory of

Grenoble) (France)
LMP Locational Marginal Price
LP Linear Programming
LSE Load-Serving Entity (USA)
MB Megabyte
MO Market Operator
NAG Numerical Algorithms Group

NERC North American Electric Reliability Corporation (North America)
NEMMCO National Electricity Market Management Company (Australia)
NI Non Intentional
NL The Netherlands
NOx Nitrogen Oxide
No rec. N o recommendation
NYISO New York Independent System Operator (USA)
NZ New Zealand
Ofgem Office of Gas and Electricity Markets (Great Britain)
OMEL Operctdor del Mercado EUctrico (Electric Market Operator) (Spain)
OTC Over-The-Counter
OTESS OiiTilpour PEtiide des Scrviccs Systbm (Tool for the study of ancillary services)
P Proportional
PACF Partial Autocorrelation Function
PHP PHP: Hypertext Preprocessor
PI Proportional Integral
PMU Phasor Measurement Unit
POD Point of Delivery
Pri. Primary frequency response (Great Britain)
PSS Power System Stabilizer
RAM Random Access Memory

RCT Reserve Connecting TSO (UCTE)
REE Red Electric# de Espaiia (Spanish Electrical Grid) (Spain)
RI Reserve Indicator
RMS Root Mean Square
RRT Reserve Receiving TSO (UCTE)
RSI Residual Supply Index
RTE RJseau de Transport d'Electricite (Electrical Transmission Grid) (France)
RTO Regional Transmission Organisation (USA)
RWE RW E Transportnet^ Strom GmbH (Germany)
SE Sweden
Sec. Secondary frequency response (Great Britain)
SO System Operator
SO-CDU System Operator-Central Dispatching Upravlenie (Russia)
SS System Service
Stem Statens Energimyndighet (Swedish Energy Agency) (Sweden)
SVC Static var Compensator
SvK Svenska Kraftnat (Sweden)
T Total
TO Transmission Owner
TQ Third Quartile
TSO Transmission System Operator

UCEI University of California Energy Institute (USA)
UCPTE Union for the Co-ordination of Production and Transmissionof Electricity

(Europe)
UCTE Union for the Co-ordination of Transmission of Electricity (Europe)
UPFC Unified Power Flow Controller
UPS Unified Power Systems of Russia (Europe and Asia)
VCG Vickrey-Clarke-Groves
VET Vattenfall Europe Transmission GmbH (Germany)
VOLL Value of Lost Load
XML Extensible Mark-up Language *

SYMBOLS
A C E nerc ACE of zone %according to NERC (in W)
ACE[,crB ACE of zone âccording to the UCTE (in W)
B susceptance of the capacitance (in S)
Bz frequency bias setting of zone ^ (in W/0.1 Hz)
Cj cost paid by user i (in €)
C capacitance (in F)
C(q) cost of the SS deployed (in €)
CD dispatch cost of the day D (in €)
^dr—optimisation de-optimisation cost due to frequency control of the day D (in €)
hydro dispatch cost of the da)'- D (in €)
Statin h-optimisaim, relative de-optimisation cost due to frequency control of the day D (no
unit)
thermal dispatch cost of the day D (in €)
C l dispatch cost while providing reserves during the day D (in €)
33
34 SYMBOLS
^without „Jt™ dispatch cost while not providing reserves during the day D (in €)
x %oj wi/hi mm* dispatch cost with the demand for reserves equals to X % of the initial
demand of the day D (in €)
C ^s annual cost o f a given ancillary service for zone ^ (in €/year)
annual wholesale energy cost for zone %(in € / year)
C I^S cost indicator for a given ancillary service for zone ^ (no unit)
D self-regulation of the load (in %.Hz-1)
or day (an integer)
deviation between the empirical F*(x) and the modelled F(x) series, used in
the Kolmogorov test
ELwtytm hourly average energy consumption of the zone %(in MWh/h)
hourly average energy production of the zone ^ (in MWh/h)
f (x px r ..xn) objective function (usually in €)
f actual system electrical frequency at the considered point (in Hz)
fn nominal electrical frequency of the power system (in Hz)
/ quasi-steady-state electrical frequency of the power system (in Hz)
ft target system electrical frequency (in Hz)
f'm system electrical frequency measured by generating unit i (in Hz)
f'K the nominal frequency of the power system set in the controller of generating
unit i (in Hz)
SYMBOLS 35
system electrical frequency measured by zone y; (in Hz)
F(.v) modelled series related to the empirical series F*(.v)
F* (x) empirical series
^ generating unit number
hK estimated power d elation of the zone ^related to P ^v<!aj (in %)
H hour
i current (in A)
or user i in a Groves-Clarke tax system or in the Aumann-Shapley method

(integer)
or time step i in A PO G EE (integer)
I root mean square current (in A)
I complex representation (phasor) of the current (in A)
base current for a given per-unit representation (in A)
Ir characteristic current of a line (in A)
^■conductor maximal permanent current in a line conductor (in A)
Ih complex representation (phasor) of the current flowing in the electrical node k

(in A)
I~ME factor to compensate the difference between the integration of the
instantaneous power exchanged and the demand’s energy measurements of

zone ^
j constraint number (an integer)
J total moment of inertia of the rotor masses (in kg.m2)

36 SYMBOLS
k electrical node
K proportional gain of a first order system (in output unit/input unit)
KK K-factor of zone ^ (in W /Hz)
L inductance (in Id)
or Lagragian function (usually in €)
U Lerner index (no unit)
number of variables to optimise (an integer)
net value got by the user / (in €)
Nr. number of generating units providing speed control
P instantaneous power (in YA)
A number of poles of an electrical rotating machine
A instantaneous active power (in W)
A instantaneous reactive power (in var)
P(omttmpUon total active power consumed in the power system (in W)
' m u w vp titm estimate of the internal consumption of the zone ^ (in MW)
Pmaxtenmmhtion estimate of the maximal internal consumption of the zone ^ (in MW)
P average instantaneous active power or simply active power (in W)
P complex representation (phasor) of the active power (in W)
electrical active power consumed and produced before the perturbation (in W)
electrical active power (in W)
g iniratm total active power generated in the power system (in W)

SYMBOLS 37
Pm active power flowing through the electrical node k (in W)
Pi power loses in the process (in W)
Pm mechanical active power sent to an electrical rotating machine (in W)
p„ nominal active power produced by the generating unit (in W)
P'demand demand for power generation for the time step i (in MW)
P'dispatch power generation dispatched during the time step i (in MW)
PG
!O active power set-point of the generating unit i without any frequency control (in
WO
p; nominal output active power of the generating unit i (in W)

1Gn
P.C
tnfercotmecltou active power exported by the zone ^ to the power system (in W)
p~inttmmtrtion0 scheduled active power exported by the zone \ to the power system (in W)
P5
intercomiecliontn measured value of the total power exchanged by the zone with other zones,
where a positive value represents exports (in W)
maxgeneration estimate of the peak generation for the zone y; for the day (in MW)
Pz
PC price cap (in €/MW)
Penallyp penalty due to the power mismatch for the time step i (in €)
Penalty^ penalty due to the primary frequency control reserve mismatch for the time
step i (in €)
Penalty^ penalty due to the secondary frequency control reserve mismatch for the time
step i (in €)
Pf power factor (in per-unit)

38 SYMBOLS
q* clearing quantity (in good unit)
q* quantities of SS actually used (e.g., in MWh)
q,- quantity of SS actually used by user i (e.g., in MWh)
j2 maximum of the instantaneous reactive power or simply the reactive power (in
var)
Q complex representation (phasor) of the reactive power (in var)
Qk reactive power flowing through the electrical node k (in var)
Q„ nominal reactive power produced by the generating unit (in var)
Q.pqd reactive power flowing through the point of delivery (in var)
rxy cross-correlation between two series (no unit)
R resistance (in Q)
RY primary frequency control reserve of the zone ^ (in MW)
^P"'demand demand for primary frequency control reserve for the time step i (in MW)
Ppddjspal(h primary frequency control reserve dispatched during the time step i (in MW)
dispanhth frequency control reserve dispatched on thermal units during the time
step i (in MW)
secondary frequency control reserve of the zone ^ (in MW)
Psc'demand demand for secondary frequency control reserve for the time step i (in MW)
Psc'dispatch secondary frequency control reserve dispatched during the time step i (in MW)
PsJdispatchth secondary frequency control reserve dispatched on thermal units during the
time step i (in MW)

SYMBOLS 39
R IP reserve indicator for the primary frequency control reserve of the zone y (in %)
R /y reserve indicator for the secondary frequency control reserve of the zone ^ (in
%)
S apparent power (in VA)
S complex representation (phasor) of the apparent power (in VA)
Shajf base apparent power for a given per-unit representation (in VA)
Slh:e apparent power admissible by the line (in VA)
Stt nominal apparent power produced by an electricalmachine (in VA)
s'G droop of tire generating unit i (in per-unit)
share[b thermal reserve share for time step i (no unit)
t time (in s)
Cphjmcnr deployment time a (in s)
deployment time b (in s)
T cycle time (in s)
or time constant of a first order system (in s)
Tf electrical torque (in N.m)
Tw mechanical torque exercised on the rotor of an electrical rotating machine (in

N.m)
u voltage across a dipole (in V)
U root mean square voltage of u (in V)
Ud!m network dimensioning voltage (in V)

40 SYMBOLS
U" network nominal voltage (in V)
Vf value put by the user i (in €)
V average voltage between two electrical nodes (in V)
V bcl5t base voltage for a given per-unit representation (in V)
Vk root mean square voltage from the ground to theelectrical node k (in V)
Me complex representation (phasor) of the voltage from the ground to the

electrical node k (in V)
P'unt-imi the voltage line-to-line of the line (in V)
constraint function (in various units)
window the width of the extrapolation (an odd integer larger or equal to three)
-V data of die temporal series (in series’ unit)
x extrapolated extrapolated data to complete the temporal series (in series’unit)
x r: primal variable to optimise (various units)
X reactance of an inductance (in Q)
or the amount of the initial demand for reserves (in %)
% power system zone number (integer)
Z impedance (in Q)
P'lmt base impedance for a given per-unit representation (in Q)
Zf characteristic impedance of a line (in Q)
8 angle between two voltages (in rad)
e accuracy of measurement (in % or in Hz)

SYMBOLS 41
A%C®(_optlmjsa!joil relative variation of the de-optimisation cost for D in comparison to D—1
(no unit)
Af quasi-steady-sate frequency deviation from the nominal frequency^, (in Hz)
Ajf' measured frequency deviation by the generating unit i (in Hz)
AP quasi-steady-state power unbalance
APmilsm/Jp!iail consumption change following a quasi-steady-sate frequency deviation (in W)
^Pgentmthu generation change following a quasi-steady-sate frequency deviation (in W)
APG change in the active power set-point of the generating unit i for any frequency
deviation (in W)
AP^„wmplio„ consumption change in zone y following a quasi-steady-sate frequency
deviation (in W)
^generation generation change in zone y following a quasi-steady-sate frequency delation
(in WO
APfotemnmrfiian change in the power transfer between the zone and the other interconnected
power systems (in WO
AH voltage difference between two electrical nodes (in V)
(p lag between current and voltage (in rad)
X instantaneous frequency characteristic (in W /Hz)
X. Lagrange multiplier related to the constraint J (usually in €/unit of the
constraint)
Xp marginal cost of power for the time step i (in €/MWtii)
X'R weighted marginal cost of reserves for the time step i (in (€/MW0/h)
42 SYMBOLS
Xp marginal cost of primary frequency control reserves for the time step i (in
(€/MW)/h)
Xk marginal cost of secondary frequency control reserves for the time step i (in
(€/MW)/h)
A frequency characteristic o f the power system for a given frequency deviation (in
W /Hz)
A' frequency characteristic of the zone % for a given frequencydeviation (in

MW/Hz)
k clearing price (in currency unit/ good unit)
% simulated competitive price (in currency unit/good unit)
Tcy price paid by user i (in €/MW)
a standard deviation (in the unit of the value considered)
t delay (in s)
electrical angular frequency (in rad.s-1)
(*)w angular velocity of an electrical rotating machine (in rad.s-1) ■

CHAPTER 1
INTRODUCTION
Never be entirely idle; but either be reading, or writing orpraying

or meditating or endeavouring something/or the public good.
Thomas a Kempis (1380 - 1471)
1.1 In creasin g W elfare of P ow er S y ste m U sers
USER connected to a power system (e.g., a generating unit or a consumer) wishes to
A have access to a system that meets a certain standard of quality. In particular, this
user expects that the frequency and voltage will stay close to their nominal values because
most electrical appliances are designed for a particular frequency and a given voltage.
Failure to meet these frequency and voltage standards would lead to losses for the users that
may range from a burnt bulb to the loss of production in an expensive process (e.g., in an
automobile factory). Efficient frequency and voltage controls are thus essential to maintain
a high welfare for all power system users.
The frequency and voltage control services are called system services (SS) because they
are delivered by the power system to all the users. Some users of the system, such as
generators, contribute to these system sendees by acting on the frequency of the system or
the voltage at the point where they are connected to the system. Because these sendees
A Comprehensive Assessment of Marketsfor Frequency and Voltage Control A n d lla y Services, Yaiin Reborns, 2008 43
44 CHAPTER 1 INTRODUCTION
provided by users are ancillary to the production or consumption of energy, they are called
ancillary services (AS). This distinction between system and ancillary services is depicted in
Figure 1.1.
Power system
System services
Ancillary services (e.g., frequency and
voltage controls)
Some users
The other users
>.g., generators)
Figure 1.1: D istin ctio n b etw een ancillary and system services. B a sed on Eurelectric (2000)
Ancillary services have been provided by users of the power system since its early
days, more than one hundred years ago. However, it is only since the recent liberalization of
the electricity sector that ancillary services have been treated as a commodity by themselves.
Indeed, the reform of the electricity sector has led to a separation between network and
generation activities (see section 1.2). Markets for ancillary services resulting from this
separation have been developed independently across countries, depending on the previous
historical procedures or market architecture. Markets for ancillary services are thus currently
very different across systems. In addition, while ancillary services are commodities that
differ in many ways from the electrical energy product, efforts in the liberalization process
were concentrated on the main product, i.e. electrical energy and its transmission.
Therefore, the theoretical framework is much less advanced for ancillary services than it is
for energy and transmission markets. Lastly, the structure of power systems is currently
evolving fast, driven by high energy prices, aging infrastructures, increasing environmental
constraints, an intensified competition and the appearance of new technologies. For
example, new generation technologies are developed, interconnections between countries
are used closer to their limit and consumers are getting more active. Hence, markets for
ancillary services have to be adapted to this evolving structure.
In summary, markets for ancillary services are disparate across countries; they lack
from a consistent theoretical framework; and they have to be constantly adapted to the
changing structure of power systems. It is thus essential to assess current markets for
ancillary services to avoid inappropriate architectures that would lead to inefficiencies and
thus a reduced global welfare. However, stakeholders do not have a general and systematic
1.1 INCREASING WELFARE OF POWER SYSTEM USERS 45
tool that would help them assess the current markets for ancillary sendees and thus improve
current practices. Therefore, tills thesis proposes to fill this gap by providing the tools to
perform a comprehensive assessment of markets for ancillary services along three aspects:
the technical definition of ancillary sendees, the cost of provision and the market design.
First, contrary to previous works that were concentrated on specific aspects of

markets for ancillary sendees, this thesis gives a complete picture of the issues by
developing the technique, the cost and the market design together. Indeed, these three
aspects are linked. For example, the technical definition of an ancillary sendee will have an
impact on the cost to provide it (e.g., a more complex sendee is likely to be more expensive
to provide than a simpler one); the cost structure influences the market design (e.g., a cost
that is constant over time makes a short-term market unnecessary); and the technical
characteristics of an ancillary sendee may not be suitable for some particular market design
(e.g., it is useless to build a market over a large geographical area if a product is useful only
in a given power system region).
Second, the proposed assessment is based on a framework that can be applied to

any system. Tills framework is developed as follows: (a) identifying features; (b) expressing
the issues related to each feature; (c) describing the actual solutions adopted in various
systems across the world; (d) proposing innovative solutions. Therefore, both theoretical
and practical aspects are tackled. At the end of each core chapter, an assessment checklist is
proposed to help stakeholders improve their system by implementing new solutions or by
fostering more research on a specific feature. Indeed, by putting together issues and
solutions for each feature, a global solution to manage ancillary sendees becomes much
clearer.
This thesis is organised in five chapters. Chapter 1 presents the basic layout, the
stakeholders and the marketplaces of a power system. The basic concepts underlying
frequency and voltage controls are then introduced. Even if this thesis is focused on the
ancillary sendees provided by conventional large generating units, most o f the concepts are
applicable to any provider of ancillary sendees as well. The explanations are intended to be
suitable for all readers irrespective of their technical background. In particular, the concept
of reactive power is explained. Nevertheless, readers familiar with frequency control will
notice that the new concepts of average and instantaneous frequency characteristics are
defined. Lastly, technologies providing ancillary sendees are presented.
Chapter 2 focuses on the delivery of ancillary services. First, the needs of users in
terms of system services are examined. To meet these needs, ancillary services have to be
provided by some users of the system. Therefore, the specification of the quality o f ancillary
services is discussed. Lastly, the optimal quantity and location of the ancillary services are
debated. In particular, Chapter 2 shows that the amount of ancillary services currently
provided does not correspond to the actual needs of users in terms of system services.
Chapter 3 describes the main costs incurred by the provision of ancillary services by
a producer. A practical methodology to evaluate the cost of frequency control due to the
day-ahead capacity reservation is then proposed and successfully applied to Electricite de
France (EDF) Producer’s portfolio. In particular, this study gives interesting insights on the
parameters affecting the cost of frequency control. Lastly, the cost of the time control (i.e.,
the cost to maintain the frequency average at 50 Hz) is assessed for France.
Chapter 4 reviews the numerous issues related to the procurement of ancillary

sendees, namely: (a) nominating the responsible entity of procurement; (b) matching supply
and demand; (c) choosing the relevant procurement method; (d) defining the structures of
offers and payments; (e) organizing the market clearing procedure; (f) avoiding price caps;
(g) providing appropriate incentives; (h) assessing the procurement method. Practical and
innovative solutions are proposed for each issue.
Lastly, Chapter 5 provides a summary of this thesis, some scenarios o f evolutions

and possible future work.
1.2 S ta k eh o ld er s
Broadly speaking, the elements of a power system are physically divided into three main
categories: generating units, the network and loads. The network, which provides the
electrical link between loads and generating units, is actually divided into two main parts
(see Figure 1.2). The transmission network is meshed and operated at high voltages (e.g., 63 kV
to 400 kV in France), while the distribution network is usually operated in a radial fashion and
at lower voltages (e.g., 400 V to 20 kV in France). Networks are mainly constituted of lines,
1.2 STAKEHOLDERS 47
transformers1 and various controllers. Conventional large generation (e.g., coal, nuclear, large
hydro, gas or fuel) are connected to the transmission network, while smaller generation
(e.g., wind, small hydro, combined heat-power plant or photovoltaic) tend to be connected
to the distribution network, which led to the term of distributed generation (DG) to designate
this kind of generating units. Lastiy, consumers withdraw energy from the system, either at the
transmission level (large consumers) or at the distribution level (small consumers).
The various elements of a power system are owned by different parties. Prior to
liberalisation, most of them were owned by vertically-integrated companies, which owned at
the same time generation, transmission, distribution and retail. Currently, the ownership of
these activities tends to be separated. Large generating units are owned by generation
companies, which most of the time finds then roots in the historical vertically-integrated
companies. The transmission network is owned by a few entities (Transmission Owners, TO,
or transmission companies), whereas distribution networks are usually owned by many entities
(Distribution Owners, DO, or distribution companies), such as the city councils in France. Lastly,
end users are obviously owned by a large number of stakeholders. Therefore, end users are
usually gathered in consumer associations to get a more powerful representation. In addition,
they deal with retailers (or suppliers), which buy large volumes of energy from generation
companies and is in relation with the intermediate stakeholders, such as distribution and
transmission operators. In certain cases, retailers can also buy the electricity produced by
the end users who own generating assets. Note that retailers generally do not own
significant physical assets, except intelligent meters in some cases.
A liberalised power system is complex to run because of the number of participants

involved. In addition, it has been recognized that the operation of electrical networks is a
natural monopoly. Therefore, independent bodies have to be designated to manage the
power system (the system operators, or SO). The Transmission System Operator (TSO)
operates the power system at the transmission level, while the Distribution System Operator
(DSO) is in charge of the distribution level. Therefore, from the TSO’s perspective, a DSO
is equivalent to a large consumer. Note that the terms Distribution Network Operator
1 A transform er allows to converts energy from a given voltage to another level with the help o f two
windings around a magnetic circuit. T he first w inding brings pow er into the magnetic circuit, while the second
winding withdraws pow er from it. A difference in the num ber o f turns for each winding leads to a voltage
change, bu t w ith a similar pow er transm itted. N ote that Figure 1.2 did n o t explicitly display any transform ers.
(DNO), Regional Transmission Organisation (RTO) and Independent System Operator

(ISO) can also be found, with responsibility and authority varying amongst systems.
The competitive sectors of the electricity industry, i.e. generation and retail, need
marketplaces to trade products (see next section). The Market Operators (MO) are in charge
of these marketplaces. In addition, other participants may provide additional services in the
marketplace, such as the brokers, who help bring buyers and sellers together.
To foster fair relations between stakeholders, three basic functions should be

performed: (a) setting the rules; (b) monitoring that the rules are respected; and (c)
enforcing the rules if necessary. Arrangements to perform these three functions vary from
one system to another. The rules are usually set by the legislator, which is most o f the time
the legislative body of the government. The regulator, which is independent, then checks
whether the rules are respected by the participants. It also collects complaints from
stakeholders. It may also propose some rule modifications to the legislator. Lastly, it
oversees the quality of services provided by participants. Depending on the countries
considered, the power of regulators to enforce rules varies. Some of these powers may be
entrusted to other entities, such as an anti-trust commission or a market monitoring entity.
Frequency and voltage controls involve all the stakeholders described above.
Indeed, generation companies, system operators and end users modify the frequency and
voltages through their actions, as explained in Chapters 1 and 2. In addition, the market
operators, the regulator and the legislator design the rules used to manage frequency and
voltage controls, as shown in Chapters 2 and 4.
1.3 MARKETPLACES 49
Transmission
network
Large
consumer
—v Centralized
generation
Small ^
consumer
Distribution
network Distributed
generation
Figure 1.2: B asic layout o f a pow er system
1.3 M a rk etp laces
In order to facilitate exchanges of products between stakeholders and to send signals to all
the participants, markets have been developed along with the liberalization. Section 1.3.1
presents the basic features of a marketplace, while sections 1.3.2 to 1.3.5 introduce the main
marketplaces that one is likely to find in the electrical industry.
1.3.1 Basic features of a marketplace
The goal of a marketplace is to bring a buyer and a seller together, so they can exchange a
given good at an agreed price. If the delivery of the good is instantaneous (e.g., when one
buys some Camembert cheese at a favourite daily shop), or if the good cannot be resold by
the buyer before the delivery (e.g., when one buys a Norman wardrobe to be delivered to
the buyer die next day), the market is called a spot market. Otherwise, the quality of die good
has to be described, as well as the date of delivery and the price to be paid at delivery. Such
a market where commodities are traded for delivery in the future is named futures (or
sometimes forward) market. In practice, there are several futures markets, which deal with a
particular product (e.g., 1-year or 6-month delivery). Obviously, the products of the 1-year
futures market can be traded 6 month later in the 6-month futures market. Futures markets
are useful to participants to hedge against risk associated with the volatility of the spot
prices.
To trade these products (spot or future), the marketplace can be organised in two
ways. A centralised market is cleared by a unique entity that collects the offers to sell (supply
curve) and the bids to buy (demand curve). A unique price 7t* is seen by both buyers and
sellers, and a quantity q* is traded. This price and this quantity correspond to the point
where supply and demand match (see Figure 1.3). Note that it can be easily proven that the
price 71* is equal to the marginal cost of the producer2 providing the last offer (considering a
perfecdy competitive market3). On the other hand, in a decentralised market, sellers and buyers
can enter directly into contracts to buy and sell (sometimes without knowing each other’s
identity). The transaction is thus bilateral. In such a market, there is no "official price”, but
there may be mechanisms that allow all the participants to be informed of either the price
of the last trade or a weighted average of recent transactions.
An important feature of a marketplace is liquidity. Market liquidity characterises the

market’s ability to quickly match any bid to buy with an offer to sell without changing the
market price. The liquidity incorporates four features: the tightness (i.e., the capability to
avoid a large spread between the highest demand price and the lowest supply price); the
depth (i.e., the capability to absorb large trade volumes without significant price changes);
the immediacy (i.e., the capability to quickly meet the demand to sell or buy); and the
resilience (i.e., the capability to recover after a price change) [IMF (2006)].
The participants can also decide not to meet in a public marketplace, but to agree
on a bilateral contract outside any organised structure. This kind of transaction, which Is very
popular in the electricity industry to manage long-term contracts, is called over-the-counter
(OTC)4. Such transactions are usually facilitated by brokers. If the bilateral contract is firm,
it is called a fonvard contract. On the other hand, if the delivery is optional, it is called an
option. Several types of options are possible: European (a unique date of delivery), American
(which can be exercised once before a given date), swing (which can be exercised several
2 T he marginal cost is the cost to provide an additional good in the delivery (one M W h in the case o f
an energy market).
3 A perfect competitive m arket is a m arket w here no participant uses its dom inant position to distort
the signals sent by the market to participants.
4 O n die down side, O T C contracts cannot guarantee the pro d u ct provision, whereas the m arket
operator has the provision liability in an organised market.
1.3 MARKETPLACES 51
times, but there is a limit in terms of energy), Asian (which is a variant of European: the
average price of the good is taken instead of the spot price) or Bermuda (which can be
exercised at some given dates). Kluge (2006) describes in depth methods to price options in
electricity markets.
Both organised and OTC markets lead to an equilibrium, i.e. a set of operations by
stakeholders that equals supply and demand. This equilibrium can be either Pareto efficient,
i.e. it is impossible to increase the benefit of a party without decreasing the benefit of the
others, or qualified as a Nash equilibrium, i.e. in which no participant has interest to change
its position. Note that a Pareto equilibrium takes into account potential cooperative
behaviours by stakeholders, whereas a Nash equilibrium is obtained only with the help of
unilateral decisions. Therefore, a Nash equilibrium is often not Pareto efficient.
Kirschen and Strbac (2004a) provide a comprehensive introduction to the

marketplaces in the electrical industry. Varian (1999) gives a more general and more
theoretical view on marketplaces. Furthermore, Chapter 4 is dedicated to the design of a
marketplace for ancillary sendees.
price
Demand
Tf
Supply
quantity
Figure 1.3: Market equilibrium
1.3.2 Generation markets

Generation markets help generation companies find buyers for their products, i.e. energy
and ancillary sendees. Because generating units have a large investment cost, may require a
long building period (e.g., around ten years to design a nuclear power plant and up to seven
years to build it) and have lifetimes that usually span over tens of years, generation
companies need to find buyers for their products over a long period to reduce risks. In
particular, sufficient revenues are essential for generation companies to invest and thus to
maintain sufficient capacity in the long-run5. Therefore, long-term markets are essential in
generation markets. In practice, the generation companies secure their investment with
long-term OTC bilateral contracts or vertical integration. However, futures markets in
electricity are deemed to provide unreliable signals. In fact, the price of futures tends to be
constant whatever is the delivery date (e.g., 1-year, 2-year or 3-year delivery) [e.g.,
Powernext (2008)], whereas the price of electricity is likely to increase in the future.
Therefore, long-term generation markets are still an issue in electricity markets.
Once the generating units have been built and most of the power traded in an OTC
manner, generation companies and the buyers of their products enter in the short-term
futures markets (i.e., one-month or shorter) to balance their positions. Finally, the market
closest to real-time is the spot market. This spot market can be organised either with a
centralised (pool) or a decentralised (exchange) unit commitment, as discussed in
section 4.6.1. If the delivery delay of the spot market is too long (e.g., more than
15 minutes), an additional market is necessary to balance the positions of the participants.
This additional market is called the balancing mechanism (or balancing market). This kind of
market is more restrictive than the day-ahead market6 in order to avoid the exercise of
market power by some participants (e.g., prices cannot be changed easily) and is usually
operated by the system operator.
Markets for ancillary services, which are described in depth in Chapter 4, are part of
the generation markets. In particular, markets for ancillary services and for energy are tightly
linked, since a generation company can make the choice to allocate one MW of production
capacity as an ancillary service or as energy.
For further information on generation markets, see for example Wilson (2002),
Kirschen and Strbac (2004a), Baldick et al. (2005) or Joskow (2006). D G EM P/D ID EM E
(2003) provides reference prices for generating plants, as well as their building time.
5 Building sufficient generating capacity in the long-term is p art o f the pow er system adequacy issue
described in section 2.2.1.
6 T he day-ahead m arket trades one-day futures.

1.3 MARKETPLACES 53
1.3.3 Transm ission markets

To allow sellers and buyers to trade In the generation markets, a transmission network is
necessary. Such a network exhibits some particularities. First, electricity follows physics and
not die financial rules, so the impact of an injection or a withdrawal of electrical power does
not have necessarily a logical consequence from a business point of dew (e.g., an increase in
electrical power consumption can reduce the price of this power). Second, it is difficult to
build competing transmission lines, so the access to the transmission network has to be
managed by an independent entity (the TSO)7.
Since the transmission network has a cost, participants have to pay for the right to
use it. To charge users, Transmission System Operators rely on two methods. First, they
can charge the users with an ex-ante price, i.e. the users know in advance the price that they
are going to pay for their transmission use. A typical example of an ex-ante price is the
postage stamp, for which the users pay a transmission fee that does not depend on die
location of the energy generation or consumption. This tariff is designed to recover the cost
of transmission expansion and operation, but the signals sent to participants may be too
weak to foster an optimal use of the transmission network (e.g., to encourage the
construction of new generation plants in a congested area). On the other hand, an ex-post
price is possible. Such a price is calculated after (and not prior) the actual use of the
network. To allow hedging, some ex-post price estimates are given to the participants. E x
post prices are often used for competitive procurements of the available transmission
capacity. However, the actual implementation of such competitive procurements is still
under debate. A popular approach in USA is to link transmission and generation markets by
defining energy prices at each node of the system. Such prices are called Locational
Marginal Prices (LMP) and are precisely computed ex-post. The differences between the
LMPs then provide an income to the owner of the line between the two nodes. Such nodal
energy markets are completed with financial instruments that allow a party to hedge against
price differences between two nodes of the network (the Financial Transmission Rights, or
FTR). Quintana and Bautista (2006) give a short and comprehensible introduction to this
topic.
7 H ow ever, the actual construction o f the lines may be done by entities under com petition (the
Transm ission Owners).
In the same manner as generation markets, transmission markets involve both short
and long terms. Therefore, the network charges should allow the transmission owners to
raise enough money to develop the network, in addition to give short-term signals to
optimally use the available transmission capacity and to charge the users fairly. But because
generation and transmission investments are linked it may be difficult to reflect the optimal
allocation between generation companies and transmission owners.
For further reading, Kirschen and Strbac (2004a) provide a good introduction to the
issues and solutions of transmission markets. Pignon (2003) analyses the transmission tariffs
in Europe, while Rious (2007) discusses the coordination between generation and
transmission investments in a deregulated environment.
1.3.4 Retail markets
The idea of a retail market is to allow all end-users to freely choose their supplier of
electricity. Retail markets are deemed to improve services and prices offered to the end
users. However, competition in retail markets is currently quite limited [e.g., see CRE (2008)
for France]. In addition, retail markets are not complete. Indeed, end users sign only long
term OTC contracts with their supplier because of the technical limitations. It is thus not
possible for a user to swap from one supplier to another in real-time, as it would be the case
in a prefect decentralised market.
1.3.5 Other markets
The markets described previously are directly linked to the generation, transmission and
consumption of electricity. However, these markets rely on other commodities. First, they
are sensitive to the various fuel markets (e.g., coal, gas or nuclear). Second, they are also
impacted by the environment markets that price the pollution, such as the nitrogen oxides
(NOx) and the carbon dioxide (CO^, or guarantee the renewable provenance of electricity
(the so-called green certificates in Europe). Lastly, the cost of generating units may be
impacted by the prices of other commodities, in particular raw materials such as steel or
copper.
1.4 FUNDAMENTALS OF FREQUENCY CONTROL 55
1.4 F u nd am entals o f F requency Control

Tills section introduces the main principles and definitions that are necessary to understand
frequency control. In particular, the new concept of instantaneous frequency characteristic
is introduced in section 1.4.6.
1.4.1 Frequency and active power

The electrical frequency f of a power system is the same across the entire synchronous
zone8. Therefore, all the rotors of the alternators from this area rotate at the same angular
velocity (0m (in rad.s'1)9, times a factor that is a function of their number of poles p, as
shown in (1.1) [Grainger and Stevenson (1994)].
/ =
P © (1.1)
2 2%
As shown in Figure 1.4, the electrical power provided by a generating unit comes
from the alternator. The mechanical rotation of the rotor creates a rotating magnetic field,
which is then converted into an alternative current in the stator. The rotor is driven by the
prime mover, which can use different types of fluids such as steam, water or wind. The
mechanical power PMprovided to the prime mover creates a torque Tm on the shaft (the
mechanical torque). On the other hand, the electrical power Pf delivered by the generator
creates an opposite torque Tt on the rotor (the electrical torque). If these two torques equal
each other, the rotor’s speed will not change. However, if the mechanical torque is larger
than the electrical torque, the rotor will accelerate. Conversely, if Tm is lower than T,, the
alternator will slow down. This relation can be easily modelled mathematically with the so-
called swing equation given in (1.2), where J is the total moment of inertia of the rotor
masses (in kg.m2), cl? the rotor acceleration (in rad.s"2) and Pt the power lost in the
process (in W) [Grainger and Stevenson (1994)].
8 A synchronous zone is an area interconnected through alternative current. In transient state, the
frequency actually varies slightly from one part o f the network to another, especially in a large interconnected
netw ork. A n illustration with the N o rth A merican network is given by Z h o n g et at (2005), while K undur
(1994) describes the theoretical fundamentals.
9 A pow er system is like a tandem , on which all the cyclists have to pedal at the same speed to
transm it power.
Electrical power
Mains
Mechanical power
(steam, water, wind...)
1—\pm )
r*
y
Rotor
-V Prime mover
\i
Stator Te
L ?
Lost power Alternator = rotor + stator
Figure 1.4: Principle o f a gen eratin g unit
d~co
Jo>.— r = P * - P - P < (1.2)
at
Equation (1.2) is very important for dynamic studies since it shows the behaviour of
a particular unit in response to mechanical or electrical power changes. From the generating
unit’s perspective, Pr is imposed by the network, as it is equal to the difference between the
consumption and generation of active power at the terminals of the generator10. If the
electrical power drawn by the power system decreases (i.e., if P, decreases), the electrical
machine will start accelerating. In addition, all the generating units across the system will
also accelerate, according to the same reasoning11. Therefore, the mechanical power injected
into the machines has to be reduced to slow down the rotors. Conversely, if the electrical
power drawn by the system increases (i.e., if Pe increases), the generating units will slow
down and the mechanical power will have to be increased to maintain the same speed.
In conclusion, the active power injected in or withdrawn from the power system
affects the frequency of the system. By adapting the mechanical power provided to
generating units, the rotor speed (and thus the electrical frequency) can be controlled.
Similarly, the frequency can also be controlled by adapting the consumption of electricity by
the loads.
,0 Obviously, the active pow er provided by the generator is n o t included in this difference.
11 This reasoning is true if the netw ork capacity is sufficient. Otherw ise, pow er oscillations can
appear, i.e. som e generating units can accelerate while others are slowing down.
1.4.2 Dynamic and quasi-steady-state frequency deviations

As any physical value, the frequency goes through a transitory state in response to a
perturbation before stabilising to a new value. The maximum deviation from the target
frequency during die transitory period is called the dynamicfrequency deviation. The deviation
between the target value and the final value is called the quasi-steady-state deviation (see
Figure 1.5). The term “quasi” is used here because a steady-state analysis is performed on a
dynamical system12. Though, the dynamics considered are slow enough to allow such a
definition. The quasi-steady state is usually defined with the help of a maximum gradient
that the frequency should not exceed [UCTE (2004b)]. Note that similar definitions exist
for voltage deviations.
frequency
target
frequency quasi-steady-state
frequency deviation dynamic frequency
deviation
time
F igure 1.5: D y n a m ic and quasi-steady-state frequency deviations. B a sed o n U C T E (2004b)
1.4.3 Self-regulation of load

The active power consumed by a load depends in part on the frequency, especially for
motors. In general, a frequency increase leads to a load increase, and vice-versa. Flowever,
some loads such as arc furnaces have an opposite behaviour JXundur (1994), p. 310].
To make explicit this relation between load and frequency, let us consider a stable
power system state, in which the same amount of active power P0 is consumed and
produced at the nominal frequency f r A quasi-steady-sate frequency deviation A f from the
nominal frequency f (i.e., usually 50 Hz or 60 Hz) will lead to a consumption change
U P „ ,p,„„ according to the Equation (1.4).
At = / „ - / , (1-3)
12 A pow er system never reaches a steady state because various param eters are always fluctuating. F o r
example, the frequency usually comes back to its target value after a perturbation w ith the help o f various
controls (see C hapter 2).
A Pm!sô„ = Dy<P0 x A f ( 1.4)
where /jlf is the quasi-steady-state electrical frequency of the power system (in Hz); and D,
positive and expressed in % /H z, is called self-regulation of the load. It depends on the structure
of the power system, in particular on the number and the types of loads on-line. The term
“self-regulation” is used because this effect tends to be naturally opposed to frequency
variations. However, it is usually not sufficient to compensate imbalances without a large
quasi-steady-state frequency deviation since the value of this parameter is usually between 1
or 2 % /H z [UCTE (2004b)].
1.4.4 Speed control of generators

Most generating units have a controller that adjusts the mechanical power Pm provided by
the prime mover as a function of the rotor speed (0m (see Figure 1.4). This speed control is
performed by a simple feedback loop (see Figure 1.6) with an adjustable parameter 1/ sGi
where s'G is called the droop of the generating unit i (in per-unit and positive). The relation
between the power set-point of the generating unit, the droop and the frequency deviation
is given in (1.5) [UCTE (2004b)]. Therefore, the lower the droop, the stronger is the
generating unit’s response. This Equation can be compared to (1.4) concerning the self
regulation of load.
AJ fw f - fJ n‘
J m
=
AP' P c -P L
1 Git 1 Gn (L5)
p i= -—LT xPLxAf:
In X JG
where APG is the change in the power set-point of the generating unit i for any measured
frequency deviation t f m
’ (e.g., dynamic or quasi-steady state), P G
‘ tt the generating unit’s
nominal output power, PG0 the power set-point without any frequency control and f ’n the
nominal frequency of the power system registered in the generating unit’s controller. Note
that die negative sign can be sometimes omitted to define the droop sG, which thus leads to
negative droops (e.g., —4 % instead of 4 %).

Consequently, the production of N G generating units providing speed control will

increase by APge„m,;0„ following a quasi-steady-state frequency deviation Af APgtmrafio„
expressed as follows (if the maximum outputs of the generating units are not reached):
Nr.
A Pgeneration = 2 > 4
£=1
f Ng
v -nk
<;,xaa//,A
(1.6)
'^-J J„ v A
\k=i G J
AP.generation AT
V^=i sc J
The droop is individual to each generator and is generally set at around similar
values in order to avoid the non-linearities that can appear if generating units reach their
maximal outputs for different frequency deviations13. In practice, droops range between 4
and 20 % (see section 2.3.4.2), which lead to very different responses.
Change in
power set- Measured
point AP g Speed frequency fn
T ransducer
controller
Electrical Mechanical
power set- + power Pm
Process
Speed co,
point P g o - Prime mover
controller + Alternator * Electrical
power Pe
F igure 1.6: Sim plified regulation sch em e o f a g en eratin g unit
1.4.5 Combined effect of speed control and self-regulation

At the quasi-steady state, consumption and production are equal. Therefore, a quasi-steady-
state power imbalance AP (a positive value indicates an excess in consumption) is
compensated by the combined effect of generating units AP (via the speed regulation)
and loads /SPnns„mpiiBn (via the self-regulation), as shown in (1.7).
13 The non-linearities can be spotted easily w ith the concept o f instantaneous frequency characteristic
developed in section 1.4.6.
AP + AP.consumption APgeneration = 0
<=> AP - APgeneration - APconsumption (1.7)
( T}k \
Gn
<=> AP
E
\k=\ sg y
x ^ j — D x P Qx A f
Jn
Therefore, the quasi-steady-state frequency deviation following a power imbalance is equal

to:
AP
(t8)
o x ii+ 7 - Z %
Jn k^l SG
The number N G and the droops sG of the generating units providing speed control
actually vary over time, as well as the self-regulation D. However, for a given imbalance AP
(i.e., for a few tens of minutes), these parameters can be assumed to be constant. In
addition, the nominal powers PG and the nominal frequency^, are also constant, so the
following simple linear relation can be deduced:
— —C* > 0 (1.9)

A/
In other words, a quasi-steady-state frequency deviation depends exclusively on this

constant if only the droops and self-regulation are considered. This constant, expressed in
MW/Hz, is sometimes called the composite frequency response characteristic or the stiffness of the
system [Kundur (1994)].
1.4.6 Frequency characteristics
In addition to the automatic actions of the generating units (i.e., speed control) and loads
(i.e., self-regulation), the entity responsible for the power system (i.e., the system operator)
can take several additional actions to restore frequency:
S With regard to the generation:
■ Modify the generation output;
■ Stop or start new generating units.

S With regard to the consumption14:

■ Modify consumption (e.g., a residential load may shut down its electrical heating
upon request);
■ Disconnect or connect loads.
Therefore, in the same manner as for (1.9), but considering any actions that can be
taken as a function of frequency, a simple linear relation can be assumed between the power
imbalance AP (a positive value indicates an excess in consumption) and the quasi-steady
state frequency deviation Af.
(1.10)
A/
where A(A/"), positive and expressed in MW/Hz, is the frequency characteristic of the power
systemfor the quasi-steady statefrequency deviation A/r[UCTE (2004b)].
The definition of the frequency characteristic described above does not show the
regulating capacity actually available for a frequency different from /). To fill this gap, it is
proposed to introduce the concept of instantaneous frequency characteristic X, defined for an
elementary quasi-steady-state variation of the frequency d f as in (1.11). The partial
derivation is used since the active power consumed or generated in the system can vary
along with other parameters such as voltage. The frequency characteristic for a given
perturbation A, which represents the average of the instantaneous frequency characteristic,
can then be easily calculated through an appropriate integration, as shown in (1.12).
canmmptwa^fqss ) generation^\-fqrs)
(1.12)
14 N etw ork assets are considered here as consum ers o f active pow er (e.g., see section 1.5.4 for the
consum ption o f a line).
Figure 1.7 shows the interest of this new definition. First, the instantaneous
frequency characteristic A. of the regulating devices can be easily calculated. Therefore, the
frequency characteristic A(A^) can be conveniendy deduced for any frequency deviation.
Second, the contributions of the different regulating devices are clear in the top graph (e.g.,
the participation of a particular generator becomes recognisable), whereas it is much more
difficult to identify them in the bottom graph. Last, non-linear behaviours of the system are
spotted easily (e.g., in this case, the system does not react anymore to frequency changes
beyond a certain point).
instantaneous frequency characteristic A (MW/Hz)
” ^ disconnection of loads
-— generators with a small droop
generators with a large droop
frequency characteristic A (MW/Hz)
0 A f (Hz)
Figure 1.7: E xam ple o f in stantaneous and average frequency characteristics o f a pow er system
1.4.7 Management of an imbalance in an interconnected power

system
As frequency is a value common to the whole interconnected power system, an active
power imbalance in one part of the system will lead to a reaction from across the whole
system. Figure 1.8 shows a zone ^ which is interconnected with the other zones of the
system and exports a power •

other zones
consumpUOi
interconnection
/generation
zone z
F igure 1.8: C onfiguration o f a zo n e z
In the case of a quasi-steady-state frequency deviation Af, the generation and the
consumption of the zone is affected, because of the Equation (1.10), as expressed in (1.13).
In other words, a frequency change leads to a deviation from the scheduled power transfer
between the zone ^ and its neighbouring zones.
gmmttion (mmmptiou inltmnmlion
■ » - A ‘(A r)x 4 f-A P 4 ™ « tf. = 0 (U3)

» A P i m^ , = A '( A / ) x A /
Then, let us suppose that die zone 0 loses generating unit 1 that produces (see
Figure 1.9).
other zones
consumptioi
■' © ©
© ©.
generation interconnection
zone 0
F igure 1.9: C onfiguration o f the z o n e 0 w ith the lo ss o f a g en eratin g u n it
Following this loss, the frequency of the whole power system starts to decrease (see
section 1.4.1). Therefore, each zone starts to increase its exchange according to
Equation (1.13). For all the zones except zone 0, the change in generation AP*„eratio„ is only
due to frequency control, so the exchange deviations are equal to:
A J £ _ » , = -A *(A /)x A / , V* * 0 (1.14)
O n the other hand, zone 0 has the following exchange deviation because both P^ and
frequency control have to be considered in AP^lltraliffll ■
AT’1 . ,- ., , = -A °(A /)x A / - P'G (1.15)
The algebraic sum of all the exchanges at the various interconnections is zero because all
the exchanges are considered twice: once positive and once negative, which leads to:
= 0 (1.16)
Therefore:
A*(A/)x A /-A ° (A /)x A f - P ^ = 0

(1.17)
» P , i = -A /-x 2 > * ( A /)
1.5 FUNDAMENTALS OF VOLTAGE CONTROL 65
In conclusion, the imbalance between generation and consumption in zone 0 is

compensated by the whole interconnected power system (1.17). However, unscheduled
exchanges appear at the interconnections, as shown in (1.14) and (1.15). Therefore, the
frequency has to be brought back to its target value in order to remove these unscheduled
exchanges. This is done using complementary controls described in section 2.3.3.1.
1.5 F u n d am en tals o f V o lta g e Control
This section introduces the main principles and definitions that are necessary to understand
voltage control. In particular, the concept of reactive power is explained in section 1.5.1.
1.5.1 Apparent, active and reactive powers

Let us consider a passive dipole13, in which a current i(t) flows and with a voltage u(t)
between its terminals16 (see Figure 1.10). The instantaneous powerp(t) absorbed by this dipole
is equal to the product of i(t) and n(t). Figure 1.11 shows these three values as a function of
time when they are sinusoidal quantities, assuming a 50-Hz frequency f and a 30° lag 9
betweencurrent and voltage17. Equations from (1.18) to(1.21) give themathematical
expressions of i(t),u(t) and p(t), where CO is called the angularfrequency, U the Root Mean
Square (RMS) voltage and I the RMS current.
u(t)
— H h-------
m
F igure 1.10; A dipole (receipting convention)
U) “ 27tX / (1.18)
//(/) = u 4 l sin(co/) (1.19)
15 A passive dipole is an electrical elem ent w ith two nodes and w ithout any internal energy source.
16 T h e current quantifies the debit o f electrons and the voltage the electrical pressure that is exerted
on the electrons.
17 A lag o f 30° is equivalent to 7 1 /6 radians. As the current lags the voltage in this case, the dipole is
said to be inductive.
i(t) = I -s/2 sin(o>/ —(p) (1.20)
p it) = u(t) x i(t) (1.21)
P(t)
i(t)
u(t)
0.005 0.010 0.015 0.020 0.025 0.030
t(s)
F igure 1.11: T em poral representation o f voltage, current and in stan tan eou s pow er (inductive dipole)
Figure 1.11 shows that the absorbed powerp(t) is sometimes negative (i.e., when u(t)
and i(t) have opposite signs), which means that the dipole provides power during this time.
Since the dipole does not have any internal source of energy, it has to sometimes withdraw
more energy from the source before giving it back later on. This storage/release can be
mathematically isolated in two components as follows [Bastard (1998)]:
S p jt), called instantaneous reactive power, is null on average and represents the alternation
between the storage and release of energy by the dipole;
S pr(t% called instantaneous active power, is always positive and represents the power actually
used by the dipole.
Equations from (1.22) to (1.26) give the mathematical expressions for p jt) and p r(t), where P
is die average of the instantaneous active power (or simply the active power) and Q the
maximum of die instantaneous reactive power (or simply the reactive power). Note that cos (p
is usually called power factor and noted pf. Figure 1.12 shows the temporal representation
of the three instantaneous powers for the same dipole as considered in Figure 1.11.
P(l) =«(/■)X /(/) =pxt)+PM (1-22)
p r(t) = P x [l —cos(2t0/)] (1.23)

P = UI cos cp
A- (0 ~ —Q.x sin(2co/) (1.25)
O —UI sin (p (1.26)
P(t)
pr(t)
px(t)
0.000 0.005 0.010 0.015 0.020 0.02S 0.030
Us)
F igure 1.12: T em p oral representation o f the three in stan tan eou s pow ers (inductive dipole)
Temporal representations are not very easy to handle because the sine and cosine
functions lead to complex calculations when added or multiplied. A practical manner to
handle sinusoidal values is to transform them into the complex plane, where the sinusoidal
values can be seen as vectors that are rotating in this plane. Equation (1.27) shows the
transformation used in this example, while Equation (1.28) gives the result of this
transformation for the three instantaneous powers, where P, O and S are complex values. S
is called the apparent power because it is related to the instantaneous power, i.e. the physical
power used by the dipole. Equation (1.29) expresses the link between instantaneous powers
and their complex representations. Equation (1.30) shows the simple link between the three
powers. Lastly, Figure 1.13 shows the representation of the three powers in the complex
plane. The projection of these vectors on the real axis gives the instantaneous physical
values, because of Equation (1.29). Note that the power vectors are rotating in a clock-wise
manner at twice the angular frequency CO18, whereas the voltage and the current rotate
trigonometrically at the angular frequency.
\(B~)
A e f t +; B e W ; X = Asm(B)\-^X = A e 2 = ^ [s in (B )- jc o s ( j3 )] (1.27)
p = p e i(-2w/-71)
jf-2u>/— 1
2) a - 28)
S = P + £ = S e l(?
A ( / ) = P + Re(p)
'pA =M g) (1-29)
p(t) = P + Re(S)
S 2 = P2 + Q 2 (1.30)
lm f
P(t) Re
F igure 1.13: R epresentation o f the pow ers in the co m p lex plan e (inductive d ip ole)
To conclude, dipoles can store electrical energy in a magnetic manner (e.g.,

generating units, lines or transformers) or under an electric form (e.g., capacitor,
underground cables) before releasing it quickly. This release/storage of energy creates a
phase difference between voltage and current, which leads to the use of reactive power.
Note first that other elements based on power electronics can also produce a phase
difference between current and voltage and thus reactive power. Second, reactive power is
essential for the good working of essential power system elements. For example,
18 T he alternating com ponent o f active pow er that can be seen in a tw o-phase system creates
vibrations in a rotating machine. H ow ever, in an equilibrated three-phase system, the sum o f the three pow er
vectors leads to a constant power, as the three alternating com ponents are cancelled.
transformers need an alternating magnetic field (and thus a storage/release of magnetic

energy) to transfer power from one winding to another. Third, reactive power is a value that
is limited to alternative-current power systems: it is never drawn away from the system. For
example, the prime mover of a generating unit only provides active power to the alternator
and the end-user motor only provides active power to the shaft. However, even if the
reactive power cannot be lost, it increases the active power losses through the Joule effect
(see section 1.5.4).
1.5.2 Voltages and reactive power

To illustrate the link between voltages and reactive power, the dipole of Figure 1.10 is
described more precisely in this section. Indeed, Figure 1.14 shows two electrical points
connected through a resistance R and a reactance X 19. In practice, such an electrical link is
equivalent20 to a short aerial transmission line (where V , and X 2 are the voltages at each
terminal, see section 1.5.4 for a more precise model) or a simplified synchronous machine
(where JX; is the internal voltage, see section 1.5.3) [Grainger and Stevenson (1994)]. The
electrical point 1 produces an active power P, and a reactive power O,. The electrical
point 2 consumes an active power P2 and a reactive power 0 2. Current I lags voltage
LJ —V 2 because of the inductance. The associated phasor representation in the complex
plane of the voltages is shown in Figure 1.1521.
Figure 1.14: E lectrical representation o f an R -X dipole
19 T h e reactance X is die imaginary p art o f the impedance Z, H ere, X is equal to _Loj, w here L is the
physical characteristic o f the elem ent and is called the inductance (see section 1.5.4).
20 A single dipole is equivalent to a three-phase electrical elem ent by considering drat the diree
phases are equilibrated and that the single dipole represents one phase. T herefore, the three-phase pow er is
equal to drree times the single-phase power.
21 See section 1.5.1 for a short explanation on the phasor representation o f electrical values. See for
example G rainger and Stevenson (1994) for tire link between voltage and current.
F igure 1.15: P hasor representation a sso cia ted w ith F igure 1.14
From Figure 1.15, by using 6 and squaring the trigonometric expressions, we

have [Bastard (1998)]:
AV -^RxRI2 + - ^ X x X I 2 +RP2 + Xj22

(1.31)
V V
where:
V =
vt+v2
(1.32)
A V = V j- V 2
For a transmission line, the term X Q 2 is predominant” , so Equation (1.31) can be

approximated as follows23 [Bastard (1998)]:
(1.33)
Therefore, supposing that J /, is constant (e.g., with the help o f a controlling device such as
a generating unit), any change in the consumption of the reactive power by the end terminal
will result in a voltage drop. Therefore, by controlling the consumption of reactive power
Q_2i the voltage 1V2 can be maintained at this terminal. In other words, by maintaining
voltage and current in phase (and thus limiting the reactive power consumed by the line),
voltage drops/rises can be limited in the transmission network. This direct link between
22 Because I is low, P2 IQ 2 ~ 3 0.95) and X / R « 10 (see section 1.5.4)

23 Because o f the strong approxim ation, relation (1.33) should be used only to give an order o f
magnitude. Otherw ise, (1.31) should be preferred.
voltage and reactive power is less strong for distribution networks because the ratio X /R is
lower (see section 1.5.4), so the term RP2 is not negligible anymore, and both active and
reactive power consumption have to be controlled to maintain voltage.
1.5.3 Reactive power from a generating unit
As shown in Figure 1.4, the mechanical power Pm of a generating unit is converted into
electrical power Pt through the alternator. The rotor of an alternator is equivalent to a large
rotating magnet. This rotating magnetic field then induces an alternative voltage (the internal
voltage) in die windings of the stator. The latter are connected to the system through the
mains, and provide the stator voltage. The internal voltage is proportional to the amplitude of
die magnetic flux. Therefore, the voltage at the stator of the transformer is controlled by
adjusting the magnetic field of the rotor, which is itself regulated by the field current
flowing in the windings of the rotor (the exciter current) [Grainger and Stevenson (1994)].
In practice, the Automatic Voltage Regulator (AVR) controls the exciter current to
maintain the stator voltage at the given value. The exact configuration and tuning of such
system varies from one technology to another and relies on die local condition of the
machine. In addition, the AVR can include a Power System Stabilizer (PSS). A PSS takes
into account die speed or the active power of die machine to act on the exciter current.
Kundur (1994) provides much detail on the configurations and tuning of AVR and PSS.
To understand the link between the stator voltage and the reactive power
produced/absorbed by the generating units, Figure 1.14 and Figure 1.15 can be used.
Figure 1.14 is an equivalent of a stator, where X and R represent the impedance of the
stator; V , the internal voltage; V2 the stator voltage; I the stator current (i.e. the current
flowing out of the terminals of the generator). Note that the internal voltage L) is physically
created along the whole stator windings. Therefore, die proposed model is only an
equivalent, in which die internal voltage source and die impedance are separated (a
Thevenin model). The stator voltage V2 always lags the internal voltage V, in generator
mode. However, the stator voltage can either lag or lead the stator current I. If the
generator is over-excited, the stator current I lags the stator voltage V 2 and the generator
delivers reactive power (as represented In Figure 1.15). In this case, die system consumes
reactive power in this case because Q 2 positive24. On the other hand, if the generator is
under-excited, the stator current leads the stator voltage, the generator absorbs reactive
power and the system provides reactive power (Q2 < 0) [Grainger and Stevenson (1994)].
The capability o f a generating unit to provide or absorb reactive power is limited,

and depends on the active power produced and the stator voltage. A P-jQ diagram is
convenient to represent such capabilities. Figure 1.16 shows a typical P-Q diagram from the
stator side for a given stator voltage25. Therefore, when a generator is producing/absorbing
reactive power, active power may have to be reduced, which may create an opportunity cost
(see section 3.2.2). However, this opportunity cost is theoretical because the connecting
conditions for generating units usually impose large reactive power capabilities (see
section 2.3.4.4).
Most of the time, a generating unit controls the voltage (and thus the reactive power
produced or consumed by the generator) at its stator. However, the system operator may
ask the operator of the generating unit to express its reactive power capability at the Point
of Delivery (POD) instead of the stator. The POD is the point owned by the generator that
is the electrically closest to the network (see Figure 3.1). As rule-of-thumb, at the nominal
apparent power Sn of the generating unit, the reactive power absorbed by the auxiliaries26,
the step-up transformer27, and the transmission line up to the POD is usually around 15%
of T,,28. Therefore, the reactive power Qp0D supplied by the generator to the network can be
estimated using Equation (1.34).
a P0 D * Q 2- 0,15xJ„
j2 k > d » J & - 0,15x 4 : <U4)
24 0 .2 " U , I sin <p according to (1.26), and sin (p > 0.

25 Similarly, a U /Q diagram can be built for a given active pow er output.
20 T he auxiliaries are all the elem ents that a pow er plant needs to run, such as pum ps, controllers o r
light.
27 T he nom inal stator voltage is generally lower (e.g., 20 kV) than th e nom inal voltage at the P O D
(e.g., 400 kV). T he producer thus needs to step-up the voltage through a transform er.
28 T he m ajor com ponent o f the 15 % is due to the im pedance o f the step-up transform er, w hich is
around 13 % (see note 74 p. 145).
Practical P-Q diagram

This diagram is different
Limit imposed by the for each voltage
prime mover
)r = Uo
Limit imposed by
Limit imposed / ■'Ah ■opportunity cost: ■A copper losses
by the rotor / y m y be incurred
edge/
temperatui^ ©
Working zone
Limit imposed
by iron losses
0 Q (M) Q
Reactive consumption Reactive production
Figure 1.16: T yp ica l P -Q diagram from the stator side. B a sed on T estu d (1991)
1.5.4 Reactive power from a line

To study die contribution of a line to the absorption/generation of reactive power, a better
model than the one described in Figure 1.14 has to be used. This new model is shown in
Figure 1.17. It is a single-phase Tt-representation of the three-phase line. R and E represent
the resistance and the inductance of the line, respectively. Their values depend on the cable
section, the type of material (e.g., steel with aluminium or copper), the number of
conductors, the configuration of the cables and the temperature [Johannet (1997)]. C, which
mainly depends on the configuration of the cables, is the equivalent capacitance of the line
lumped at each extremity29. This modelling simplifies reality because the capacitance is
actually distributed along the whole line. However, this is a good approximation for a line
of medium length (less than 200 km) and if the electromagnetic transients are
neglected [Bastard (1998)]. Lastly, the conductance, i.e. the resistance between the line and the
ground in neglected here. Table 1.1 gives typical values of these parameters for different
voltages. It is clear that the ratio E /R of the inductance over the resistance increases with
the line voltage because high-voltage line have larger sections and thus lower resistance than
low-voltage lines, whereas the inductance stays relatively constant.
29 Coo is called the smceptance o f the capacitance and is often denoted B, N o te that the reactance o f the
capacitance is equal to 1 /C o .
R jLa)
Li -2
■AAAr
2 _L 2 _L
jCco jCco ~ j~
F igure 1.17: 7i-representatiou o f a line
T ab le 1.1: Characteristics o f a 50-H z transm ission lin e im p ed an ce for o n e ph ase. B a sed on Kundur
(1994) and E D F internal docu m en ts
U nderground
Voltage Aerial line
cable
(kV) R L Leo C Cw C Cw
Leo/R
(Jfi/km ) ( m il/ k m ) (a /k m ) (nF /k m ) (fiS /k m ) (n F /k m ) (n S /k m )
45 0.22 1.14 0.36 1.7 10.3 3.25 - -
63 0.24 1.31 0.41 1.8 8.9 2.79 * 400 w 126

90 0.16 1.28 0.40 2.6 9.1 2.84 * 300 -9 4
150 0.14 1.36 0.43 3.0 8.6 2.69 - -
225 0.083 1.26 0.40 4.9 9.4 2.95 * 200 *63

400 0.034 1.08 0.34 10.2 11.0 3.46 « 150 *47
500 0.028 0.86 0.27 9.6 13.8 5.20 -
765 0.012 0.87 0.27 22.8 13.2 4.98 - -
To express the reactive power consumed or produced by the line shown in

Figure 1.17, the current flowing to the ground through the capacities is supposed to be
negligible in comparison with the current flowing into the line. In addition, the voltage
amplitudes are supposed to be identical at both terminals of the line, for example with the
help of two generators that maintain these two voltages. These assumptions on currents and
voltages can be expressed as in (1.35).
I ,* W
(1.35)
V,xV2= V
So the consumption of the line in active P and reactive Q powers is as follows, where R IZ
are the Joule losses; —B V 2 the reactive power provided by the capacitance; and X I 2 the
reactive power consumed by the inductance30 [Johannet (1997)]:
P = BJ2
(1-36)
q =x f - brx
Therefore, the following power balance can be deduced:
Pj =P2+ RI~
(1.37)
IQ} =J22 + X X - B V ^
To illustrate the reactive power consumption of the line, let us consider a

400 kV / 2 000 MVA line. To give an idea, this kind of line has usually a bundle three 570-
rnm2 conductors per phase. For this kind of conductor, the maximal permanent current
ndtif/or between 900 and 1 100 A per conductor, depending on the season, so less than
2 A /m m 2 (note that the steel core is included in this calculation). Therefore, the maximal
power Sljnf admissible by this line is between 1 870 MVA and 2 290 MVA and
mathematically expressed in (1.38), where V is the voltage line-to-line. But usually the
line is operated between 40 and 60 % of this power in order to respect the N -l criterion31.
^= 3xJ
= 3 x TVx I
r t1-38)
= V 3xK to. f e x 3 x J _ lm.
= V x 400 000 x 3 x / .....
The reactive power consumption as a function of the current has been calculated
using (1.36) and Table 1.1, and is shown in Figure 1.18 for three 400 kV / 2 000 MVA lines
of different lengths, for which the base parameters32 are given in (1.39). Four remarks can
be drawn. First, a line can produce reactive power, which leads to a voltage difference
between the two terminals of the line (see section 1.5.2). In particular, the voltage rise at no
load is called the Ferranti effect. Second, it is clear that a long line requires more reactive
30 W hen the line is heavily loaded, die term —B V 2 becomes negligible in com parison to X I 2, which
leads to the simplified m odel o f section 1.5.2.
31 T he N - l criterion states that the loss o f an elem ent (e.g., a line) should n o t jeopardize the security
o f the system.
32 T he absolute value is found by multiplying the base value by the per-unit value.
support than a short line. Third, for any line length, the reactive power consumption is
always null for a given power, called the characteristic power, which corresponds to a
current It given by (1.40), where Zt is called the characteristic line impedance. Lastly, it is
important to note that an underground cable produces much more reactive than an aerial
line because of its high susceptance B (see Table 1.1). This high reactive power leads to
strong voltage rises that limit the length of underground cables.
0 .2 5 -
0 .2 0 - 2 0 0 km
0 . 15 -
1 5 0 km
0 .0 5 -
5 0 km
0.00-
- 0 .0 5 -
0.0 0.1 0.2 0.3 0.4 0.6 0.6
I (p.u.)
F igure 1.18: R eactive pow er co n su m p tio n o f a 400 kV / 2 000 M VA line as a fu n ction o f the p h ase
current
Vim = 400kV
.5^ = 2 000 MVA
V, 2
z #„ = - f ^ = 80Q (U9)
^ base
^ , = ^ l = 5000A
base
V
1 = (1.40)
z = (1.41)
In conclusion, an increase in the reactive power Q j produced at one terminal of the

line leads to an increase in the current I flowing in the line for an identical active power
1.6 TECHNOLOGIES USED TO PROVIDE ANCILLARY SERVICES 77
transmitted P, (see section 1.5.1). This larger current causes more reactive power O to be
consumed in the line (see Figure 1.18). Therefore, the reactive power sent by one side of
the line is not fully transmitted to die other side. This phenomenon also appears for active
power, but to a lesser extent because the resistance is much lower than the reactance (see
Table 1.1). Therefore, it is not possible to transmit reactive power over long distances.
1.5.5 Management of a voltage drop in a power system area
Contrary to frequency control, a voltage change in a particular area of a power system will
usually not be seen by the whole interconnected area because the impedance is getting very
high when the bus considered is far from the voltage drop. The mitigation of voltage drops
is however more complex than for frequency drops. Indeed, a voltage drop may be created
by a fault, such as a short circuit between a line and a tree. Injecting more reactive power is
useless in this case, as it would lead to very high current. Therefore, the fault needs to be
isolated first. Fault isolation is one of the areas that the power system protection discipline
covers. See for example Anderson (1999) for more details on the adopted strategies.
If the voltage is still low when all the faults have been cleared (i.e., considering a
quasi-steady state), more reactive power has to be injected as close as possible to the voltage
drop. This compensation is mainly done by using the capability of generating units, as
described in section 1.5.3. Other techniques can also be deployed and are described in the
following section. As there are hundreds of buses in the transmission network (and
thousands considering the distribution network), the voltage is actually controlled at key
points, which leads the other bus voltages to follow. The control of key points is described
in section 2.3.3.3.
1.6 T e c h n o lo g ie s U sed to provide A ncillary S e r v ic e s
This section provides an overview of the current technologies that can be used to control
frequency and voltage. For further details, Flirst and Kirby (1998) provide a comprehensive
comparison between most of the technologies mentioned in this section.
1.6.1 Generating units

Generating units are the main providers of active and reactive powers in the system. A
distinction can be made between two main families:
Thermal units, which are based on a temperature difference between a hot and a cold
source that makes a fluid flowing (usually steam, but also hot gas, for example in the
case of a gas turbine or a diesel generator). This family uses energy resources such as
coal, fuel, gas, nuclear or the Earth’s heat in the case of geothermal energy;
S Non-thermal units, which are based on the natural flow o f a fluid, such as wind or
water, or other technologies, such as photovoltaic electricity generation. This family is
getting more important, since they usually produce little greenhouse gas (e.g., C 0 2) over
their lifetime.
Generating units are technically limited in terms of frequency and voltage controls.
These limits are due in particular to the boiler output flow, to the lag induced by controllers,
to the thermal gradient that is bearable by the unit, to the inertia of the process, or to
environmental constraints (e.g., the maximum water level variation for a hydro unit).
Table 1.2 gives an idea of the possibilities of some typical generating units, where the
possible generation variation is expressed in percentage of the nominal output power of the
unit per minute. The difference between hydro and thermal units essentially comes from the
inertia and the thermal constraints of the boiler of the thermal units. The differences across
thermal generating units are due to the various complexities in changing the flow of
combustible in the process [Wood and Wollenberg (1996)].
T able 1.2: P ossib le gen eration variation as a fu n ction o f u n it type. B ased o n U C T E (2004b)
Lignite33 N uclear Coal33 Fuel or gas Hydro

1 to 2 % /m in 1 to 5 % /m in 2 to 4 % /m in 8 % /m in 90 to 150 % /m in
33 Coal is constituted betw een 75 and 91.5 % o f carbon, whereas lignite has betw een 50 and 60 % o f
carbon [Hoinkis and Lindner (2007)].
1.6 TECHNOLOGIES USED TO PROVIDE ANCILLARY SERVICES 79
To control the temperature of the boiler, most of the thermal generating units rely
on the variation of the energy flow. However, in the case of nuclear power plant, nuclear
material cannot be removed or added in operation. Therefore, nuclear generating units use
two principles, which have an impact on the possible variation of active power [Gautier
(1990)]:
V Change the boron concentration in the reactor core. This technique has a large time
constant (tens of minutes to change output by a few tens of percent);
V Use the control rods to regulate the neutron flow. Control rods can be either “black” or
“grey”. The former can only shut down the reactor, whereas the latter ones can tune
more precisely the reaction. The technology of “grey” control rods was developed
during the eighties. Therefore, only nuclear plants built after this period and that have
installed this system may modulate their active power relatively quickly. Power plants
with the regulating control rod can vary their power up to 5 % /min.
1.6.2 Basic transm ission and distribution a ssets

Transmission and distribution networks consist mostly of lines and transformers.
Therefore, it is efficient to use their capabilities to control frequency and voltages. First, the
transformation ratio of the transformers can be changed within real-time operation. This is
done by selecting the appropriate position amongst around ten taps [Grainger and
Stevenson (1994)]. However, such a system is quite slow. For example, it takes 30 s for
French transmission transformers and 10 s for French distribution transformers to change
their first tap (and then 1 tap/10 s). Moreover, this technique creates short over-voltages, is
limited in amplitude and can lead to a voltage collapse when a transformer tries to absorb a
constant amount of power that the system is not able to provide.
Second, because lines and transformers have a relative high inductance, they draw
large amounts of reactive power. In addition, the resistance of theses elements induces
active power losses (see section 1.5.4). Therefore, the impedance of the network can be
modified by taking transmission lines in or out of sendee. This method creates over
voltages, increases the chances of incident during the manoeuvre and is limited in
amplitude. In conclusion, the basic transmission and distribution assets are useful but have
a limited manoeuvrability by comparison with the generating units.
BO CHAPTER 1 INTRODUCTION
1.6.3 Purpose-built devices

To increase the manoeuvrability of the system in terms of frequency and voltage controls,
purpose-built devices can be installed.
First, passive elements such as capacitances and inductances can be added to change
the reactive power consumption/production at different points of the network. They can
be switched on or off either manually or remotely. They are cheap and very helpful, but
create transients, can introduce resonance problems and their reactive power compensation
depends on the voltage.
Some power electronics can be added to these passive elements to increase their
capabilities. These devices belong to the large family of the Flexible Alternating Current
Transmission Systems (FACTS). Their cost is obviously higher than for passive elements, but
they allow a continuous regulation similar to generating units. Some FACTS are dedicated
to voltage control, such as Static var Compensators (SYCs), while others are intended for
flow control but can perform voltage control as well, such as Unified Power Flow
Controllers (UPFCs) [Hingorani and Gyugyi (2000)].
High-inertia rotating machines can be used as well. They can provide both high
quality reactive power control (they are said to work as synchronous compensator.s) and a short
term active power control through their inertia (they are said to work as energy storage devices).
They are usually expensive because o f their moving parts, and their capabilities in terms of
frequency control are limited. Therefore, the improvement in the storage of electricity may
help the development of new devices purpose-built to provide ancillary services.
1.6.4 Loads
The power system end-users are large consumers of ancillary services because their
consumption of active and reactive powers varies a lot over time. They can thus have a big
role to play to improve the control of frequency and voltages. However, their availability is
often not predictable, their rating is typically small (which increases management costs), and
their participation is most o f the time not continuous. However, the development of low-
cost information systems in parallel with high energy prices (because of more
environmental constraints and less available energy resources) may foster a more active
participation of loads in the future.
1.7 SUMMARY 81
1.7 Sum m ary

Consumers, generating units and system operators expect an appropriate standard for the
frequency and the voltages across the system. Such a standard refers to the system services.
In practice, the global frequency and tire local voltages depend on the balances between
consumption and generation of active and reactive powers across the power system.
However, these balances are constantly perturbed by users. Therefore, some users have to
act on the active and reactive power balances in order to improve the frequency and
voltages, instead of perturb them. Such contributions by the users are called ancillary
services. Users hence can benefit from system services, consume them, provide ancillary
sendees or do all of these.
The developments in information technology, power electronics and electricity

markets forces all users to face their double responsibilities as a consumer of system
services and a provider of ancillary services. However, to grasp fully this opportunity, the
users need an appropriate framework to manage ancillary sendees. Unfortunately, current
frameworks are disparate, and none can claim to be fully efficient. Therefore, this thesis
provides the tools necessary to assess any market for ancillary sendees and gives essential
elements to improve it. The assessment developed in this thesis is based on three objectives:
define the technical needs and solutions to meet the expectations of the users (Chapter 2);
assess the cost of these solutions (Chapter 3); and design an efficient marketplace for
ancillary sendees (Chapter 4). It is then the responsibility of the appropriate stakeholders to
apply this assessment to their system. ■
CHAPTER 2
DELIVERY OF ANCILLARY
SERVICES
- Vincent: You know what they call a Quarter Pounder with Cheese in Paris?
- Jules: They don't call it a Quarter Pounder with Cheese?
- Nah, man, they got the metric system. [...] Thy call it a "Royale with Cheese".
Pulp Fiction movie (1994)
2.1 Introduction
Chapter 1 has shown that users expect a certain standard for the frequency and voltages. In
addition, users can provide ancillary sendees to help maintain this standard. Therefore,
Chapter 2 assesses the delivery of ancillary sendees. First, section 2.2 makes explicit the
needs of users in terms of system sendees. Section 2.3 then introduces the various
specifications that define the different qualities of ancillary services that users can provide.
From the needs of the users and die available qualities of ancillary sendees, the quantity of
each AS quality can then be determined, as presented in section 2.4. Lastiy, section 2.5
discusses die issues related to the delivery location of ancillary sendees.
A Comprehensive Assessment of Marketsfor Frequency and Voltage ControlA nailmy Sendees, Yann Rebours, 2008 83
84 CHAPTER 2 DELIVERY OF ANCILLARY SERVICES
2 .2 N e e d s o f U sers for S y ste m S e r v ic e s
The needs of users can be expressed as appropriate reliability, power quality and power
system utilisation. System services, by controlling reactive and active power balances, help
to meet these three objectives. This section reviews these objectives and shows that they
may be contradictory. Therefore, a trade-off has to be found and the best objective function
depends on the system considered.
2.2.1 Reliability
The reliability of a power system is the probability that the system will be able to serve
consumers. Adequacy and security underlie the concept of reliability. An adequate system is a
system with sufficient facilities (e.g., transmission lines and generation) to satisfy the
consumers, while considering static conditions. A secure system is able to respond to
contingencies, i.e. considering dynamic conditions [Billinton and Allan (1996)]. Therefore,
day after day, the adequacy and the security of a power system constitute its reliability.
Obviously, system services are essential to maintain the security of the system (see
sections 1.4 and 1.5). But they also affect the adequacy decisions. First, enough capabilities
to provide reactive and active powers have to be built in the long-term in order to respond
to contingencies in the short-term (see section 4.3.1). Second, AS capabilities and
transmission capacity are linked (e.g., reactive support can increase transmission capacity
and transmission capacity can increase frequency control capabilities), so they have to be
coordinated. Therefore, system sendees deeply influence the reliability of power systems
from both security and adequacy aspects.
Even if reliability (and therefore security) is a quality and not a good, reliability can
be considered as a public good in economic language [Abbott (2001), Boucher et al. (2006)].
Indeed, the characteristics of a public good are (a) the impossibility (or high difficulty) to
exclude a consumer and (b) the inexistence of competition in the consumption: what one
consumes does not deprive another one [Varian (1999)]. Reliability fits this definition,
because separating the reliability of two users is very difficult, especially with more and
more interconnected systems, and because any user can enjoy the reliability of the power
system without depriving the others. Note that system services are not a perfect public
good since over-consumption by a participant may hamper the welfare of the others, in the
2.2 NEEDS OF USERS FOR SYSTEM SERVICES 85
same manner as the over-consumption of a freeway reduce the welfare of drivers because
they are travelling more slowly.
Unfortunately, users of a same power system have very different points of view on
what the reliability of die electric system, i.e. the level of public service, should be. To
consider a well-known example, a semi-conductor factory and a residential consumer do
not give the same value34 to a reliable source of electricity. In addition, reliability is at least
100 times more valuable to consumers than it is to generators [Kirschen (2002)]. The value
of security35 is even more complex to calculate than the value of reliability because the value
of security varies in real-time whereas reliability is defined over a given period. For example,
a residential consumer would value short-term security much more while watching the final
of the world cup on TV than while sleeping. Easdy, this problem increases with the size of
the synchronous network. For example, all the users of the synchronous continental
European network have to agree on consistent reliability criteria. Consequentiy, an
independent party has to define the appropriate level of security for the power system36.
This independent party may be eitiier the system operator or the regulator on behalf of the
consumers. Two approaches can be adopted to define the necessary level of security.
On the one hand, deterministic security criteria can be adopted, such as the famous N -l
criterion. According to tiiis criterion, the power system has to be able to continue operating
without a major component (e.g., a transmission line or a large generator). This criterion is
therefore binary (die system is declared as secure or not) [Kirschen (2002)] and relatively
easy to calculate. A flaw of this system is to considerer only a predefined set of
contingencies, and to ignore completely events with a low probability of occurrence.
Another flaw is to consider that all die events are statistically independent, whereas the
recent North American37 and European38 experiences show diat the increasing complexity
34 T he customer damage function gives the cost o f an outage for a given customer. F rom the custom er
damage functions, the Value O f L ost Load (VOLL), which is determ ined through surveys, expresses the value
that the average consum er places on being deprived o f one kW h w ithout w arning [Eariuki and Allan (1996),
Joskow and Tirole (2004)],
35 Practically, T he Expected Energy N o t Served (EENS) may be used to measure the security o f a
pow er system [Kirschen and Jayaweera (2007)]. N ote that the E E N S does n o t take into account the value o f
energy.
36 In any case, a pow er system is never totally secure [Kirschen (2002)]. H owever, it can have a
certain level o f security. This level o f security has then an im pact on the reliability o f the system,
37 T he N o rth -E ast o f the USA and the South-East o f Canada had a blackout on the 14* o f A ugust
2003 that affected around 50 millions people [U.S.-Canada Pow er System O utage T ask Force (2004)].
of the system multiplies dependencies [Kirschen and Strbac (2004b)]. In addition, for a
given operating point, the risk of outage (and so the value of security) relies heavily on the
weather conditions [Kirschen and Jayaweera (2007)]. Therefore, deterministic criteria of
security do not reflect the real level of security of the system, but are very practical in the
day-to-day operation of the system.
On the other hand, a risk-based approach may be preferred. In such a case, an

optimisation problem has to be solved. For example, it may consist in maximizing the value
minus the cost of security, based on the probability of occurrence of different incidents
within the system. The cost of security can be computed because its cost components are
calculable (e.g., the cost o f building an additional line, the cost of installing a larger
transformer...), even if some of them are rather difficult to determine (e.g., a faster aging of
generators because of providing reactive support). However, the value of security is very
difficult to compute because it differs widely between users and varies with time as
explained previously. This last problem may be solved in part if a certain reliability
separation can be attained in the future with individually interruptible loads [Cramton and
Stoft (2005)]. Another problem is that computing resources are today too low to achieve a
probabilistic approach on-line. However, new deterministic criteria could be adopted like
the Adaptive Deterministic Security Boundaries (ADSB) proposed by Kirschen and
Jayaweera (2007).
2.2.2 Power quality

Power quality is “the concept of powering and grounding sensitive equipment in a manner
that is suitable to the operation of that equipment” [IEEE Std 1159-1995 (1995)]. In
particular, power quality is related to electrical phenomena such as voltage sags or
harmonics, which are local problems, whereas security is a global problem. However, a clear
separation between power quality and security can be difficult. For example, a voltage
within the security margins increases both quality and security. Moreover, increasing the
meshing of a network increases the reliability because it makes more paths available for the
delivery of energy, but it may also increase the number and severity of voltage sags seen by
consumers. On the other hand, some quality parameters, like the level of harmonics, are not
38 T he w hole Italy (around 60 millions people) was plunged in the dark on th e 28th Septem ber
2003 [UCTE (2004a)] and 15 millions E uropean households were disconnected during the incident o f the 4th
N ovem ber 2006 that originated in G erm any [UCTE (2007)].
2.2 NEEDS OF USERS FOR SYSTEM SERVICES 87
directly related to the security of the system. Hence, quality and security can be separated
only for a small set of quality parameters.
In conclusion, frequency and voltage controls impact the quality of power supply.
Therefore, the quality has to be included in the objective function for system sendees.
2.2.3 Optimal utilisation of resources

A system with appropriate levels of security and power quality is not necessarily run in an
optimal manner. For example, more reactive power injected at the terminals of a transmission
line can increase the active power transit capacity (see section 1.5.4). Similarly, if some
transmission capacities are reserved to allow the supply of ancillary services, less power for
energy can be transmitted. Moreover, the provision of ancillary sendees plays a role in the
amount of losses and impacts the aging of infrastructures. More generally, the consequences
on the various resources of the power system have to be taken into account while using
ancillary services in order to use the resources of the system in an optimal manner.
2.2.4 Specification of the needs

Because system sendees are public goods, it seems logical that an independent party with a
system perspective, such as the system operator (SO), be in charge of specifying the needs
for system sendees (see section 1.2). Ideally, this specification should take into account the
expectations of all the users in terms of the quality of sendee (i.e., reliability, power quality
and utilisation) that they expect for a given period, as well as the price that they are willing
to pay for this quality. Since different qualities of sendee cannot be delivered with the
current technology, the SO must define a single quality of sendee that represents a
compromise between the expectations of all the users using an objective function39. This
quality of service should then be converted into the quantity, quality and location of system
sendees that the SO has to provide during the period considered.
Unfortunately, this ideal procedure is very difficult to implement in practice. First,

since it is impossible to ascertain the expectations of all the users, a representative cross-
section of the population has to be chosen instead. Second, defining the quality of service
39 In particular, in interconnected systems, policies regarding the necessary system services m ust be
decided in coordination w ith the other systems, because the security o f the w hole interconnected pow er
system is at stake.
accurately is difficult and translating this definition into the specification of a quantity and
quality of system services is even more difficult, as will be shown in sections 2,3 and 2.4.
Third, system services, especially frequency control, are shared by all the areas of an
interconnected power system (see section 1.4). Therefore, all system operators within an
interconnected system should have coherent policies, which may be difficult if users have
contractory objectives. Last, the ideal objective function is difficult to implement.
A first objective function that can be adopted is the cost-effectiveness (also called
the rational-buyer) approach, which tries to minimise the procurement cost for an ex-ante SS
specification. Obviously, diis simple policy does not guarantee that a Pareto-efficient SS
specification is attained (see section 1.3.1). However, it is very easy to implement because
the procurement cost is usually better known than the value of system services.
On the other hand, the cost-benefit approach tries to maximize the global welfare, i.e.
the sum of the individual profits and surpluses. This situation is always Pareto-efficient in
theory [Kirschen and Strbac (2004a)]. However, determining the surpluses or the profits of
the participants depending on the system services can be quite complex, because users give
different values to security and these values are hard to define, as shown previously. In
addition, it often implies a probabilistic computation, which is computationally demanding
(e.g., the Monte-Carlo method) [Bhattacharya and Zhong (2001)].
In other words, the cost-effectiveness objective function only takes into account the
cost of system services, while the cost-benefit approach takes into account both the cost
and the value of system services. The latter objective function is the most desirable, but it is
also the least practical. Therefore, the optimisation problem is often solved by subdividing
the system into subsystems (e.g., transmission, energy, frequency control...), then by
optimising the necessary system services separately with a cost-benefit objective function,
and lastly by refining the needs to take into account the dependencies.
2 .3 S p e cifica tio n o f th e Q uality of A ncillary S e r v ic e s
The first task required to fulfil the expectations of users in terms of system services is to
define the ancillary services that are able to meet the needs. As the definitions of AS differ
widely from the classical energy product by many ways, this section investigates how to
separate AS in different products according to various specifications (namely: optimal,
2.3 SPECIFICATION OF TH E QUALITY OF ANCILLARY SERVICES 89
elemental')', functional, actual and standardised) and therefore to be able to specify the
quality of ancillary services needed.
2.3.1 Optimal specifications

A power system usually has numerous potential AS providers with different capabilities.
Ancillary sendees that can be provided are much more differenciated than energy
products40. The differentiation for each characteristic of a product41 can be either vertical or
horizontal. Vertical differentiation is related to the quality of the product by itself (the common
understanding of the quality is the same everywhere, such as an oily fislTn chip is of lower
quality than a grilled fish). O n the other hand, horizontal differentiation is related to the
different perception that two users may have of the same product (e.g., a fish A chip cooked
in Boston has more value in Boston than in Paris). In particular, frequency products are
mainly vertically differentiated, whereas voltage control ancillary services tend to be
horizontally differenciated.
As product differenciation reduces competition, a policy maker may want to

liomogeneise products by for example: (a) reducing the product space; (b) fixing prices; (c)
aggregating firms [Tirole (1988)]. On the other hand, removing differentiation may lead to a
decrease in quality down to the one providing the lowest quality (assuming that quality has a
cost).
Optimal specifications have to be found. They correspond to a trade-off between

too general (i.e., a few differentiated AS products) and too precise (i.e., a lot of
differentiated AS products) specifications. This trade-off differs from a power system to
another because motivations may vary (e.g., a system may want to give priority to the
reduction of the transaction costs). General specifications have the advantages of increasing
the number of potential providers and fostering innovation. On the other hand, precise
specifications take into account the specificities of the providers, increase the understanding
of the AS but can potentially increase the complexity of trading (see Table 2.1). Note that
precise specifications are not always a barrier to innovation because tight rules may force
the players to find innovative solutions that follow new rules at the least cost. For example,
140 M ore generally, w hen getting closer to the real-time, the products are getting m ore and m ore
differenciated (e.g., base/peak, dien h o u r by hour, then 15 m in ...).
41 Several products can be bundled to obtain a given bundle o f characteristics.

older wind farms cannot withstand voltage sags as severe as those that more recent ones
can withstand, because of the new rules that have been imposed.
T able 2.1: G eneral versus precise sp ecification s o f AS qualities
Specifications
General Precise
Increase the number of providers for a given
+++ 4”
specification
Foster innovation +++ ++
Reduce transaction costs ++ +
Recognize the capabilities of providers + +++
Help understand the behaviour of providers + +++
The SO should specify the ancillary services needed for both the long and short
term. Regarding the long term, each power system has to define at the time of connection of
new users (i.e., as part o f the connecting conditions) the compulsory quality needed. A good
principle is to avoid scarcity for ancillary services to guarantee the adequacy of the system,
but without making the connection o f new participants unreasonably expensive (see
section 4.3.1). A good practice is probably to implement a basic requirement for every unit
and to allow the grouping of a few devices to provide the final required quality. In case of
compulsory provision, these connecting requirements should be remunerated in order to
have a transparent energy price and to avoid distortion of competition between two
producers that were connected at different times under different rules. Lastly, it may be
useful for developers to benefit from standard connecting requirements across several
systems, because this will reduce the cost of procuring standard devices (e.g. a standard
alternator for a given power). Note that since frequency is shared by all the participants,
frequency control ancillary services are more likely to benefit from standardised connection
rules than voltage control ancillary services.
In the short term, the SO has to define the global specification for the system (see
section 2.2.4). To fulfil this need, it can either require a standardised quality for each
provider, or the quality can be part of the provider’s offer. The choice will impact the
market design (see section 4.4.1). In practice, most systems define a standard quality that
any provider has to meet if it wants to provide the service considered (see section 2.3.4).
Such a practice limits the number of participants but eases the settlement. Some devices
could be grouped in order to achieve the requirement together (but without necessarily
fulfilling the requirements alone). In other words, several devices could be considered as
one AS provider.
2.3.2 Elementary specifications

As explained in sections 1.4 and 1.5, the frequency is shared by the whole synchronous
system, while voltages are local to each node of the system. Frequency and voltages can be
used as indicators to monitor the global balance between production and consumption of
active power, and the local balances between production and consumption of reactive
power.
To maintain the frequency and voltages within their defined ranges and therefore to
keep the balances of active and reactive powers, devices with one or more of the four
following capabilities can be used:
>4 Devices that are able to control the active power and therefore mainly help control the
global frequency:
■ Up active power balancing, devices that increase the active power injection or decrease
the active power absorption;
■ Down active power balancing, devices that decrease the active power injection or
increase the active power absorption.
Y Devices that are able to control the reactive power and therefore mainly help control
local voltages:
■ Up reactive power balancing, devices that increase the reactive power injection or
decrease the reactive power absorption;
■ Down reactive power balancing, devices that decrease the reactive power injection or
increase the reactive power absorption.
Such devices basically consist of two main components. First, the available resource
(e.g., the headroom of a generator) and the different actuators (e.g., the exciter of a
generator) form the active part of the device that we will call capability of control. This
capability can be discrete (e.g., a capacitor bank) or continuous (e.g., a synchronous
condenser), which leads to different impacts on the stability of the system. Most of the
time, discrete capabilities of control are cheaper and simpler, but usually stress more the
system during switching (see section 1.6).
Second, the capability of control has to be governed by a controller. Controllers can

be gathered in two families. On the one hand, independent controllers are fully independent
and respond to local conditions. The main advantage of these controllers is their
robustness, because they just rely on themselves. However, they can be difficult to design.
The speed control of a generating unit is a good example of an independent controller (see
section 1.4.4). On die other hand, dependent controllers rely on data sent by other devices.
Obviously, such controllers simplify the coordination and can facilitate bilateral contracts.
The main drawback of such devices is that they rely on the transmission of data, which may
Increase die risk of failure and may be more expensive. The Internet technology helps
develop the implementation of such controllers [Kirby and Hirst (2003)]. Nevertheless, the
security of the data channel is an important issue because an Internet server does not work
without electricity and is sensitive to cyber attacks. As illustrated in Figure 2.1, a dependent
controller may be controlled by a central entity (e.g., in the case of secondary frequency
control, see section 2.3.3) or may communicate directiy with other devices (e.g., a generator
following a specific load, as described by Nobile et al. (2001), and used for example in
Sweden). In any case, because of the complexity of electrical systems, controllers cannot be
completely autonomous. Human oversight is always included at some point in order to
maintain the stability and the security of the system, even if power systems tend to be more
and more automated (e.g., concepts of the European “smart grid” or the American
“intelligrid”). In some cases, the capability relies solely on sole human judgment, so there is
no specific controller.
Controller
Main controller
device A ,
Controller
Controller Controller device B J Controller
device A , Controller device C , device C
device B J
(a) (b)
F igure 2.1: (a) C entralised and (b) decentralised d ep en d en t controllers
In conclusion, an ancillary service consists of a capability of control and a controller.

A capability can be a combination of four different types: active or reactive power and up
or down. This specification is very general, so it is useful only to identify devices that can
potentially provide ancillary sendees. The following sections improve these elementary
specifications.
2.3.3 Functional specifications

From a functional point of Hew, frequency and voltage control ancillary sendees can be
classified into six categories commonly agreed by the whole community, but with different
vocabulary (see section 2.3.4.1). The idea is to separate ancillary services in function of their
controllers, as shown in Table 2.2.
T able 2.2: C apabilities and controllers related to fu n ction al ancillary services
Functional ancillary service Capability Controller

Primary up & down independent
Secondary up & down dependent centralised
Tertiary up & down none
2 . 3 . 3 . 1 F r e q u e n c y co n tr o l
Figure 2.2 provides a schematic representation of the three functional frequency controls
provided by a generating unit. Primaryfrequency control is a local automatic control that adjusts
the active power generation of the generating units and the consumption of controllable
loads to restore quickly the balance between load and generation and counteract frequency
variations. In particular, it is designed to stabilize the frequency following large generation
or load outages. It is thus indispensable for the stability of the power system. All the
generators that are located in a synchronous zone and are fitted with a speed governor
perform this control automatically (see section 1.4.4). The demand side also participates in
this control through the self-regulating effect of frequency-sensitive loads, such as induction
motors (see section 1.4.3), or the action of frequency-sensitive relays which disconnect
some loads at given frequency thresholds42. However, this demand-side contribution is not
always taken into account in the calculation o f the primary frequency control
response [UCTE (2004b)]. The provision of this primary control is subject to some
constraints. Some generating units that increase their output in response to a frequency
42 F o r small frequency thresholds, loads usually participate on a voluntary basis, as for example in
Australia JN EM M CO (2004)1- H owever, for large frequency excursions, the system operator has to perform
com pulsory load-shedding through a defined protection scheme [e.g., U C T E (2004b)]. Lastly, we can also
imagine systems in w hich additional loads are connected as a function o f frequency.
drop cannot sustain this response for an indefinite period of time. Their contribution must
therefore be replaced before it mns out.
Secondary frequency control is a centralised automatic control that adjusts the active
power production of the generating units to restore the frequency and the interchanges with
other systems to their target values following an imbalance [Jaleeli et al (1992)]. In other
words, while primary control limits and stops frequency excursions, secondary control
brings the frequency back to its target value. Only the generating units that are located in
the area where the imbalance originated should participate in this control, as it is the
responsibility of each area to maintain its load and generation in balance (see section 1.4.7).
Note that loads usually do not participate in secondary frequency control. Contrary to
primary frequency control, secondary frequency control is not indispensable. This control is
thus not implemented in some power systems where the frequency is regulated using only
automatic primary and manual tertiary control. However, secondary frequency control is
used in all large interconnected systems because manual control does not remove overloads
on the tie lines quickly enough. For example, Figure 2.3 shows the action of the French
secondary frequency control on a day. The regulation signal is comprised between —1 and 1,
and the deployed secondary frequency control power depends on the reserved capacity.
Tertiaryfrequency control refers to manual changes in the dispatching and commitment

of generating units and loads. This control is used to restore the primary and secondary
frequency control resemes, to manage congestions in the transmission network, and to
bring the frequency and the interchanges back to their target value when the secondary
control is unable to perform this last task. Some aspects of tertiary control relate to the
trading of energy for balancing purposes. This thesis does not deal with these aspects
because they do not represent a service provided to the TSO by the market participants but
a mechanism for the participants to balance their financial positions.
By counting the number of oscillations o f the alternative current of the electrical

supply and by knowing the nominal frequency of the network (e.g., 50 or 60 Hz), one can
estimate the time. However, since the frequency of the system actually varies over time, the
time given by the electrical network, also called synchronous time, may not be accurate.
Therefore, frequency control can be used to perform the so-called time control. This historical
control was very useful when a lot of synchronous clocks were used, in particular for trains.
However, with the development of quartz- or GPS43-based devices, synchronous clocks are
now getting rare, though still used by some household appliances such as washing
machines [Bosch (2006)],
A certain amount of active power, usually called frequency control reserve^ is kept
available to perform these controls. The positive frequency control reserve designates the
active power reserve used to compensate for a drop in frequency (it is thus provided by
devices with up active power balancing capabilities, as described in section 2.3.2). On the
other hand, the deployment of negative frequency control reserve helps decrease the
frequency.
Primary control
(automatic)
' prim ary control
= APL Governor
GO produced Interconnected
Generator
network
tertiary control
Secondary
Tertiary control controller
s e c o n d a r y control
(managed by TSO &
interconnection 0
generating unit)
Secondary control
(managed directly by TSO)
F igure 2.2: T h e three functional frequency controls con sid erin g a gen eratin g unit
43 GPS stands for Global Positioning System. This system is based on a netw ork o f satellites that
send inform ation to ground devices, such as the universal time or the position o f the device.
1200 iaoo
T*npt
Figure 2.3: Frequency and French regulation sign al on 31 M arch 2005 [T esseron (2008)]
2 .3 .3 .2 S p in n in g r e s e r v e
The term “spinning reserve” is very popular to designate some frequency control reserves.
In particular, many authors use this term without defining it because they assume that its
meaning is obvious and unambiguous. However, a partial survey of the literature produces
very different definitions:
Y British Electricity International (1991): “the additional output which a part-loaded

generating plant is able to supply and sustain within 5 minutes. This category also
includes pumped-storage plant [...] operating in the pumping mode, whose demand can
be disconnected within 5 minutes”;
Y Wood and Wollenberg (1996): the total synchronised capacity, minus the losses and the
load;
Y Hirst and Kirby (1998): “generators online, synchronized to the grid, that can increase
output immediately in response to a major outage and can reach full capacity within 10
minutes”;
Y Zhu et al. (2000): “the unloaded section of synchronized generators that is able to
respond immediately to serve load, and is fully available within ten minutes”;
Y NERC (2006): “Unloaded generation that is synchronized and ready to serve additional
demand”.
These definitions disagree (or remain silent) on some important issues:

Y Who provides spinning reserve? Is it limited to generators or can the demand-side
participate?
Y What is the time frame for responding to a request? When should it start and end?
Y How is this reserve activated? Does it happen automatically or is it only done at the
request of the Transmission System Operator (TSO)?
Therefore, a definition which fits all systems would be helpful. First, it seems
essential to remove the idea of time, as each system has its particularities. Second, there is a
system operator in any system. Third, the term spinning reserve is used in the optimization
literature as a capacity that can be re-dispatched (scheduling objective), which can be
different from the operational capacity (stability objective). Fourth, it seems interesting to
get detached from the terms such as “generator” or “demand-side”, to reduce ambiguities.
The following definition is therefore proposed in this thesis: the spinning reserve is the unused
capacity which can be activated on decision of the system operator and which is provided by devices which are
synchronised to the network and able to affect the activepower.
Some important comments must be made on this definition:

Y The beginning of the reserve deployment and the duration of the deployment do not
appear in the definition, as they are not relevant for a general definition of the spinning
reserve. In fact, each country has its own definition, depending on parameters such as
the size of the synchronous network or the market structure (see section 2.3.4);
Y The primary frequency control reserve, which is not controlled by the SO, has to be
excluded from the spinning reserve. Moreover, the self-regulating effect of the loads,
which has an effect similar to the primary reserve, is also excluded from the spinning
reserve. In fact, these two items are more important for system stability than for
balancing consumption and production over a longer period of time;
Y The secondary frequency control reserve should be considered as spinning reserve. In
fact, the power deployed by the SO through this reserve equilibrates the consumption
and the production and has to be kept as long as required;
Y The spinning reserve also includes the synchronized tertiary frequency control reserve,
as this reserve is deployed on the instruction of the SO;
^ If a generator decides not to provide reserve, its spare synchronized capacity is not
spinning reserve, as it cannot be activated by the system operator. However, in many
systems, generators have to offer all their spare synchronized capacity in the balancing
mechanism (see section 1.3.2). Therefore, in this case, the system operator has the
possibility to call upon all the synchronized capacity;
S A consumer can provide spinning reserve, if it agrees to be disconnected or to reduce
its load upon request from the SO. For example, pump loads are good candidates for
die provision of spinning reserve from the demand side.
2 . 3 . 3 . 3 V o l t a g e c o n tr o l
Figure 2.4 provides a schematic representation of the three functional voltage controls
provided by a generating unit. Primary voltage control is a local automatic control that
maintains the voltage at a given bus (at the stator terminals in the case of a generating unit)
at its set point. Automatic Voltage Regulators (AVRs) fulfil this task for generating units.
Other controllable devices, such as static voltage compensators, can also participate in this
primary control (see section 1.6).
Secondary voltage control is a centralised automatic control that coordinates the actions
of local regulators in order to manage the injection of reactive power within a regional
voltage zone. This uncommon technology is used in France and Italy [Paul et al (1990),
Corsi et al. (1995) and Vu et al (1996)].
Terliay voltage control refers to the manual optimisation of the reactive power flows
across the power system. This is performed by either modifying the set-point of pilot
voltages in case of a secondary control, or by directly modifying the voltage set-point of the
unit. Tertiary voltage control also includes the startup of additional units and the operation
of other means such as capacitor banks.
Primary control w<

Local voltage
(automatic) pnmarQonfro,5efpo,'n'
Reactive + /O” b A w anted

Voltage j
controller Generator
controller
- tertiary controls
stator
Interconnected
p ro duced network
Voltage
representative
Q s e c o n d a r y + tertiary controls S econ d ary of the zone
controller
pilot w a n te d
s e c o n d a r y control setpoint
Secondary control
(managed by TSO) tertiary control
-i Tertiary control (managed by TSO)
F igure 2.4: T h e three fu n ction al voltage controls con sid erin g a generating unit
2.3.4 Actual specifications

The functional specification presented in section 2.3.3 classifies ancillary sendees into six
categories and is sufficiently general for any power system. However, the actual
specifications are more detailed in practice. This section compares the features and the
values of the actual specifications in different power systems amongst those listed in
Table 2.3. This comparison is centered on the UCTE interconnected system44 to illustrate
the differences that can coexist within a single synchronous zone. The values for Great
Britain (which is not synchronized with the UCTE), the Unified Power Systems of Russia
(UPS) and NERC43 put this comparison into perspective. Only large generating units
connected to the transmission network are considered here. Exceptions granted to small
units and distributed generators are beyond the scope of this survey. The issues related to
the transmission of data to and from the provider of ancillary sendees are not included in
this survey either.
Note that similar surveys have been performed by Eurelectric (2000), Cali et al
(2001), Arnott et al (2003) or Raineri et al (2006). However, the present sui'vey differs from
44 U C T E is the U nion for the C o-ordination o f Transm ission o f Electricity, which is the association
o f the T SO s operating w ithin the synchronous system o f mainland E urope. T he U C T E establishes the
security and reliability standards fo r this interconnected system.
45 N E R C is the N o rth A merican Electric Reliability Council. Like the U C T E , N E R C is n o t a system

operator and hence does n o t intervene in the operational working o f the system.
these in terms o f the systems studied and the framework used to compare the features of
the ancillary services. In addition, the reader should keep in mind that specifications change
over time. Therefore, such a survey should be updated regularly to remain relevant.
T a b le 2.3: B a sic inform ation on system s studied
System Abbreviation Regulator TSO

Australia AU AER NEMMCO
Belgium BE CREG Elia
California CAL FERC CAISO
France FR CRE RTE
EnBW, E.ON, RWE
Germany DE BNA
and VET
Great Britain GB Ofgem National Grid
The Netherlands NL DTe TenneT
Electricity
New Zealand NZ Transpower
Commission
PJM PJM FERC PJM ISO
Spain ES CNSE REE
Sweden SE Stem SvK
Federal tariff
UPS UPS SO-CDU
service
2 .3 .4 .1 V o ca b u la ry
The profusion of terms used for ancillary services may lead to some misunderstandings and
confusion when one tries to compare services from different power systems or
jurisdictions. Table 2.4 classifies the services considered in the survey with the help of the
functional framework developed in section 2.3.3. The names are shown in the original
languages to avoid misinterpretation. In this table, for each type of control, the reserves are
ordered from the fastest (left) to the slowest (right). Reserves within a given column are
thus similar. Within a system, terms are classified from the most general (up) to the most
specific (down). A dashed line means that positive and negative reserves have different
names. For instance, in Great Britain, the terms “primary response” and “secondary
response” are used for the positive primary frequency control reserve and “high frequency
response” for the negative one. In some cases (e.g., Australia), the name of a service is used
as the corresponding reserve does not have a name.
A number of observations can be drawn from this Table. First, it is clear that care
must be taken when comparing services defined in different systems. For example, the
terms “secondary response” (GB), “secondary control reserve” (UCTE) and “secondary
2.3 SPECIFICATION OF THE QUALITY OF ANCILLARY SERVICES 101
reserve” (PJM) describe three completely different sendees. Even the term “reserve” may
lead to some misunderstandings. For example, in Australia, “reserve” designates
investments in generation capacity and not the frequency control reserves as defined in this
thesis. “Spinning reserve” is another example of an ill-defined term, as explained in
section 2.3.3.2.
Besides differences in terminology, there are also significant differences in

implementation. In Sweden, Great Britain, New Zealand and Australia, the primary
frequency control reserves have been divided in different categories. O n the other hand, in
the other systems, a single reserve is defined for this type of control. This difference may be
explained by the fact that smaller systems are subject to larger frequency deviations than the
large interconnected systems of North America, mainland Europe and Russia. Reserve that
can be used for very fast primary frequency control reserve is therefore more valuable in
these smaller systems and making a distinction between different categories of primary
reserve is technically and commercially worthwhile. Only one type of reserve is defined for
performing secondary frequency control, except in Sweden and Great Britain where
secondary control is not used. Since die whole of the British system is operated by a single
TSO (and is not synchronous with die Western European interconnected system),
secondary control is not needed to correct deviations from interchange schedules. While
Sweden is interconnected with other countries, its TSO can rely on numerous fast manual
tertiary control offers provided by the large hydro generation capacity (see section 2.4.3.1).
Tertiary frequency control reserves do not lend themselves to easy comparisons because
TSOs have adopted widely different limits for deployment times and various approaches to
the treatment of non-synchronized generating units.
From the perspective of providers of voltage control services, it is convenient to

divide the production of reactive power into a basic and an enhanced reactive power
sendee. The basic or compulsory reactive power service encompasses the requirements that
generating units must fulfill to be connected to the network. The enhanced reactivepower service
is a non-compulsory sendee that is provided on top of the basic requirements. The
terminology of voltage control is much more uniform than for frequency control and does
not need to be discussed further.
T ab le 2,4: V ocabulary u sed to n am e frequency control reserves in various sy stem s46
Primary Secondary
frequency control frequency control Tertiary frequency control reserves
reserves reserves
Reserve
Frequency response Operating reserve
> 30 min
PJM Regulation Primary reserve Secondary reserve
Synchronised Quick start
reserve47 reserve
Replacement reserve and

(no name) Operating reserve
supplemental energy
CAL Regulating reserve Contingency reserve
Non-Spinning
Spinning reserve
reserve
Resen t depuissancepour Reserve depuissancepour
BE riglageprimaire reg/age seenndaire
Reserve depuissancepour rig/age teniaire
DE Primiimgelresem Sekunddrregelnserve Minutenreserve48
Reserva de regulation Reserva de regulation

ES primaria secundaria
Reserva de regulation iertiara
Reserveprimaire Reserve seenndaire Reserve tertiaire

FR Reserve a
Reserve tertiaire
Reserve tertiaire rapide 15 minutes echeance ou
compUmentaire 30 minutes
dijfirie
NL Primaire reserve Seenndaire resene Tertian reserve
Operating reserve (does not exist) Operating reserve Contingency reserve

Response Fast start Short-time Balancing mechanism start-up
GB
Primary | Secondary'
High frequency
Frckvcnsstyrd
SE Normaldriftsrescrv and (does not exist) Seven different types of reserves
Stomingsreserv
U PS Penitcbnyi reserve Vtoritchnyi reserve Tretitchnyi reserve
Regulating services and

Contingency services Short-term capacity reserve
AU network loading control
Fast Slow Delayed
Frequency' regulating (or

Instantaneous Reserves (no name)
keeping) reserve
NZ
Fast j Sustained
Over frequency
46 References - AU: N E M M C O (2001), N E M M C O (2004) and N E M M C O (2005); B E: Belgium

econom ic departm ent (2002); CAL: CAISO (2005a) and CAISO (2005b); D E : B D E W (2007a); ES: Spanish
industry and energy departm ent (2006b); FR: R T E (2004); GB: N ational G rid (2007) and N ational G rid
(2008a); N L: T enneT (2003); N Z : N Z IE R (2005); PJM: PJM (2006); SE: N ordel (2004).
47 In 2006, PJM changed the denom ination “spinning reserve” to “ synchronized reserve” .
48 G erm any recently rem oved Studenreserve and N otreserve from tertiary frequency control
reserves.
2 . 3 . 4 . 2 P rim a r y f r e q u e n c y co n tr o l
Policy makers tend to specify various parameters that have a significant effect on primary
frequency control. The most important technical parameters for frequency-related ancillary
services are the deployment times. The maximum amount of time that can elapse between
the request from the TSO and the beginning of the response by the sendee provider will be
called the deployment start. Full availability is the maximum time that can elapse between the
moment when the provider receives the request and the moment at which it delivers its full
response. Lastly, deployment end is the maximum amount of time during which the service
must be provided starting from the time of the request.
The accuracy of the measurements is another important issue because it affects the
efficiency of the control and die payments to the providers. For example, if the
instrumentation at a generating unit overestimates the frequency, its response to frequency
deviations will be inadequate and the generating unit may be paid more than what it
deserves. However, it is generally in the interest of producers to measure accurately so they
can argue more persuasively with the TSO in case of a dispute.
The droop of a generating unit is an important parameter of primary frequency control

(see section 1.4.4). A lower droop increases the response of a unit but would cause more
stress in die generating unit as it would react more strongly to each deviation. On the other
hand, a unit with a low droop is more likely to succeed in switching to islanded mode in
case of a major disturbance and tends to reduce the quasi-steady state frequency deviation
following an imbalance. Adjusting the droop is not always easy because it often requires
that the plant be shut down.
The frequency deviation for which the entire primay frequency reserve has to be deployed is also
an important parameter. Indeed, using this information, the value of the droop and
Equation (1.5) of p. 58, one can calculate the maximal share of the nominal power output
that a generating unit must keep in reserve to provide the required primary frequency
control power. For example, Table 2.5 shows the frequency deviation for which the entire
primary frequency reserve is deployed as a function of the droop and the primary frequency
control reserve.
The frequency characteristic represents the total action of the primary frequency control
provided by generators and the self-regulating effect of the load (see section 1.4.6). In
North America, the terms of “Beta” or “frequency governing characteristic” are preferred
to designate this parameter [Ingleson and Ellis (2005)].
The insensitivity of a primary controller is the frequency band within which the
controller does not change its output. Two insensitivities should be distinguished: the non-
intentional insensitivity, which is intrinsic to the controller, and the intentional insensitivity,
which is added on purpose. In Europe, the non-intentional insensitivity is sometimes simply
called insensitivity while the intentional insensitivity is called dead band. The sum of these
two insensitivities gives the total insensitivity. If two generators have different total
insensitivities, it means that the one with the smallest insensitivity will participate to the
primary frequency control before and probably more frequently than the other one.
T ab le 2.5: F requency deviation for w h ich the entire primary frequency reserve is d ep loyed as a
fu n ction o f th e droop and the primary frequency control reserve
D roop s'G
__________________________ 4% 5.7 %
Reserve relative___ 2 %_____40 mHz 57 mHz____
to 7 % 140 mHz 200 mHz
Table 2.6 summarizes the primary frequency control parameters in nine different
systems'19. Since units providing primary frequency control must respond immediately to a
change in frequency, the “deployment start” does not appears in this Table.
Deployment times are the same all across the systems within the UCTE because it is
important to have a homogeneous response in the synchronous zone. On the other hand,
NERC does not make any recommendations on the value of this parameter. A faster
primary frequency control is used in Great Britain because the size of this system makes it
more susceptible to frequency variations.
The frequetuy characteristic of the Eastern interconnection of North America was

around 31,000 M W /Hz in 2004 [Ingleson and Ellis (2005), NERC (2005)]. The
requirements for the national systems within the UCTE are estimated based on their annual
electricity production [IEA (2005)] and the total UCTE’s requirement
(20,570 M W /Hz [TenneT (2006)]). Lastly, the frequency characteristic in Great Britain is
49 T he following abbreviations are used: N o rec.: no recom m endation; Pri., Sec. or Hi.: primary,
secondary o r high frequency response; I: intentional; N I: non intentional; T: total.
estimated from requirements in the tendering process for primary frequency control
sendee [National Grid (2008b)].
Within the UCTE, the fu ll deployment of the primmyfrequency reserve must occur before a
deviation of ± 200 mHz; has happened. Hence, TSOs with a tight requirement for the droop
(i.e., a small droop, such as in France) would need to reserve a large headroom for their
generating units if they want to provide their primary frequency control continuously until
+ 200 mHz. However, in practice, units with a small droop attain their maximum before the
limit of ± 200 mHz, so they do better than what the UCTE recommends. In Belgium and
Germany, no general condition is applied to the droop as this parameter is agreed between
the generators and the TSO during the procurement process (see Chapter 4). Lastly, the full
deployment of the British primary frequency reserve happens for larger deviations than in
the UCTE because the British power system has a smaller inertia.
Policies on controller insensitivity are similar in all systems, except in Great Britain,
where the requirement is less strict. NERC has recently changed its policy and no longer
recommends any insensitivity. Before, a total insensitivity of 36 mHz (30 mHz if a 50/60
coefficient is applied to account for the difference in nominal frequency) was
required [NERC (2004)], which is three times the UCTE’s requirement (10 mHz).
However, the UCTE’s requirement applies only to the controller [UCTE (2004b)], while
NERC former recommendation was for the “governor” [NERC (2004)]. This last term is
not formally defined, but is commonly understood to include both the primary frequency
controller and the actuators (e.g., jacks and valve). Therefore, in this table, the insensitivity
of the controller only and the insensitivity of the whole generator unit are not distinguished.
Finally, the accuracy ofthefrequency measurement is not addressed explicitly in most systems.
Table 2.6: Technical comparison of primary frequency control parameters in various system s 106
NERC UPS UCTE DE FR ES NL BE GB

French economic Belgium
Spanish industry National Grid
NERC (2004), NERC UCTE (2004b), BDEW (2007a), and industry TcnneT (2003), DTc economic
'
R eferences and energy (2008a), National
(2005), NERC (2006) TcnncT (2006) BDEW (2007b) department (2003), (2005), TcnncT (2006) department
department (1998) Grid (2008b)
l R TE (2004) (2002)
o
o
VI
•'d y
tn
VI
VI
ft. oo
Full availability No rec. <30s <30s <30s <30s

CHAPTER 2
T'S
2 ^ 2
VI VI VI
lft 1/1 V5
Pri.: > 30 s
Sec.: > 30 min
o
c
c
c
c
6
c
Al
c
Al
Vs
tn
m
in
Al
Al
Al
Al
m
in
U1
At
Deployment end
E
C
■E
■g
Mi.: as long as
required
Frequency | 10 %o f the balancing

authority’s estimated Variable
o0
It
N
T
©
O
U
01
U
No rec, 20,570 MW/Hz
-T
N
~ 4,200 MW/Hz
N
characteristic ~ 600 MW/Hz
oOO
1 yearly peak = 2,000 MW/Hz
requirem ent demand/Hz
5 % in 2004; no rec. 5-60 MW: 10 %

Droop of generators 3-6 % No rec. No rec. 3-6 % < 7.5 % No rec. 3-5 %
anymore > 60 MW: 4-20 %
!
Is an adjustable 5-60 MW: No rec.
2
o
No rec. Yes No rec. Yes Yes No rec. Yes

droop com pulsory? > 60 MW: Yes
i I
Accuracy of the ± 10 mHz
recommended, Within Within
§
?
c
N
O
No rec.
-H
frequency No rec. No rec. No rec. No rec.
+ 5 mHZ ± 10 mHz ± 10 mHz
m easurem ent desirable
T: ± 10 mHz T: ± 10 mHz
T: ± 36 mHz in 2004;
-fl
T: ± 10 mHz NI: No rec. T: ± 10 mHz NI: No rec. T: ± 10 mHz T: ± 15 mHz

DELIVERY OF ANCILLARY SERVICES
Controller no rec. anymore

2
NI: No rec. I: should be NI: No rec. I: should be NI: ± 10 mHz NI: No rec.
NT. No rec.
^ si
insensitivity
2 2 o
H ci «
+1
I: No rec. compensated I: ± 0 mHz compensated I: No rec. I: No rec.

? A 2
G y ^ o
g& M g °^
+i ? +i V
I: No rec.
0 h-I
S o o +!
within the zone within the zone

o
^
&
2iT)
Full deployment for

©
*ri
^
1©
©
©
O
+l
to
iS
m
bi
2
0
2
u
-H
1
o
R
+i
©
©
o©
NI
01
CO to
>4 <
+1
N
± 200 mHz
fN
or before a deviation ± 200 mHz ± 200 mHz

o
§
%
A
&
P. M E
S« £hL X
© ©O
1 E S
ai-> *j
h
© L©O
of:
1c
X
+l
©
©
O
O
iS
N
j®
ir>
2 . 3 . 4 . 3 S e c o n d a r y f r e q u e n c y con trol
Secondary frequency control can actually be organized in three ways: centralised, pluralistic
and hierarchical [UCTE (2004b)]. In a centralised organisation, the control is performed by a
single controller for the whole control area. In a pluralistic organisation, the system is split into
independent zones, each of which having its own controller and regulating capacity. In a
hierarchical organisation, the organisation is similar to the pluralistic approach but a main
controller coordinates the action of all the other controllers.
The Area Control Error (ACE) of a zone g, which charaterises die imbalance of the
zone, is calculated as follows according to the UCTE (A C E ^crE ) or NERC (A C E ^brc):
CEf F^hikrmmiutionm îiiltrammction0 ^ Cfat ft )
A C E nerc = Pjitenvimtrtioum ~~ 0 1073" ft ) ^ME (2>2)
where is the measured value of the total power exchanged by the zone with
other zones (a positive value represents exports). Note that E>leminmthll w is slightly different
from the actual exchanges E>ttK0Himlhii because of measurement errors. f z is the measured
network frequency by the zone, is the target frequency, which can differ from the nominal
frequency when controlling the synchronous time (see section 2.3.3.1). The additional term
Jl!B introduces a very small correction factor. This factor compensates the difference
between the integration of the instantaneous power exchanged and the demand’s energy
measurements [NERC (2006)]. K z is the IC-factor of the control area (in M W /Hz and
positive) and Bz the frequency bias setting (in W/0.1 Hz and negative). K < and Bz are an
overestimate of the frequency characteristic of the zone A~ (see section 1.4.6). An

underestimate would indeed lead to a conflict between the primary and secondary frequency
controls. To illustrate this fact, Table 2.7 shows the impact of the K-factor on the
secondary frequency control power when a disturbance appears outside the control area, i.e.
when Pmttramuedion
\ m
—ITiiikrwmKCtiou 0 is equal
1
to —AM/p
\J m
—Jf,)
I /
(in
\
other words,5 the imbalance of
tire area is only due to the action of the primary frequency control). Note that a positive
term K z —A z indicates that K z is an overestimate of the frequency characteristic A".
Furthermore, when the term f z —f t is negative, the primary frequency control power
increases and Ace versa. Lastly, when siC E UCfE is negative, the secondary frequency
control power increases and Ace versa. It is clear that conflicts appear only when K z is
inferior to A z .
T able 2.7: Im p act o f the K-factor o n secondary frequency control
A CP — Primary Secondary Conflict?

K '-Az fl-ft control control
(k' - A ' ) x ( £ - / , ) power power
>0 <0 <0 Increases Increases No
>0 >0 >0 Decreases Decreases No
<0 <0 >0 Increases Decreases Yes
<0 >0 <0 Decreases Increases Yes
When the target frequency f of the power system is different from the nominal
frequency fn, a part of the primary frequency control reserve is undermined. In fact, primary
frequency controllers are based on the nominal frequency (see section 1.4.4). Therefore, this
use of primary frequency control has to be compensated. For example, requirements for
primary frequency control reserves in France are changed as follows50 [RTE (2004)]:
A Iff — 49.99 Hz51, the reserves are increasedhy 150 MW;
A If ft — 50.00 Hz, no change;
A If j t — 50.01 Hz, the reserves are decreased by 150 MW.
Secondary frequency control usually relies on a Proportional Integral (PI) controller,

filters and heuristics to bring the ACE back to zero. The exact configurationvaries from a
system to another [e.g., Jaggy and Longley (1991), Jaleeli et al (1992),UCTE (2004b) or
PJM (2007)].
Because of data management problems, the frequency and the power flows through
interconnections are measured with discrete time steps. Therefore, the time steps of
frequency and power exchanges measurements, as well as the cycle time of the controller,
should be considered carefully. Shorter cycle times produce a more efficient secondary
control, but also imply higher data management costs.
50 T he policy to change the requirem ent for primary frequency control reserves as a function o f the
target frequency is not adopted in all systems, which creates security breaches.
51 In G reat Britain, time control is perform ed by changing the nom inal frequency by ±0.05 H z or
±0.10 H z [National G rid (2008a)]. In N o rth America, the target frequency is changed by ±0.02 H z [NERC
(2006)].
The speed of the effective frequency correction by the secondary control depends
on the error si2e. Nevertheless, the frequency normally should begin to come back to its
target value (neglecting the frequency oscillations) no later than the full deployment of the
primary frequency reserve. Indeed, at that time, the balance between consumption and
production should have been re-established using the primary frequency control and the
secondary frequency control should have started its action.
Table 2.852 shows the parameters adopted for secondary frequency control in eight
different systems. Great Britain is not included in this table because secondary frequency
control is not used in that country.
Deployment times in European countries are generally smaller than what the UCTE
recommends, probably to provide some additional margin or to remain consistent with
former local policies. For its part, NERC does not provide any specific recommendation.
However, it imposes some performance criteria through its Disturbance Control Standard
(DCS) and its Control Performance Standard (CPS) [NERC (2006)]. To give an example of
US practice, within PJM, the secondary frequency control power should be deployed within
five minutes and for at least one hour (i.e., the duration of the regulation market) [PJM
(2007)]. Lastly, UPS does not impose a deployment time but sets a minimum gradient that
must be respected.
With a requirement of 0.8 mHz (coefficient of 50/60 applied), NERC recommends

a more precise frequency measurement than the 1.5 mHz UCTE specification. The cycle times of
German controllers are the smallest amongst the systems considered.
Unlike the UCTE, NERC does not make any particular recommendations regarding
the design of the secondaryfrequency controller\ Where this data is available, the proportional term
is very small, while the integral term is about 100 s. Contrary to the frequency characteristic
requirements given in Table 2.6, the K-factor (fC~ or —10Bz) cannot be estimated because
it represents the actual frequency response of the system, which is not publicly available
data. This parameter can be variable in real-time in North America, whereas it is constant in
Europe. As guidelines for the K-factor, UCTE asks for a 10 % overestimation of the
frequency characteristic while NERC recommends the exact value.
52 T he following abbreviations and symbols are used: N o rec.: n o recom m endation; e: accuracy; T:
cycle time; P, I o r PI: proportional, integral and proportional-integral controller.
Table 2.8: Technical comparison of secondary frequency control parameters in various system s 110
NERC U PS UCTE DE FR ES NL BE
Spanish industry
and energy
'
R eferences UCTE (2004b), TcnncT BDEW (2007a), department TcnncT (2003), Belgium economic
NERC (2006) (2006) BDEW (2007c) RTE (2004) (2006a) DTe (2005) department (2002)
Immediate or
om
VI
No rec. <30s
■§
oHI
Deployment start <30s No rec.
C
< 10 s
CHAPTER 2
£ 5 min
< 15 min
VI
As soon as possible
c
a
‘S
VI
VI
U“)
E
No rcc. < 300-500 s
lO
Full availability < 5 min since < 10 min
® &
^ VI
(gradients imposed)
w
15 May 2008
6
-o
Al
m
As long as
Al
m
At
U“i
No rec. As long as required
C
Deployment end As long as required As long as required
C u
5 *o
required j
1
U
u
is«
*3
Tj
Cl
L)
Control organisation No rcc. No rec. Pluralistic Hierarchical Pluralistic
C
Centralised
O
Frequency
N
N
1.0<e< 1.5 mHz s: Unknown
N
vJ h
p 'a
t S
- >
VI L
VI H
v ,H
ii
Vf II
1 »
2 "
ij 'S
VI H
£ «
VI H
^ VI
o VI
T: No rec.
u-> w
m easurem ent T=2s
E xch an ges
VI
0)
e-
VI
fO
e: Unknown s < 0.5%
to £
£ -
mH
VI H
in in
“ VI
II
IT) T-
^ VI
^ 11
— VI
c w
lO "St
T = 4s T: Variable
VC c/3
4
m easurem ent
VI £—
VI
vo
m
M-
Controller cycle tim e 1-5 s 1-2 s 4s j 5s
I or PI, PI, with
S3
Controller type No rec. I or PI depending on the additional PI
regulation zone heuristics
DELIVERY OF ANCILLARY SERVICES
o
o
Proportional term No rec. 0-0.5 Unknown Unknown 0.5 0-0,5
Integral term No rec. 40-120s 50-200 s Unknown 115-180 s 100 s 100-160 s 50-200 s
K-factor for The frequency 110 % of die frequency

s
E
It
V
O
Unknown
^3
Unknown Unknown Unknown

N
900 MW/Hz
VO
O
measuring the A CE characteristic characteristic

2.3 SPECIFICATION OF THE QUALITY OF ANCILLARY SERVICES 111
2.3.4.4 V o ltage control
Contrary to the primary frequency control, the parameters of the primary voltage control
cannot be compared between systems, and most of the time, even between units. In fact,
the parameters of the voltage controller must take into account the configuration of the
surrounding network. However, the requirements in terms of reactive power capabilities can be
compared, as they measure the ability of the unit to control voltage and because one Mvar
from a hydro unit is generally equivalent to one Mvar from a thermal unit. To allow a
benchmarking between systems, the reactive power requirement given at the terminals Q 2 is
converted to the Point O f Delivery (POD) 0 POD(see section 1.5.3).
Sometimes, reactive power requirements may be given for the dimensioning voltage
Udim instead of the nominal voltage of the network Un. The dimensioning voltage is the
actual average voltage at which the generating unit is connected, which may vary from U„.
The possible reactive power/voltage pairs are given by the U /Q diagram of the unit. To
convert from the reactive power requirement at Uflm/ to the one at U/n a typical U /Q
diagram should be used on a case by case basis. In this survey, we assume that U/limand L/„
are equal.
Concerning deployment times, the primary voltage control is immediate and must
operate continuously, as it maintains the voltage at its nominal value and insures stability.
From an accounting perspective, cycle times and accuracy are less important for the
voltage control than for the frequency control, because the utilization cost of reactive
power is generally considered to be too low to require a precise metering [Da Silva et al.
(2001)]. However, for the stability of the system, the measurement accuracy and the
parameters of the voltage controller should be considered carefully.
Most of the TSOs make recommendations regarding the effectiveness of the control\ and
in particular the stability margins [Spanish industry and energy department (2000), Belgium
economic department (2002), French economic and industry department (2003) and
BDEW (2007a)]. However, these requirements were not considered in this survey because it
focuses on features of ancillary sendee controls.
Table 2.9i3 shows the reactive power requirements in eight different systems. While
these requirements are often specified using a complete P /Q diagram, this table can only
show the maximum requirements for Pn.
NERC and the UCTE do not give any recommendation on reactive power
requirements because it is a local problem. UPS requirements are negociated case by case.
Spanish requirements are quite low in comparison to the other systems in the Table.
Therefore, connecting conditions are less stringent on this point in Spain than in the other
systems. However, reactive power connecting conditions do not imply that one system uses
more or less reactive power than another one since TSOs can contract for additional
reactive capabilities or use some of their own reactive power sources (see Chapter 4).
53 T he following abbreviations and symbols are used: N o rec.: n o recom m endation; pf: pow er factor;
Pn: nom inal power; P O D : P o in t O f Delivery; Udim: dim ensioning voltage; Un: nom inal voltage.
Furtherm ore, note that the expressions lagging and leading refer to the stator voltage and the stator current
PQ
O
W 0, jOP
pa SN
s Q
9 cO
u
« P, « «
o P
o <a
"Q Q
o ,C
o. *2
o ”3
C/i 16*1^ c , P-i e

■9■ s8U 1§H§ co
CH nO £P
a ^â
aCL.O
8, .1?1\ oQ
V>
. a
o* &* 3 * o, E s &, a
’o Ph
is a o8 I"3 .9
M l
o a .3
i*\ ^ «o Q n n
ou "=?O So
'j
8P LT5 Q _
a
~ Q P ° uT3
£ O -o s " g
o Q
ao II 9 v.
w ® ,0
■«a * UJ
&
w
H
u
P
3
C\
c4 C/D
a
<u p
3«
H
u
PCi
a
£
*
3
art •8a B4a ^
& 5r ^
rt «u ^
u 4- .
a a a o
o S '3 9 bJD
'5 h a u
& §
9 •§ TJ
3 uM d
■a ’a 55
3 *■ a-
S .3S
pci <i a w &a a
2.3.5 Standardised specifications

As shown in section 2.3,4, actual specifications for ancillary services use time-related
parameters (the deployment times). Since primary and secondary ancillary services rely on
automatic controllers, the time-related specifications make sense only for a given input signal (e.g., a
step function)54. For example, as shown in Figure 2.5, the two deployment times fd h t
and t biirphjmcnl are totally different, whereas it is the same AS provider that is considered in
both cases. Consequently, if two policy makers take different typical input signals, the
capability of a provider cannot be translated directly to the other policy maker. Therefore, a
standardised description may help the providers communicate their capabilities.
Second, actual specifications for ancillary services are with good reason focused on
the quasi-steady state (e.g., frequency characteristic or full deployment for a given frequency
deviation)53. Therefore, the dynamics of the providers are ignored .Providers that react fast are
more useful to the system56 and have to respond more often, which increase their provision
cost (see Chapter 3). Hence, these providers should be entitled to receive a premium for a
provision of AS with an improved quality (see Chapter 4). In addition, the SO is eager to
know the actual dynamics of the AS providers to better understand the reactions of the
system. In particular, after collecting the data on dynamics from all the AS providers, the
SO can run simulations to check whether its security criteria are fulfilled or not.
Lastly, the current frameworks define their requirements for each type ofproviders (e.g.,
conventional power plants, distributed generation, purposely-built devices or loads). In
order to foster innovation, it would be useful for the policy maker to benefit from a
framework that describes any provider.
54 F o r manual ancillary services such as tertiary reserves, the inp u t signal is m anaged by the provider
itself (e.g., it can activite its fast units very close to the end o f the deployem ent time).
55 T h e short response (< 30 s) o f a frequency control provider is linked to stability, whereas the
longer-term response (beyond 30 s) is linked to the active pow er actually provided by the prime
m over [Proctor (2001)].
56 Dynamics that are too fast can also lead to an unstable system.
input
o u tp u t o u tp u t t
t'-deployment
a t tdeployment
b t
(a) (b)
F igure 2.5: R epresentation o f resp on ses o f an AS provider for tw o different inputs
Because of the limits of current specifications, Zhong (2003) and Doherty et al

(2005) have proposed specifications based on the actual d3>namics of the control. Moreover,
an innovative approach for classifying ancillary sendees has been proposed by a Cigre Task
Force (2001). The idea is to convert the AS needs and responses from the time domain to
the frequency domain by applying a Fourier transformation. Therefore, each AS is classified
into different frequency categories in function of its response time, as shown in Figure 2.6.
Still, these innovative methods do not take into account all the specificities of an AS
provider such as deadbands or delays. Therefore, an original specification of frequency and
voltage control ancillary sendees is proposed here and illustrated in Figure 2.7. Primary,
secondary and tertiary controls are only differentiated by the input signal, which is either
computed locally, either sent by the SO or set by the device operator. Lastly, note that
capacitor banks and generators do not provide the same reactive power capability because
the actual reactive power supplied by a capacitor varies with the square of the voltage at its
terminals [Fourment et al. (1995)]. This variation can be taken into account in this
framework by using the tenninals voltage as input.
In all the cases, the device responds to the input signal57 with its own dynamics,
characterised by five parameters. First, the equipments have a delay of response, generally
associated with the transmission of data. This delay can also be very useful for providers
57 AS providers may also respond to several input signals. F o r instance, the active pow er
consum ption o f a load usually depends on b oth voltage and frequency o f the electricity supply [K undur
(1994)].
that need a certain time before being able to react, such as disconnected gas turbines.
Second, all equipments have a certain insensitivity, which is given by the deadband. In
addition, this insensitivity can be used by all-or-nothing providers, such as loads fitted with
a low-frequency relay. Third, the response of the device to the input signal is approximated by a
first order system (see Figure 2.8). Obviously, a generating unit is not a first order system,
but in most of cases, the approximation is sufficient to check whether the security may be
guaranteed. This feature is important because slow and fast controllers have different values
for the SO. If the provider agrees (and is able) to give a more precise description of its
device (e.g. a second order system), this better description would be easily implemented in
the model. Fourth, the process is limited by its maximum and minimum outputs (i.e., the
negative and positive reserves) and by the maximum slope. The fifth and last parameter is the
maximum duration of the control. This maximum duration may be expressed in terms of
energy, which can be very useful for hydro units, storage devices or loads.
It is worth noting that classical control schemes for generators, FACTS and loads
can fit this framework. In the particular case of generators, this structure may be used for
different levels o f active power production, since the dynamics of a unit are a function of its
output. However, it is not clear whether this structure is appropriate for future forms of
regulation such as adaptive commands, neural networks or fuzzy logic.
output
A .<5 -1
Low frequency Medium frequency High frequency
Figure 2.6: C lassification o f AS b a sed o n frequency-dom ain characterisation

Input K
signal
1+r-s
Delay D eadband P ro c e s s Limiter Duration
deadband K, T min, max, slope max

duration
Figure 2.7: P rop osed standardised definition o f quality o f a frequency or voltage control AS
input
output
actual
response
1+ T- s
estimated
response
delay
F igure 2.8: R epresentation o f the actual resp on se o f the group versus its estim ate
This standardised framework can be used to describe any AS provider. All the
stakeholders would thus speak a common language. The particularities of providers would
be recognized and the SO would have a better understanding of the system. In addition, to
help identify types of AS providers that are either abundant or scarce, or to create a
market38, AS providers can be grouped in categories as a function of specific profiles, as
shown in Figure 2.9. However, such profiles require an input signal, so it reduces the
universality of the specification.
58 Allowing product substitution betw een different qualities w ould be necessary in such a market.
input
output
actual
response
Calibre
Dynamic Quasi-steady-state
requirements requirements
F igure 2.9: U tilisation o f a profile to define a category o f AS quality
2 .4 Q uantity o f A ncillary S e r v ic e s
The previous section has provided various frameworks to specifiy the quality of a given
ancillary service. None of the proposed frameworks stands out from the crowd. Therefore,
the chosen specification depends on the constraints of the power system considered. Using
these specifications, this section focuses on the quantity of system (and thus ancillay)
services needed to meet the expectations of users in terms of reliability, power quality and
utilisation. The optimal method is first presented (section 2.4.1), before reviewing and
discussing the practical methods used within the UCTE (sections 2.4.2 and 2.4.3). The
requirements in various systems are then given. Some original methods to define the
appropriate quantity of system services are introduced (section 2.4.4). Lastly, actual
requirements across systems are compared (section 2.4.5).
2.4.1 Optimal definition of the quantity of system services
As shown in section 2.2.4, the optimum quantity of system services is the one that maximizes
the global welfare. However, this optimum is very difficult to reach because many
parameters have to be considered. In particular, the value and cost of system services are
2.4 QUANTITY OF ANCILLARY SERVICES 119
difficult to compute39. Therefore, as illustrated in the following sections, the practical

methods rely on heuristic (i.e., based on empirical past information) and deterministic (i.e.,
predefined) methods, because they are the easiest to handle while operating the system.
If the volume of SS can be computed depending on the operating conditions, the

frequency of review of these requirements would have to be adequate. In practice, depending
on the system sendee and the country considered, the requirements can be reviewed from
every five minutes to once a year (see section 4.6.6). Obviously, an annual requirement does
not provide a constant level of security throughout the year.
Moreover, it is probably more efficient to define asymmetric volumes for system

sendees, i.e. the needs in quantity of up balance and down balance are likely to be different
Lastly, when the timeframe considered gets longer, the problem of volumes of
system sendees moves from the security domain to the adequacy domain. The SO should
give the right signals to the investors in order to obtain enough AS capacities in the long term.
This issue remains open and is deeply linked to the investment in energy production
capacities (see section 4.3.1).
2.4.2 Definitions used within the UCTE

The UCTE does not recommand any particular reactive power AS quantities [UCTE
(2004b)]. Indeed, such reactive power reserves rely on the local structure of the system (e.g.,
strength of the power flows or rating of the lines). Therefore, the definition of these
capabilities is not discussed in this section. On the other hand, frequency control reserves
depend less on the local constraints of the system and have an impact on the whole system.
Therefore, the policy on frequency control res elves across systems is defined by the UCTE.
The primary frequency control reserve within the UCTE is based on the largest
power devia don to be handled, which is 3 000 MW, "assuming realistic characteristics
concerning system reliability and size of loads and generation units” [UCTE (2004b)]. This
reserve is linked to the frequency characteristic of the system and is thus shared by the
59 Section 2.2 shows the difficulty to express the value o f system sendees, while C hapter 3
dem onstrates that it is im possible to calculate precisely the cost to provide ancillary services.
whole interconnected network, as explained in section 1.4.7. The burden o f this reserve is
then shared across the whole interconnected power system according to the generation of
each system.
On the other hand, the secondary frequency control reserve is activated by a given
zone to balance its own production and consumption. Therefore, the reserve of a zone
varies in function of the size and portfolio considered. Within the UCTE, the calculation of
secondary reserves originates from an empirical formula [UCPTE (1991)]. This formula was
developed during the seventies from half-hourly measurements of production and
consumption. The means used at that time where relatively limited, such as hand-calculation
based on information given by phone by operators of generating units. The first formula
has then evolved over time. Table 2.10 gives an overview of the evolution of the secondary
reserve R(y , by considering a few systems within the UCTE and depending on the available
information. The current UCTE formula (i.e., since 2004) is a compromise based on all the
previous formulas. However, TSOs prefer to take margins to increase the security of the
system, especially during fast variations of the load. Lastly, note that the UCTE
recommends that the loss of the largest unit in the zone be compensated by both secondary
and fast (i.e., deployed in less than 15 min) tertiary frequency control reserves [UCTE
(2004b)].
T ab le 2.10: R ecom m en d ation s for secondary reserve in so m e sy stem s w ithin the U C T E
At
least System Formula References
since
1991 Germany 1 1 2 500 * * UCPTE
(1991)
2 Vv Pz
maxgttitralt'on
PL-grnrtu, = estimate of the peak generation for the
day (in MW)
/A —estimated power deviation in % related to
Pmax frnmiiion • Ttie average of I) was around 3.5 %, and die
maximal value 9 % during rapid load fluctuations (6-
9 h)
1991 France UCPTE
Rxr = 2.8 x p mimipt!on
(1991)
R ;r = 0.02 x during rapid load fluctuations
’P<v»snmptm ~ estimate of the internal consumption (in
MW)
1991 Italy Re UCPTE
sec = 0.0125 x Econsumption
(1991)
1998 Spain Rse
^c = 3X A// maxconsumption Spanish
industry
Ri = 6 XVP L mmumption dul'lng mPid lo£ld fluctuations and energy
department
P L — prim, = estimate of the maximal internal (1998)
consumption (in MW)
2004 UCTE UCTE
K = V10x + 22 50 0-150
(2004b)
2006 France R;, = min {5OO , ^ 1 0 x + 22 500 -15o) Texier-
Pauton
demand gradient (2007)
sec s
6
if the demand gradient is higher than 12 000 M W /h
2.4.3 D iscussion of the UCTE recommandation
Clearly, the UCTE formula is a compromise between previous methods, which were
developed during the seventies, when the portfolio and the measurement means were
different than what is available today. This section thus attempts to put into perspective the
current UCTE formula.
2 .4 .3 .1 Link b e t w e e n s e c o n d a r y a n d tertia ry c o n tr o ls
Tertiary and secondary frequency controls have the same objective: bringing back the
exchanges of the area to their targets values (or the frequency to its target in a case of an
area without any interconnection). A trade off is thus possible between secondary and fast
tertiary reserves. Consequently, the definitions of the necessary secondary and fast tertiary
frequency reserves have to be considered together. However, the current requirement of the
UCTE ignores this possible trade-off.
Figure 2.10 represents the compensation of an imbalance with the help o f both
secondary and tertiary frequency controls. Despite having the same objective (i.e.,
compensate the imbalance of the zone), the tertiary frequency control is less flexible than
the secondary frequency control. Indeed, tertiary frequency control offered by generating
units or consumers usually include delays and minimum durations of use. However, some
power systems can benefit more from flexible tertiary frequency control than other systems.
Hence, each power system should have its own requirements for secondary reserves as a
function of the available tertiary frequency control offers (which vary over time).
Actual generation - Actual consumption (in MW)
Imbalance to
compensate
Controls (in MW)
Energy of tertiary control
Tertiary control
Energy of
secondary control
'I-A N.
", N Secondary control
F igure 2.10: D ep lo y m en t o f secondary and tertiary controls
2 . 4 .3 . 2 N o rm a lity
If the necessary secondary and tertiary frequency control powers can be determined to
compensate a given imbalance, one can plot die probability density of the necessary
frequency control powers60 for any imbalance. Figure 2.11 gives a stylised representation of
such a density. As it would be too expensive (and actually impossible) to have always
enough frequency control power, policy makers have to accept that not all positive or
negative imbalances will be covered. This is translated by two limits on the probability
density (a limit for the negative tail and another for the positive tail, as shown in
Figure 2.11).
The choice to cover 99 % of the cases is usual. For example, this criterion
correponds to the one chosen in France for the margin of the morning peak [RTE (2004)].
60 W hen the regulation signal is equal to —1 (see section 2.3.3.1), tire positive secondary frequency
control pow er is equal to twice the secondary? frequency control reserve.
Then, this criterion has to be distributed amongst positive and negative imbalances,
according to the priority given and eventually the related cost (e.g., 0.5 % each tail).
The requirements within the UCTE during the nineties (see Table 2.10) made the
assumption that the necessary secondary frequency control power follows a normal law.
Hence, the necessary secondary frequency control power to be covered during 99 % of the
cases can be easily deduced: it should be ±2.576 x cr, where cr is the standard deviation of
the normal law considered, as shown in Table 2.11 [Saporta (1990)].
Probability
Necessary
negative Necessary positive
Imbalances not control power control power to be Imbalances not
covered to be covered covered covered
Necessary frequency
control power
Lower cover limit Average Upper cover limit
F igure 2.11: Probability d en sity o f the n ecessary frequency control pow er
T ab le 2.11: R epartition o f a pop u lation for a norm al distribution
D eviation from the Probability to have a

m ean lower deviation
±1.645 x cr 90.00 %
±1.960 x <7 95.00 %
±2.576 x cr 99.00 %
±3.000 x cr 99.73 %
±3.291 x <7 99.90 %
The normality hypothesis was made when the available techniques were limited.
However, because of the current computation capabilities, recent reserve requirement can
now be based on more complex density functions. To give an example, Figure 2.12 shows
the empirical cumulative distribution function of the activated French tertiary control
power in 200661, and the corresponding normal cumulative distribution function with an
61 The example is based on the tertiary frequency control because the data on the activated secondary
frequency control pow er was n o t available.
average of -4.5 MW and a standard deviation of 893 MW. The activated tertiary control
power is defined as twice the energy of tertiary frequency control deployed over half an
hour (see Figure 2.10). In other words, it represents the average power needed to provide
the activated energy62. The data are from RTE (2007a). Note that the transmission losses
are included in these data, since RTE is treated as a consumer that uses active power to
cover the network losses.
To test whether the two cumulative distribution functions are equivalent, the
Kolmogorov test was performed. This test validates an empirical series F* (jQ by
comparison to another distribution F(x) by calculating the maximal difference DKolmagurw,

between the two functions, as shown in (2.3). If DK(llmo„olw, is inferior to a given value, the
empirical series F* (x) can then be assimilated to the series F(x) with a given
confidence [Saporta (1990)].
<2-3>
In this case, jDKb, m, = 0.055 is obtained, whereas the threshold for the Kolmogorov
test is 0.0081 for 17 520 elements63 with a confidence of 80 % 64. Therefore, the series is far
from following a normal distribution. This difference is explained by the threshold that can
be seen around zero (see Figure 2.12). In other words, it is highly probable that the
activated tertiary frequency control power is null over a half-hour. In fact, participants have
incentives to reduce their imbalances, as the balancing mechanism charge users that are in
imblance.
It is difficult to extrapolate directly these results for the secondary frequency

control. In addition, the UCTE recommendation is as function of the forecasted
consumption of the zone , whereas our study has considered all the data. However,
because of the link between secondary and tertiary frequency controls, secondary frequency
control power may also not follow a normal distribution. Nevertheless, the same calculation
62 T herefore, up and dow n balancing can com pensate each other over a h alf hour.
G3 T he series has 17 520 elements because there are 365 days x 48 half-hours/day.
1-073
64 0.0081 is obtained w ith the following formula: , [Saporta (1990)].
^ 17 520
should be performed to assess the foundation of the policy for the secondary frequency
control reserves.
"■—Empirical cumulative distribution

ft©-
“ -Normal cumulative distribution
2a
a>
>
JH
3
£3
o
-8000 -6000 -4000 -2000 0 2000 4000 6000

Activated tertiary control power (in MW)
F igure 2.12; E m pirical and norm al cum ulative distribution fu n ction s o f the activated tertiary control
pow er in 2006 in France
2 . 4 .3 . 3 S ta tio n n a r ity
Making assumptions on future imbalances relies on the hypothesis that the series
considered is stationnary, i.e. that its standard deviation and its average do not change over
time (see Appendix A. 1). However, since the portfolio of a system evolves in time,
especially with the increase in distributed generation, it is possible that the necessary
frequency control power not be stationnary. Therefore, the reserve requirements should be
updated frequently.
2 . 4 . 3 . 4 S y m m e tr y o f r e s e r v e s
As shown in Figure 2.11, the necessary frequency control power can be on average different
from zero. In other words, there may be an expected imbalance. This can happen when the
participants have an incentive to under- or over-estimate their production or consumption.
In practice, this happens when the balancing mechanism has assymmetric prices. Therefore,
symmetrical reserves do not guarantee that sufficient reserves are available to hedge against
imbalances. However, the secondary reserve is only dimensioned as a positive power within
the UCTE. Therefore, the current recommendation does not guarantee enough negative
control power capacity.
2 . 4 .3 . 5 C o s t o f o u t a g e a n d c o s t o f p r o v isio n
The probability density function discussed above does not take into account the cost of
being short of system services (e.g., the cost of the outage resulting from by the lack of
reserves). Indeed, as shown in section 2.2, this cost varies over time, so the risk cover factor
(and the frequency control reserves) should also vary with time. For example, RTE’s
objective is to have a probability of using exceptional means (e.g., load disconnections)
inferior to 1 % at the morning peak and 4 % at the evening peak [RTE (2004)]. However,
the UCTE currently does not recommend such a differentiated requirement over time.
In addition, the method to calculate the frequency control reserves is marred by

imprecision. Moreover, the cost to provide the necessary frequency control reserves by
users (i.e., the cost ofprovision) varies with time. An inelastic demand (i.e., constant whatever is
the cost of procurement) can thus be discussed (see section 4.3.2).
2.4.4 Innovative m ethods

The previous section has demonstrated the limits of a method based only on the probability
of having insufficient resources. In particular, it has been shown that the needs of the users
are not well reflected by such a method. Therefore, a few innovative methods are proposed
here to contribute to this debate. As system sendees are public goods (see section 2.2), one
might want to apply the relevant economic concepts in then procurement. A first possibility
to procure a public good is to use private provisions. In this scheme, the users of the power
system have to cooperate and truthfully reveal how much they value the system sendees.
Then, each user would pay for the system sendees according to the value that it gives to
each sendee. Such an approach is theoretically the best, but does not work in practice. In
fact, users may not be willing to reveal the true value in order to try to pay as little as
possible, hoping that others will pay. This phenomenon is called free riding. Therefore,
incentives are generally too poor for providers when a voluntary equilibrium for a public
good is sought [Varian (1999)].
Voting may also be an alternative. In such a system, users would vote for say three
levels of system sendee A, B and C. When a level of system sendee is chosen, the cost is
split equitably amongst all the users. However, this system may lead to some difficulties. For
example, let suppose that A is preferred after the vote. However, it is possible that a
majority of users prefer B to A, a majority prefer C to B and that a majority prefer A to C!
And even if such cycles can be avoided, the result may not be efficient because a vote is
binary while the volume of system services is continuous. Lastly, voting does not necessarily
provide a good incentive for users to honestly reveal their true preferences. For example,
Varian (1999) illustrates these flaws using the case of a vote in the Congress of the United
States in 1956.
In the case of a continuous public good, Varian (1999) shows that a Pareto-efficient
amount of the good is obtained when the sum of the absolute rates ofsubstitution between the
private good and the public good for all consumers equals the marginal cost of providing an
extra unit of the public good. However, it is not clear what private good can be substituted
for security in a power system. For example, such a private good might be a generator close
to the load, a private line, load-shedding or a small network with a DC link. A system can
also be imagined in which a load or a distribution company gives instructions directly to an
AS supplier in order to follow the consumption of the load or the distribution
company [Tyagi and Srivastava (2006)]. Anyway, the marginal rate of substitution is unlikely
to be useable directly since it is hard to compute.
Instead o f the rates of substitution, a related idea could be used: the cost of insecurity65.
Indeed, the cost of insecurity is specific to every user of the power system. It can thus be
proven very easily that the optimal quantity of system services is obtained when the
marginal cost of insecurity equals in absolute the marginal cost of system services (see
Figure 2.13). This optimisation problem seems therefore to be straightforward. However,
even if the cost of system services as a function of the quantity can be calculated,
computing the cost of insecurity is much harder for three reasons. First, determining the
value that users put on electricity is not easy and this value varies with time. Second,
supposing that the value of electricity is known for all the users, determining the related
cost of insecurity for different quantities of system services requires huge amounts of
computation. Lastly, the cost of insecurity is represented here by a smooth curve. In
practice, it is very likely that this cost exhibits discontinuities. For example, if just one MW
65 T he security is a public good, while the damages o f insecurity lead to private losses for a user.
T herefore, the cost o f insecurity can be seen as a private good th at can be substituted for security.
of reserve is missing, the TSO may have to disconnect a whole feeder, which usually
represents several MW. Despite these difficulties, this method is one of the main current
research axes. Amongst others, see for example the recent contributions by Papadogiannis
and Hatziargyriou (2004), Galiana et al (2005), Abiri-Jahromi et al (2007) or Ortega-
Vazquez and Kirschen (2007). However, note that these papers do not consider the impact
of system services on power quality and the utilisation of the power system (see section 2.2).
C ost
Total cost
Cost of
system services
Cost of insecurity
q* Quantity o f system services
Figure 2.13: C ost curves o f system services
Another possible solution relies on a tax, called the Groves-CIarke tax, described by
Varian (1999). In this system, each user gives the value g of benefits it derives from the
public good. In addition, if the public good is provided, everybody will have to pay a certain
amount cn known in advance by all the users. The net value //,- can then be computed. This
value is negative if the user loses money by buying the public good and positive if it benefits
from the public good. Pivotal agents are the users that change the sign of the total net value,
i.e. E;/,. In others words, pivotal agents are essential to the decision to procure the public
good. The pivotal agents then pay the harm that they do to non-pivotal agents, i.e. the total
net value without the participation of the pivotal agent. This tax disappears from the
system: it is not given to a participant in order to avoid gaming. Tills scheme is really an
incentive on participants to reveal the true value of the product. However, it works only
when all the Pareto efficient allocations of the public good lead to the same amount of
public good (a situation called quasi-linear indifference). In addition, the Groves-CIarke tax
is not Pareto efficient since the tax disappears from the system. However, in the case of a
large number of participants, this tax may be rather small, because nobody would be pivotal.
This method helps determine the efficient amount of public good, but does not help with
the cost allocation66. Lasdy, note that the cost allocation scheme has an impact on the value
of the tax, but does not affect the decision of buying or not the public good.
Table 2.12 illustrates the Groves-CIarke tax mechanism for a public good that costs
100 €. It can be seen that the proposed cost allocation c-t is different from the value v{ given
by the participants. Therefore, this mechanism may be considered as unfair by some
participants. On the other hand, participants have incentives to reveal the true value of the
good. In fact, if they underestimate it, the good may not be bought at all and the
participants may thus loose the benefit from the public good. But if the participants
overestimate the value of the public good, they may be pivotal and pay a tax that will
undermine any potential extra benefits.
T able 2.12: E xam p le o f a Groves-CIarke tax
Cost (c,) Value (v,) N e t value (n,) Pivotal? T ax

Bidder 1 70 € 50 € -20 € No 0€
Bidder 2 20 € 30 € +10 € No 0€
Bidder 3 10 € 40 € +30 € Yes 10 €
Total 100 € 120 € +20 € - 10 €
Let us imagine a practical example for a given SS67. First, the TSO determines the
SS cost for a given period of time (e.g., one hour of the day) as a function of the quantity,
and allocates it to, say, the four groups of system users A, B, C and D (see Figure 2.14). For
example, these groups of users could be the balancing entities68 in France.
66 See section 4.8.2 for a discussion o f cost allocation.

67 T he present example is an adaptation o f the initial Groves-CIarke tax, w hich was initially designed
for discrete public goods and n o t for continuous public goods like SS.
68 A balancing entity acts on the balance betw een production and consum ption w ithin a specific
perim eter o f the market.
C o st total cost
o f SS
Q u an tity of S S
F igure 2.14: A llocation o f SS costs b etw een four groups o f users
Then, each group of users reveals the value that it gives to the quantity of SS for the
given period of time (see Figure 2.15 - note that a discontinuous curve could also be
envisioned). To be able to let the value of SS be estimated by users, the TSO has to give
them some system information. For example, for a given quantity of SS, the probability of
level of quality of supply is estimated. All the values of SS are then added in order to obtain
the total value expressed by all the users to the SS. Unfortunately, this step is likely to be
very difficult to implement in practice because it may be very difficult for users to estimate
the value of system sendees. In addition, the TSO may not be able to give sufficient
information on the impact of a given quantity of system sendees.
Value
of SS
Q uantity of SS
Figure 2.15: V alue o f SS for a group o f user
With the cost and the value of the SS, the TSO can find the optimal quantity q* of
SS, which maximizes the difference between value and cost, i.e. the net benefit of SS (see
Figure 2.16).
€ i
V alue of S S
C ost of S S
Quantity
of SS
V alue m inus cost
F igure 2.16: R epresentation o f th e optim al quantity q * o f SS
The last step consists in chatging the tax to the pivotal agents as explained with the
example of Table 2.12. The product of this tax can then be given to the regulator to help it
to finance market audits or other related activities.
The advantage o f the proposed method is that it switches the burden of calculating
the cost of insecurity from the SO to the users of the system. In addition, it gives an
incentive to the users to bid the real value of SS for them. On the other hand, the TSO has
to be able to provide precise information on the impact of a given level of SS, which is
likely to be very difficult to compute. Therefore, building the curve of the quantity of
system services versus their value (see Figure 2.15) is the thoughest point to solve to make
this method practical.
2.4.5 Actual requirem ents acro ss countries

Systems across the world have adopted different methods to calculate the needs for system
services, which lead to different quantities of ancillary services. These quantities affect both
the SOs and the providers. For example, with such data, SOs can figure out whether they
could save money by reducing the volume of ancillary services that they purchase, while
providers can check whether they could sell more ancillary services to their SO (see
Chapter 4 for a discussion on the procurement of ancillary services).
To compare volumes, system specificities and especially the size of the power
system need to be taken into account. Therefore, a specific indicator, called Reserve Indicator
(RI), is proposed in this thesis. It is calculated by dividing die amount of reserve in the
system by the hourly average energy production or consumption (in MWh/h), depending
on the type of reserve. Hence, the SO of a system with a high RI procures more reserves
per unit of energy produced or consumed than one with a lower indicator. One advantage
of this indicator is that it can be calculated easily on the basis of parameters that are easy to
obtain. While the concept o f reserve indicator could also be applied to reactive power
sendees, not enough information is available on reactive power volumes to include a
reactive reserve indicator in this work. Usually, the amount of primary control reserve is
linked to the production. For example, the UCTE contribution coefficients for primary
control are based on the annual production of each system [UCTE (2004b)]. Therefore, the
reserve indicator used for the primary reserve R/C (in %) is obtained by dividing the
average primary frequency control reserve RV (in MW) by the hourly energy production
(m MWh/h), as shown in (2.4).
R
(2-4)
E genemttor)
In most systems, the secondary control reserve is linked to the estimated peak
consumption during a given period. Therefore, the reserve indicator used for the secondary
reserve R iy (in %) is obtained as shown in (2.5).
R IICCl = ^
(2.5)
V }
consumption
where Ry. (in MW) is the average secondary frequency control reserve and E^nji/mp/kii (in
MWh/h) the hourly energy consumption.
Figure 2.17 shows the primary and secondary frequency control reserve indicators
calculated in 2004 or 2005, depending on data availability, for various systems69. For the
sake of clarity, primary and secondary RIs are not separated in this graph. However, in
69 Australia and N ew Zealand have been included in order to have a different view from n o n
interconnected systems. F o r data on electricity production and consum ption, see IE A (2005). F o r data on
reserves, see the websites o f TSO s and m arket operators o f the systems.
order to explain differences, primary and secondary frequency control RI will be discussed
separately.
Except in Australia, the primary reserve indicator is higher in small systems with no
interconnections or limited interconnection capacity such as Great Britain70, New Zealand
and Sweden. Indeed, the primary frequency control reserve is si2ed to counteract the loss of
the biggest unit, which is about the same in all the synchronous zones. Within the UCTE,
Spain has a higher primary frequency control RI because of its compulsory provision (see
section 4.4.2). The values of this indicator for Germany and France are similar. Finally,
systems outside the UCTE have a non-symmetrical primary frequency control reserve
(positive different from negative) and PJM’s R^- was not calculated because, contrary to the
secondary frequency control, there is no specific requirement for the volume of primary
frequency control reserve yet [NERC (2005)].
Germany has a high secondary frequency control RI because of the pluralistic

structure that has been adopted (see section 2.3.4.3). Indeed, each of the four German
TSOs must be able to compensate for any realistic imbalance within its control area.
Therefore, the total of reserves represents a large amount for Germany as a whole. New
Zealand, PJM and France seem to have similar policies in terms of secondary control
volume. Spain has a higher because the Spanish TSO uses an extended N -l criterion
(loss of the largest group and the largest line). Note that its reserves are not symmetrical,
contrary to the two other countries of the UCTE (France and Germany) included in this
comparison. Australia has relatively small amounts of secondary control reserves, because
they are mainly used to control the synchronous time. Lastly, New Zealand is split into two
islands connected by a DC link, so frequency control RIs have been calculated for each
island separately.
70 N ote that for G reat Britain, primary and secondary frequency responses have been added to obtain
the total primary frequency control reserve.
2.5 LOCATION OF ANCILLARY SERVICES 135
Average positive reserve requirement

Minimum positive reserve requirement
Maximum positive reserve requirement
Average negative reserve requirement
Minimum negative reserve requirement
I I Maximum negative reserve requirement 3 ZS
o O O
cn 2H
q; LU CLUD DQ LU N N N
a 0 CD < Z Z Z CL
0) _ 5 .
3
XT : Primary negative frequency control reserve unknown
© T Primary frequency control reserve not formally specified
.MO-J
F igure 2.17: F requency control reserve indicators in 2004-5 across system s surveyed
2 .5 L ocation o f A ncillary S e r v ic e s
In addition to the quality and the quantity of ancillary sendees, the location of the ancillary
sendees has to be considered carefully.
2.5.1 Impact of the location of ancillary services

Frequency control ancillary sendees act on the global frequency (see section 1.4), so one can
think that their location is not important. However, their physical locations should be
considered while procuring ancillary sendees for various reasons. First, congestions of
transmission lines can affect the reliable provision of frequency control AS71. Therefore, a part
of the transmission capacity has to be allocated to frequency control AS, for example as
proposed by the UCTE (2005). If enough transmission capacity is not available, the affected
zone has to secure enough ancillary services from within its perimeter. More generally, as
reserving transmission capacity may be too expensive, the contributions to frequency
71 T h e location issue is in this case relatively similar between active pow er AS and electrical energy-
only products.
control are likely to be distributed across the whole interconnected network to reduce
unplanned power transits following a large generation outage.
Second, in case of a grid separation, the separated parts of the network cannot stay
stable without any frequency control system service. Therefore, a uniform distribution of
frequency control services is desirable. In particular, distributed generation could be very
useful to provide such a uniform distribution of reserve.
Third, balancing production and consumption is much more important at the

transmission level than it is at the distribution level. Indeed, a distribution network with an
imbalance can be disconnected easily from the rest of the system and therefore is not very
threatening for security of the whole. However, an imbalance at the transmission level
cannot be removed easily because of its meshed structure (see section 1.2). Therefore, an
imbalance at the distribution level has generally local consequences while an imbalance at
the transmission level has much greater global consequences. The French standards
illustrate well the importance of transmission systems. In fact, the typical commitment of
the French distributor against an unexpected loss of supply is six hours, whereas it is
around three minutes at the transmission level72.
On the other hand, reactive power ancillary services are local products (see
section 1.5), so the reactive sources have to be as electrically close as possible to the reactive
loads. Therefore, the location of providers of reactive control AS is very important. In
addition, like frequency control, voltage control AS impacts the transmission capacity.
Indeed, (a) reactive power AS need a minimal available transmission capacity to be helpful;
(b) flows of reactive power increase the active power losses; and (c) reactive power AS can
increase or decrease active power flows.
In conclusion, an appropriate location of ancillary services is essential to meet the

needs of users described in section 2.2. For example, Goel et al (2004) or Galiana et al
(2005) propose methods to take into account the location of frequency control ancillary
services while procuring them.
72 T hese values are only indicative. F o r com plete inform ation, refer to E D F (2004) for the
distribution netw ork and to R T E (2006) for the transm ission network. N o te that the actual average loss o f
supply is m uch lower than the com m itm ent. Indeed, the average loss o f supply was 60 m inutes for the
distribution netw ork in 2004 [E D F (2006)] and 1 m inute 42 for the transm ission netw ork in 2006 [RTE
(2007c)].
2.5.2 Location in actual system s
As the location of ancillary services is important, systems have to develop policies to benefit
from sufficient ancillary sendees at the appropriate places, as discussed in section 4.3.1.
Moreover, systems have to define technical procedures to allow some ancillary sendees to
be exchanged between participants. In fact, TSOs usually do not choose where the AS
providers are.
In particular, an increased integration of the electricity markets across systems is on

the way [e.g., see Meeus and Belmans (2008) for continental Europe]. Indeed, trading
ancillary sendees across systems: (a) allows more efficient use of flexible resources; (b)
reduces the potential exercise of market power; (c) diminishes imbalance exposure; and (d)
makes better use of interconnection capabilities [Frontier Economics and Consentec
(2005)]. On the other hand, a limited quantity can be traded for the reasons presented in
section 2.5.1 (e.g., 33 % of the secondary frequency control within the UCTE [UCTE
(2005)]). In addition, trading of ancillary sendees requires standardised definitions (see
section 2.3) and may make the monitoring more difficult (see section 4.8.4).
2.5.2.1 Prim ary frequency control
Within the UCTE, the requirement for primary frequency control reserve is split amongst
the various TSOs according to their electrical energy produced annually (see section 2.4.2).
UCTE (2005) proposes to allow exchanges of primary frequency control reserves between
TSOs. In this framework, the Reserve Receiving TSO (RRT) benefits from the contribution
of one or several units connected in the area of the Reserve Connecting TSO (RCT). Two
methods have been proposed:
V A n A S provider (in the area of the RCT) supplies primary frequency controlfor the RRT. In this
case, the RCT has to give its agreement;
V The RCT provides primary frequency control for the RRT. Therefore, the providers in the area
of the RCT see an increase of their provision of primary frequency control.
Such a framework is relatively easy to put in place since no specific communication

system has to be developed.
2 . 5 . 2 . 2 S e c o n d a r y f r e q u e n c y c o n tr o l
In North America, regulation (i.e., secondary frequency control) can be traded between
systems with the help of the following services [NERC (2006)]:
'C Overlap regulation service-, a control area incorporates some or all of another control area's
tie lines, frequency response and schedules into its own ACE equation (see
section 2.3.4.3);
V Supplemental regulation service, a control area takes into account all or a portion of the ACE
of other control areas in order to provide more regulating power for these areas.
The UCTE (2005) has also proposed changes to allow generating units to provide
secondary frequency control to another zone. Two architectures have been proposed,
which do not directly use the ACE:
V Direct control by the Reserve Receiving TSO (see Figure 2.18): the generating unit in the
reserve connecting area receives the regulation signal directly from the RRT. The
participation of the unit is added to the balance of the RRT and substracted from the
balance of the RCT;
V Control through the Reserve Connecting TSO (see Figure 2.19): the RCT receives the
regulation signal from the RRT. The RCT changes its ACE accordingly. Therefore, all
the generating units providing regulation in the area of the RCT participate in the
regulation effort. This service is thus somewhat similar to the North-American
supplemental regulation service.
Lastly, note that such arrangements will have consequences on the current operating
pratices. In fact, contrary to the current organisation, the exchanges at the interconnections
will not come back to their scheduled values after deployment of the secondary frequency
control power.
Reserve Receiving TSO Reserve Connecting TSO
LFC: LFC:
G= AF+K Af G= AP+KAf
-*• control signal
m e a s u r e m e n t v a lu e
Figure 2.18: D irect control by the reserve receiving T SO [U C T E (2005)]
Reserve Receiving TSO Reserve Connecting TSO
LFC: LFC:
G=AP+K A f G= AP+K A f
© t.
*■ control signal
F igure 2.19: Control through the reserve co n n ectin g T SO [U C T E (2005)]
2.5.2.3 V oltage control
Because of the local, characteristics of voltage control, this ancillary service cannot be easily
traded over long distances (see section 1.5.4). Therefore, sufficient reactive power providers
must be located across the system (see section 4.3.1). In addition, the voltage control
management requires a strong cooperation at the connecting bus between two different
entities; for example, between two TSOs of two neighbouring areas, a TSO and a DSO, or
a TSO and a generating unit. Such cooperation has to be defined with the help of a specific
contract or regulation rules.
2 .6 C on clu sion
This chapter has explored the issues related to the delivery of ancillary services, namely: the
expression of the needs, the quality, the quantity and the location of ancillary services.
Current policies can be improved by taking into account more accurately the value given by
the users to system services and by refining some empirically-defined requirements. In
particular, the following assessment list is proposed for the delivery of ancillary services:
V Needs for system services:

■ Expression of the needs', a cost-benefit optimisation is the best manner to define the
needs in terms of system services. However, the value of system services is very
difficult to estimate because system services affect at the same time the reliability,
the power quality and the utilisation of the power system. Therefore, practical
solutions tend to be based on very simplified methods, past experience or general
recommendations. As a result, the users of a power system, who are represented by
the regulator, may feel that the system services that are provided do not correspond
to their needs, either because they are too expensive or insufficient;
■ Reduction of the needs, on the other hand, system operators have little incentive to
change rules-of-thumb that have performed reasonably well for years, even if these
rules are sometimes hard to justify. System operators may also consider that their
approach within their own territory is the best, forgetting that system services affect
the interconnected system as a whole;
■ Review of the needs, the needs of users and the system constraints evolve with time
(e.g., with the introduction of new types of generations). Therefore, the needs for
system services should be reviewed frequendy by SOs.
S Quality of ancillary services: The description of AS quality is very important because AS
are technical products.
■ Differentiation of ancillary services, it is not possible to find a perfect specification. A
trade-off has to be found between too much or too low differentiation. Such
differentiation may vary as a function of the time horizon considered. In particular,
primary frequency control is more differentiated than secondary frequency control;
* Description of ancillary services, a framework to compare ancillary services has been

proposed and successfully applied to various power systems. In addition, a
standardised description has been proposed based on five parameters: (a) a delay;
2.6 CONCLUSION 141
(b) a deadband; (c) dynamics of the process; (d) a limiter; and (e) a duration of
sendee. Contrary to current specifications, this standardised description does not
rely on the input signal, does take into account the dynamics and can be applied to
any type of provider. Lastly, the term “spinning reserve” is defined in a standardised
manner: the spinning reserve is the unused capacity which can be activated on
decision of the system operator and which is provided by devices that are
synchronized to the network and able to affect the active power;
■ Rationalisation of ancillary services', in practice, specifications of ancillary sendees differ
from one system to another. Therefore, efforts should be made to unify similar
ancillary services in order to foster short-term exchanges of sendees.
S Quantity of ancillary services:
■ Responsibility for requirements', because system sendees are public goods, only the
system operator can define the requirements for ancillary sendees. In addition, the
public good characteristic makes very difficult the design of new methods that
define the requirements for ancillary sendees;
■ Definition of requirementsfor ancillary services', the requirements should take into account
(a) the possible trade-off between products; (b) the potential non-normality, (c)
non-stationarity and (d) non-symetry of imbalances; (e) the time-dependent cost and
value of ancillary sendees. However, current practices are usually opaque. Therefore,
the system operator has to explain to all the participants how the requirements are
calculated;
* Elasticity of requirements-, the fixed demand for ancillary sendees currendy used is
inelastic and marred with large uncertainty. Therefore, the benefit of procuring
more or less 10 MW of reserve is not assessed;
■ Synopsis of requirements', an indicator has been developed to compare requirements for
ancillary sendees and thus gives incentives to participants to improve their practices.
S Location of ancillary sendees:
■ Importance of location', die location of reactive power providers is particularly
important. For frequency control, die actions of providers are global but still may be
hampered by a bad location (e.g., a congested area);
* Incentives in the long rum since AS providers cannot be moved easily, appropriate
incentives should be given to have access to sufficient AS capabilities at the
appropriate locations in the long run;
■ Exchanges in the short temr. technical mechanisms should exist to allow participants to
exchange AS in the short fun to help maximize the global welfare. In addition,
standards should be developed to allow neighbouring participants to communicate
(e.g., a TSO and a DSO). ■
CHAPTER 3
COST OF ANCILLARY SERVICES
A. smile costs less money than electricity hut gives as much light.
Henri Groues (1912 - 2007)
3.1 Introduction
T IS essential for a provider of ancillaiy sendees to understand the cost of its supply in
I order to take the appropriate decisions for its investments and its contracts.
Furthermore, policy makers also need to appreciate the structure of the costs of ancillaiy
sendees in order to design the most appropriate market. This chapter therefore reviews the
main components of the cost of ancillary sendees (section 3.2). A practical methodology
based on operational data is then proposed to estimate the cost incurred by a producer
when it must reserve capacity on a day-ahead basis to provide frequency control
(section 3.3). This methodology was successfully applied to EDF Producer’s generation
portfolio and gives interesting insights on the costs of frequency control ancillary sendees
(sections 3.4 and 3.5). In addition, the cost of time control is assessed for France
(section 3.6).
A. Comprehensive Assessment ofMarketsfor Frequeue)' and Voltage ControlAncillary Services, Yann Rebours, 2008 143
144 CHAPTER 3 COST OF ANCILLARY SERVICES
3 .2 Main C o st C om p on en ts o f A ncillary S e r v ic e s
The reflection on the cost of ancillary services started in the mid-nineties, in particular with
the work by Hirst and Kirby (1996)73. At EDF, internal studies started at the same time in
order to prepare the separation between activities of transmission (i.e., RTE) and generation
(i.e., EDF Producer) o f electricity (see section 1.2). After more than ten years of various
studies, the cost of ancillary services is now much better understood than it was before the
separation o f transmission and generation activities. This section reviews the main cost
components that frequency and voltage control ancillary services cause for an electricity
producer. The additional costs for the system operators are not considered.
3.2.1 Fixed co sts

Fixed costs are all the costs that do not depend on the quantity of services provided. The
fixed costs due to frequency and voltage control ancillary services have two main
components:
Y' An increased investment cost, such as:

* Additional protections or regulators in the power plant;
■ The communication system between the system operator and the producer (e.g., to
send the signal used for the secondary frequency control);
■ The measurement system that helps track the performances of the unit;
■ The over-investment in power generation capacity in comparison with what would
be required to meet only the demand for energy. For example, frequency control
reserves represent up to 20 % of the capacity needed for energy (see section 2.4.5).
This over-investment cost is calculated by comparing the cost to develop a fictitious
adapted portfolio to the actual portfolio;
“ The over-sizing some equipment such as step-up transformers or alternators (see
Figure 3.1). Indeed, higher currents circulate through windings when voltage control
is provided. To give an idea, three 360-MVA transformers are used in France for
each 900-MW nuclear unit (instead of three 300-MVA transformers), and three 570-
MVA transformers for each 1450-MW nuclear unit.
73 In addition, various publications at that time were discussing the theoretical fram ew ork to
unbundle prices for energy and ancillary services, such as Baughman et aL (1997a), Baughm an et aL (1997b),
H ao and Papalexopoulos (1997), Z obian and Ilic (1997a), Z obian and Ihc (1997b).
3.2 MAIN COST COMPONENTS OF ANCILLARY SERVICES 145
•S The manpower required to provide ancillaiy sendees, for example for:

■ Controlling the performance of the ancillaiy sendees provided (tests and analysis of
performance) and to operate the plant in a more complicated manner than it would
have been without providing any ancillary sendees;
■ Implementing the contracts between the buyers and the suppliers of ancillaiy
sendees. To give an order of magnitude, the transaction costs between the TSO, the
DSO and the users represent 20 % of the transmission cost in France, i.e. 214 M€ in
2006 [French economic and industry department (2005)].
Producer TSO
Step-up Point of Delivery

Alternator transformer
over-sized
Figure 3.1: R epresentation o f tw o over-sized elem en ts b ecau se o f voltage control
To show tire complexity of determining the fixed costs due to ancillary services, let
us briefly discuss the additional cost of a step-up transformer used for voltage control. Step-
up transformers for large power plants are custom-built. The cost of this kind of
transformer relies on various parameters related to their design such as the labour needed to
design and build It, the materials used, its nominal power, its reactance74, its number of
poles73 and its capability to withstand lightning surges (the protection level).
In addition, the transportation cost can vary a lot from one site to another because
it is quite difficult to transport such a transformer. Typically, a site close to a large river or a
railroad has cheaper transportation costs than a power plant on an isolated site. Moreover,
if the oil tank or the magnetic circuit are reused, the cost may decrease drastically. Lastly,
74 F o r nuclear pow er plants, the reactance o f step-up transform ers ranges from 12 to 14 %. To
diminish this reactance, the winding length should be shortened. T o do so, a smaller winding radius can be
used, b u t the mechanical constraints are increased. Alternatively, a larger section may be used, bu t this m ethod
increases the space used, the w eight and the cost. A n optim um has thus to be found to satisfy the criteria o f
the generating company.
75 T hree single-phased transform ers are 15 % m ore expensive than a stand-alone three-phased
transform er.
the policies of the companies have also an important impact. Indeed, the generator may
favour different manufacturers to spread risks, while manufacturers may reduce prices to
increase their market shares.
Therefore, there is no catalogue price and it is very difficult to determine the

standard investment cost of one MVA. Nevertheless, the order of magnitude of the cost is
between 1 k€ and 6 k€/MVA.
Once the investment cost has been calculated, the actualised cost should be
determined in order to spread the investment cost over the lifetime of the transformer.
However, diis lifetime varies between 30 and 40 years for nuclear power plants, and some
step-up transformers for hydro units are still in use after 80 years of service16. In addition,
the investment cost is done in terms of apparent power (i.e., in MVA) and not in terms of
reactive power (i.e., in Mvar). Appropriate amortisation and separation between active and
reactive powers thus need to be performed. To do so, various techniques are possible. For
more details, see for example Eurelectric (1997), Da Silva et al. (2001), Alvarado et al (2003)
or Elizondo et al. (2006), who all try to give the estimated total investment cost due to
voltage control in various countries.
In conclusion, fixed costs are quite difficult to compute because costs are complex,
assets used to provide ancillary services have usually other purposes than the sole provision
of ancillary services and the lifetime to consider is not very clear. Therefore, it may be
useless to try to determine a very precise fixed cost, because calculations are likely to be
plagued by large uncertainties.
3.2.2 Variable costs

Variable costs are all the costs that depend on the quantity of services provided. The
variable costs due to frequency and voltage control ancillary services have two main
components:
^ The capacity reservation produces additional costs, called reservation costs-.

■ The de-optimisation cost: because of the provision of ancillary services, it costs
more to the generation company to produce the same quantity of energy. For
76 By com parison, transmission transform ers have a lifetime o f 40 to 60 years.

3.2 MAIN COST COMPONENTS OF ANCILLARY SERVICES 147
example, by providing ancillaiy sendees, a new generating unit may have to be

started77, the dispatch is not optimum (i.e., some cheap units provide ancillaiy
sendees instead of energy) or the efficiency is lower78. The de-optimisation cost is
calculated by subtracting the dispatch cost of Figure 3.2.b from the dispatch cost of
Figure 3.2.a. Note that this de-optimisation cost cannot be estimated easily by hand,
because the dispatch of Figure 3.2.a is not straightforward. Indeed, the unit
commitment is likely to be different. Lastly, note that the de-optimisation cost
becomes linked to the increased investment cost when it is considered over a long
period (see section 3.2.1);
■ The opportunity cost: the net profit of the company would have been higher if it
did not provide ancillaiy services (e.g., by selling additional energy on the spot
market instead of withholding capacity to provide frequency control). The
opportunity cost is calculated by subtracting the profit of Figure 3.2.a from the
profit of Figure 3,2.b. The opportunity cost may sometimes not exist, i.e. the profit
may not increase when ancillary sendees are not provided (profit of Figure 3.2.a <
profit of Figure 3.2.b). However, in practice, the opportunity cost is actually
calculated by subtracting the profit of Figure 3.2.c from the profit of Figure 3.2.b,
where a larger amount of power is sold at the spot price in Figure 3.2.c than the
optimal dispatch without ancillaiy sendees (quantity 2 > quantity 1). Even if the first
calculation is more rigorous, the second one is preferred in practice (e.g., at PJM)
because of its simplicity [PJM (2008)]. For voltage control, despite a large literature
describing this phenomenon [Hirst and Kirby (1996), Da Silva et al. (2001), Gross et
al (2002)], the opportunity cost is actually quite theoretical because alternators have
generally been designed to provide reactive power without reducing their active
power generation. Indeed, the connection rules impose such capabilities (see
section 2.3.4.4).
A The utilisation of the reserved capacity, called utilisation cost.
77 F o r example, an expensive unit may be called to provide only voltage control to support a given
area.
78 T he efficiency o f therm al units is related to the main steam pressure and the throttling o f the
turbine control valves. Bakken and Faanes (1997) show that the efficiency can decrease by 2 % w hen the
control valve is completely open, w hich can happen w hen a unit provides frequency control.
■ Deployment cost: it is the direct extra-cost to actually provide the requested service.
For example, more fuel may be used because the unit generates extra power or the
active power losses may increase [Cigre Task Force (2001), Gross et al (2002)];
■ Additional maintenance cost: wear-and-tear is increased, so elements of the unit are
aging faster than without providing any ancillary service. Note that this cost is
particularly high when a generating unit is used as a synchronous
compensator [Saluden (2006)].
M arginal c o s t M arginal c o s t M arginal c o s t

(6/MW) (€/MW) € /MW)
Spot pn ce 2 Spot p n ce 2
Spot price 1 . ////
///s /s /s /s s s s s s s s s s /s ;
'S S S S /J S S S S
//////////A
V S /////S /S ,
P o w er P o w er P o w er
Quantity 1 Quantity 1 Quantity 1 Quantity 2

(a) D isp a tc h w ith o u t r e s e r v e s (b) D isp a tc h w ith r e s e r v e s (c) D isp a tc h to e stim a te th e o p p o rtu n ity c o s t
: profit ic?Vi : pow er th at w ould h av e b e e n sold without providing rese rv e s
Figure 3.2: Schem atic d isp atch es to understand d e-op tim isation and opportunity costs
To give an idea of the impacts of ancillary services on wear-and-tear costs:

Voltage control implies:
■ Temperature fluctuations that damage mechanical parts;
■ Varying electromagnetic forces and thus vibrations that increase stress on
alternators;
* A faster aging of the step-up transformer because of increased currents79 and
because of the potential saturation of the magnetic core (a saturated magnetic core
heats up quickly).
S Frequency control implies:
■ More stress on steam circuits. The list of components that are affected depends on
the technology. Broadly, the mechanical parts that ensure a connection between two
different parts (e.g., the boiler and its feeding pipe) are subject to higher constraints;
■ For a nuclear power plant, a reduction in the lifetime of the system that commands
the control rods in the nuclear core.
79 Windings o f step-up transform ers are w rapped in paper to avoid electrical contact betw een turns.
W hen die current increases, th e tem perature in the transform er increases as well. T his tem perature increase
extends die cellulose chains that are contained in the paper, w hich may possibly lead to a rupture. This rupture
may cause a short circuit betw een two adjacent turns. In addition, the cellulose rupture releases w ater, which
may react w ith the h o t oil.
3.3 M ETHODOLOGY AND HYPOTHESES TO ESTIMATE COST 149
In conclusion, variable costs are somewhat easier to identify and quantify than fixed
costs. The cost of capacity reservation is likely to be significant for frequency control, while
it is close to zero for voltage control, except for synchronous compensator operation. As
far as frequency control is concerned, the de-optimisation cost (i.e., the difference between
the costs of dispatch with reserves and without reserves) represents the main capacity
reservation cost. The direct utilisation cost due to voltage control is lower than the direct
utilisation cost due to frequency control. However, voltage control produces higher stress in
alternators and thus extra maintenance cost.
3 .3 M ethodology and H y p o th e ses to E stim a te C ost
The previous section has presented the various components of the cost incurred by a
generating company because of ancillaiy sendees. Determining this cost is difficult because
(a) various time horizons have to be considered; (b) the cost has different origins; (c) the
allocation of the cost amongst the different products is hard, so double-accounting is
possible; (d) such a determination relies on simulations, because the system cannot be run
again without AS provision, so assumptions have to be made, which lead to uncertainties
and thus potential errors. Nevertheless, as discussed in this chapter, it is possible to estimate
the de-optimisation cost, which is one o f the most important components of the cost of
AS. This section introduces the optimisation process at EDF Producer (sections 3.3.1 and
3.3.2), describes the methodology adopted to estimate the de-optimisation cost for EDF
Producer (sections 3.3.3 and 3.3.6) and presents the main tools that have been used and the
related hypotheses (sections 3.3.4 and 3.3.5).
3.3.1 Time horizon for a generating company
A generating company has to take decisions for various time horizons, such as [Ernu
(2007)]:
Y More thanfire years ahead: investments in generating capacities;
Y Severalyears ahead: planning of shutdowns of nuclear power plants, as a nuclear power

plant cycle lasts around 18 months;
Y Around oneyear ahead: value of water, fuel provision, risk analysis;

S Up to one week ahead: adequacy analysis, timing o f maintenance for thermal units,
definition of the generation cost, option contracting;
S Up to one day ahead, dispatch of the production, trade-off with available options (e.g.,
option contracts or spot market, as presented in section 1.3.1);
'd Several hours ahead bidding/offering in the balancing mechanism and re-dispatch.
When the time horizon increases, the number of constraints diminishes but the
uncertainty increases. Therefore, probabilistic approaches are adopted to deal with long
term issues: a large amount of computing power is used to run a large number of cases with
a relatively low number of constraints. O n the other hand, when the time horizon
decreases, the uncertainty decreases but there are many more constraints. Deterministic
methods are thus preferred: the computing power is used to take into account the
maximum number o f constraints, and the uncertainty is dealt with by taking margins (see
Figure 3.4 in section 3.3.3).
Figure 3.3 shows the tools used by ED F Producer to optimise and take decisions
over the different time horizons. ARRET, GLOBAL, LOCAL, OPHELIE and ARGOS
are the names of the various software chains. ORION, MNR, COMP AS and APO G EE
refer to the models used to optimise the decisions taken. Note that all the software chains
are operated by humans. Indeed, the solutions proposed by the models need to be
supervised and experts may decide to take different decisions by considering constraints not
handled by the models.
3.3 METHODOLOGY AND HYPOTHESES TO ESTIMATE COST 151
Several years One year

Optimisation of Optimisation of
nuclear power water resources
lant shutdowns
Shutdown
planning GLOBAL
MNR One week
hutdown planning COMPAS
cost
Hydro
argin il costs of optimisation for One day
Planning of the iration each valley
nuclear power
plant value of water
shutdowns
Margins and risk

analysis
Fuel needed Value of water
Feasible planning
Indicative generation
planning
Marginal costs
Market options to buy
Figure 3.3: O verview o f the optim isation process at E D F Producer [based on E m u (2007)]
3.3.2 The daily optimisation process at EDF Producer

A software chain named ARGOS (see section 3.3.1) is the main tool that helps manage the
daily generation optimisation process. This tool is divided into several modules, amongst
which we need to describe a few:
^ APOGEE, which optimises the generation of the 110 thermal units and the 50 hydro
valleys8". APOGEE two days optimises the portfolio with a resolution of half-hourly
time steps over two days, i.e. 96 time steps81. As explained in section 3.3.1, decisions to
start up or shut down nuclear power plants must be made well in advance. Therefore,
once a nuclear unit is available, APOGEE usually commits it because the variable cost
of a nuclear power plant is very low but the shutdown costs are high. However,
APOGEE takes into account the maximum deep slow downs allowed per day and may
sometimes ask to shut down a nuclear power plant82;
80 A hydro valley is a gathering o f interdependent hydro units.
81 A P O G E E can also be used to optimise the portfolio over 12 days (it is thus designated as
A P O G E E 12 days). T he algorithm is the same as A P O G E E 2 days, but the datasets are different. F o r the
sake o f simplicity, the term “A P O G E E ” is used in this thesis to designate A P O G E E 2 days.
82 W hen A P O G E E decides to shut dow n a nuclear pow er plant, it takes into account the tens o f
MW that the pow er plant consum es for its auxiliary systems. However, A P O G E E ignores the transmission
fees that will be incurred if all the nuclear units o f a given site do not produce any power, which is very rare.
^ CANYON, which optimises the 500 hydro units by taking into account much more
hydro constraints than A PO G EE does. This optimisation tries to meet the set points
provided by A POGEE for the valleys;
S COCKPIT, the Human-Machine Interface (HMT) that helps co-ordinate the various
modules of ARGOS. In particular, COCKPIT allows operators to change the dispatch
provided by the models.
Table 3.1 presents the management of frequency control reserves throughout the
short-term process adopted at ED F Producer.
This fee is quite high in France and across E urope because m o st o f the transm ission tariff is paid by
consum ers [French econom ic and industry departm ent (2005)].
T ab le 3.1: D aily m a n a g em en t o f frequency control reserves provided by E D F Producer
Day T im e E D F Producer RTE

D- 1 11:00 RTE’s demand is entered into RTE sends the demand for
ARGOS primary and secondary
frequency control reserves to all
the producers
D -l 13:00 APO G EE is run taldng into
account all the available
constraints given by the
operators from all the sites.
Run-of-river hydro and special
agreements are not considered
D -1 13:30 From the dispatch given by
APOGEE, CANYON is
launched to obtain a feasible
hydro dispatch. The operators
re-dispatch the generation and
take into account the last
constraints
D -l 16:00 EDF Producer sends its
dispatch to RTE (programme
d’appel)
D -l RTE runs simulations for the
French power system with the
dispatches from all the
producers. RTE validates these
dispatches or makes
modifications through the
balancing mechanism to obtain
the programme de marche
D up to H -283 EDF Producer can modify its RTE uses the balancing
dispatch under some mechanism to change the
conditions schedules. RTE sends the
dispatch instructions to the
units (this task will be
performed by EDF Producer in
the future)
D H EDF Producer provides the RTE controls in real-time the
frequency control sendees in provision of the sendees (see
real-time section 4.8.4.1)
85 Actually, producers cannot re-declare their production dispatch at any time. They can do it only at
specific times every hour.
3.3.3 Principle of the de-optimisation co st calculation

The de-optimisation cost is one main component of the variable cost of frequency control
ancillary services (see section 3.2.2). Since RTE requests frequency control reserves only
one day ahead (see section 3.3.2), ED F Producer may wonder what the cost of this demand
is going to be, compared to the cost of deciding on D—1 not to provide these reserves. In
other words, this study proposes to estimate the de-optimisation cost due to the day-ahead
request for reserves from RTE.
To calculate this cost, APO G EE is used to run two simulations: one with the
reserve provision and a second one without any demand for frequency control reserves (see
Figure 3.4). Both of these dispatches are fictitious because the dispatch proposed by
APOGEE does not correspond to the ultimate dispatch. Indeed, CANYON and the
operators change the dispatch of APOGEE (see section 3.3.2). The cost difference between
the two dispatches gives the de-optimisation cost, as expressed in (3.1), where C T
f hmem> and
Without mmts are îe dispatch costs for the day D with and without providing reserves,
respectively; and C® t!mlsalion the de-optimisation cost for the day D.
C di-optimisation
D = C with
D. m in t —C without
D resent (3' r»/
\
This de-optimisation cost is a short-term value because all the previous

optimisations (from tens of years ahead to several days ahead) have considered the
provision of frequency control ancillary services, so the day-ahead constraints would have
been different without providing reserves in the long run. For example, it is probable that
the start-up of nuclear power plants could be better optimised if no ancillary services are
required. Moreover, in this study, the results obtained for one day do not modify the
constraints of the following days. For example, more water is likely to be available after one
day with the demand for reserves equal to zero than after a day with the initial demand for
reserves. However, the water levels of the following days are not modified here as the initial
operational constraints are kept for each day. Therefore, the short-term de-optimisation cost
calculated in this study is likely to give an underestimate ofthe actual de-optimisation cost.
So far, such a study has never been conducted in the literature. Indeed, previous
publications were either not focused on the de-optimisation cost or were considering a
fictitious portfolio for a limited number of days and with a non-dedicated algorithm [e.g.,
H im and Kirby (1997), Proctor (2001), Wu el al. (2004), Galiana et ah (2005), Ortega-
Vazquez and Kirschen (2007)]. O n the other hand, the proposed method is practical
because the computation burden is manageable and because it acknowledges that doing
again the whole optimisation process without considering the reserve provision is practically
impossible, since too many parameters are likely to change. This method is also realistic,
because the constraints and the portfolio are actual data. Lastly, the optimisation algorithm
used is complete and adapted to the problem, as it has been specifically developed by EDF
over tens of years.
E D F P ro d u c e r s . x x re-fuelling of water unit commitment

investment , . . « j - .. u re-dispatch
d e c is io n s power plants management 8k dispatch r
10~year ahead 18-month ahead day ahead

T im e lin e
time
Number of scenarios
simulated
S im u la tio n s
Number of constraints
managed
O u r s tu d y 1. Dataset with original 2. Dataset with

constraints no demand for frequency control
reserves
Figure 3.4: T h e tw o datasets considered to calculate the d e-op tim isation co st
3.3.4 Data considered
This section briefly presents the data used in this study.
3 .3 .4 .1 In p u ts o f t h e m o d e l
All the input data considered in this study are those available to A PO G EE at around 13h00
on D—1 (see Table 3.1). The first run of our study has the operational data as input, without
any change. On die other hand, the second run uses as input the operational data with the
reserve demand set to zero. However, even if the demand is zero, units can still provide
reserves for this second run, because:
S Units that have been pre-scheduled with reserve prior to the APOGEE run are still
providing reserves;
S A PO G EE may estimate that it is cheaper to dispatch reserve than not to provide any.
3 .3 .4 .2 A v a ila b le d a t a s e t s
A POGEE uses text files as inputs and outputs. To give an idea of the quantity of data to
manipulate, a dataset for one day represents around 60 text files and 6 MB. Table 3.2 shows
the periods studied and the available data. A non-valid day is a dataset for which APO G EE
was not able to run, usually because of files lost during the process o f archiving the
operating data. In addition, the data from 01/01/2006 to 24/06/2006 were unfortunately
not available. The present study is therefore based on 879 valid days, which is a sufficient
large number to give significant results. In addition, note that each day represents 48 half-
hourly time steps, so 42 192 valid half-hours are included in this study.
T ab le 3.2: Periods consid ered in the present study
N um ber N um ber o f
N um ber Percentage o f
From To o f valid non-valid
o f days non-valid days
days days
08/09/2004 31/12/2005 480 460 20 4.2 %
25/06/2006 30/08/2007 432 419 13 3.0 %
T otal 912 879 33 3.1 %
3 . 3 . 4 . 3 H y p o t h e s e s fo r m i s s i n g d a t a
As shown in Table 3.2, some datasets were not available. These missing data are simply
ignored when frequency distributions84 are calculated. However, it is important to complete
this missing data to capture any seasonal effects (see section 3.4.2).
These missing data cannot be arbitrary set to zero because this will unbalance the
series. Therefore, a moving average is used: the average of previous and next data is
calculated to obtain the missing data x exlrapg/a/fd, as expressed in (3.2), where x is the series
considered and window the width of the extrapolation (an odd integer larger or equal to
three). This approach is more accurate if the series has a strong autocorrelation. For the
special case of time steps, time step 1 is completed by averaging all the time steps 1 from
D—i and D+i, where i varies from 1 to the half of the window, and so forth for the 47
remaining time steps. The size of the window actually does not matter too much since the
number of non-valid data is low. However, this window was chosen as a function of the
seasonality of the series. Lastly, note that this extrapolating method does not differentiate
working and non-working days.
Sec A ppendix A .l for an explanation o f the basic statistics and m ethods used in this thesis.
x[k]+ £ * [£ ]
nvidrn—1 £=D+1
X (xmpo!atti \ = . , I
window —1
3 .3 .4 .4 O th er d a ta
The RTE’s r e s e r v e demand for EDF Producer, as well as EDF Producer’s programme d'appel
(see Table 3.1), have been retrieved as spreadsheets from 01/01/2005 to 31/08/2007.
The target frequency of die secondary frequency control for each day during this
period has also been provided (i.e., 49.99, 50.00 or 50.01 Hz, see section 2.3.4.3). These data
are useful to calculate the cost of time control (see section 3.6), since the target frequency
impacts the volume of frequency control reserve.
3.3.5 Hardware and software used

This section describes briefly the hardware and software used in this study.
3 .3 .5 .1 APOGEE
As presented in section 3.3.2, APOGEE optimizes EDF Producer’s thermal units and
hydro valleys over two days (i.e., 96 half-hourly time steps). In an optimisation problem, the
primalproblem is to minimize the objective function / ( x px ,...x j subject to the j constraint
functions mj( x ,,x 2...xn) >where (x t, x ?...xtt) are the primal variables, for which the values are
in the feasible set:
/ (xq,x*...x*) = min / (x !,x2...xn) subject to co (xpx2...x/t) = 0 (3.3)
The primal problem is often difficult to solve directly because the constraints create a
coupling between the primal variables, so the dualproblem, which leads to the same solution
if the objective function f ( x px 2...xH) is strongly convex85, is often used instead and is
defined as:
L,(Xf,x*2...xl,\*...X*.) =m ax m in L,(xvx?...xa,X1. . . \ ,) (3.4)
85 A strong])' convex two-dimension set is a set for which the segm ent betw een any pair o f its points is
contained in the set, boundaries excluded. A realjunction is strongly convex if the set o f points lying on or
above its graph forms a strongly convex set.
where ( x * i s the solution of the problem, min L ( x px r ..xn,Xv .X I) the

^ XpX2„.XH •*
Lagrange dual function and E the Lagrangian function defined as:
L ( x px 2...x)l,Xr .XJ) = f ( x px 2...x>l) + XJw1(x px 2,..xil) + ... (3.5)
and where (Xr .X .) are called the Lagrange multipliers. They represent the worst penalty that
can be applied to the violations of constraints (the worst constraints imply that the solution
found is the best). Note that the dual problem is feasible only if it is separable (i.e., if the
primal objective function and the constraints are additive, or in other words, if the
minimum of the Lagrangian function can be found for a given set of Lagrange multipliers).
This method consisting in putting the constraints into the objective function is often called
dual optimisation technique or Lagrangian relaxation [e.g., Lasdon (1968), Wood and
WoUenberg (1996), Bertsimas and Tsitsiklis (1997) or Kirschen and Strbac (2004a)].
To find the solution of the dual problem, A PO G EE uses an algorithm based on the
price decomposition (see Figure 3.5) [Lemarechal (2004)]. The coordination algorithm
increases or decreases the prices for power and reserves as a function of the mismatch
between the target and the dispatch. These prices are the Lagrange multipliers. The sub
problems are then optimised individually, where each thermal sub-problem is solved using
Dynamic Programming (DP)86 and each hydro sub-problem is solved using Linear
Programming (LP)87. Each sub-problem maximizes the profits that would arise if the
Lagrange multipliers were prices that could be turned into income. Note that the generation
cost of hydro is equal to the value of water times the quantity of water released. In addition,
every group has a penalty to switch from reserve provision to no reserve provision (and
vice versa). This cost has no physical meaning, and is only a penalty to “encourage”
APOGEE to propose a “smooth” program, i.e. that avoids erratic switchings.
Following this first optimisation, the dual variables (i.e., the prices for power and
reserves) are obtained. A second optimisation is then performed to fine-tune the first
solution. The augmented Lagrangian method is used in this case [Cohen (1978)]. To
calculate the cost of the whole dispatch, APO G EE penalizes the gap between the target
86 A D P problem is m odelled as a set o f various states. T he optim isation algorithm then tries to
maximize the gain to switch from different states.
87 In a LP problem , th e objective function to solve is a set o f linear functions. In A P O G E E , the
simplex m ethod is used to solve this problem .
and the dispatch for power and reserves. Penalties are the same for all the first 48 time steps
of the day. They then decrease by 5 % per time step.
Concerning A PO G EE’s convergence performance, the same penalties and the same
number of iterations are applied to the two dispatches of a given day (i.e., with and without
reserves). However, note that these penalties and the number of iterations have actually
evolved over the period studied, as a function of computing power and operational policies.
The reader should keep in mind the various limits of APOGEE:

>4 The modelling of hydro units simplifies the optimisation problem. Indeed, as soon as a
hydro unit provides energy, the model dispatches it to provide resemes as well. This
modelling brings a strong bias (see section 3.4.4);
S The tertiary reserve is not included in the calculated de-optimisation cost because it is
currently not optimised by APOGEE two days (whereas it could do so). Indeed, EDF
Producer and RTE agreed that specific units (some hydro units and gas turbines)
provide this reserve;
^ The re-dispatch by the operators is by definition not modelled;
S The costs considered in A POGEE are only the variable costs (e.g., the variable cost of
nuclear units is very low in APOGEE, whereas their average cost is much higher);
* Some costs, such as the value of water, are not physical. Instead, they represent an
evaluation of the use of a limited resource (e.g., the water that falls from the sky or
nuclear fuel that cannot be refilled easily). Note that the hydro cost may be negative if
reservoirs have higher levels of water at the end of the day than they had at the
beginning;
S A POGEE manages some constraints by simply giving a cost to these constraints,
whereas COCKPIT calculates the cost from the actual dispatch (e.g., APOGEE
penalises a group that switches from frequency control provision to no provision,
whereas this cost is not physical). Therefore, the dispatch cost calculated by APOGEE
is less precise than the cost issued from COCKPIT and errors of some tens of k€ may
affect the results;
As shown previously, the marginal costs are actually the dual variables in the solution of
the first optimisation problem;
APO G EE does not converge exactly. Therefore, a financial penalty has to be attributed
to the gap in order to compare the costs of dispatches. This penalty is obtained by
multiplying for each time step the gap with its corresponding marginal cost (see
section 3.3.6).
From a practical point of view, A PO G EE runs under Linux. An A POGEE

simulation lasts on average between 15 and 20 minutes on a dedicated machine (bi
processor Intel® Xeon® 3.20 GHz, 1 MB of cache and 1 GB of RAM). A PO G EE is
stored as an executable file, and a Korn Shell (KSH) script has been developed in our study
to automatically call A PO G EE for a large number of datasets. The executable file used in
this study was the A PO G EE version from August 2007. Note that the original datasets (i.e.,
with the reserve constraint) have been run again with this executable in order to be able to
compare the two dispatches. This takes care of the fact that the original datasets had been
optimized with previous versions of APOGEE.
Coordination algorithm
Prices for pow er and reserves for each time step Generation
planning
Thermal Thermal Hydro Hydro

unit 1 unit n, valley 1 valley nh
Sub-problems with local constraints
Figure 3.5: A P O G E E algorithm
3 .3 .S .2 O T E S S
APOGEE uses text files as inputs and outputs (around 6 MB/dataset). In order to manage
such a large set of data (1 758 valid days and 10 GB o f data), a specific tool has been
developed in this study, named OTESS (OuTilpour PEtude des Services Systeme - Tool for the
study of ancillary services) and described in more details in Appendix A.2. The purpose of
this tool is to modify the APOGEE input datasets, store the A POGEE output data in a
relational database and then to study the data through a web-basedinterface (see
Figure 3.6). The main features of OTESS are the following:
S Handle and modify inputs and outputs of ED F’s dispatch algorithm;
S Extract the selected data from the A PO G EE output files and store them into a
relational database;
S Store any kind of additional data in its database, such as such as the target frequency,
the balancing mechanism prices or the final dispatch planning (the programme d’appel)-,
S Easily modify data with user-defined scripts;
V Allow complex querying on cross-related tables;
^ Display data with various graph types directly from queries(e.g., scatterplots or
frequency distributions). Most of the graphs of this chapter are extracted from OTESS;
S Perform basic data mining techniques;
S Export data under various formats (e.g., CSV or XML88);
S Share data between users since it is a server-based softwarewitha web-based human-
machine interface;
S Run on both Linux and Windows environment;
'S Based on free software;
^ Is strongly modular since it is object-oriented.
1. Dataset with 2. Dataset with

Managed w ith OTESS no demand for frequency
Q original constraints
»| control reserves
E DF’s dispatch
a lgo rith m (APOGEE)
Dispatch 1 ]») Dispatch 2
Core o f OTESS
Database
otess
®?<.F J f l ||J ^ § & a U w d r e h u u s e O utputs R esource* Web-based
human-machine
Management of dataset^ ^ f ^ , script '1. Initial interface
Figure 3.6: Principle o f OT E SS
3 . 3 . 5 . 3 N A G a d d - in for E x c e l
The partial autocorrelation function (PACF) is useful to identify seasonality components in

a time series (see Appendix A.l). As the algorithm of the PACF is relatively complex, it has
88 CSV stands for Comm a-Separated Values and XML for Extensible M ark-up Language
not been implemented in OTESS yet. Instead, the Microsoft® Office Excel® statistical
add-in from Numerical Algorithms Group (2007) has been used in this study after having
exported the data under CSV format.
3.3.6 Cost calculation
In this study, the dispatch cost is directly extracted from APOGEE, and not from
COCKPIT. Therefore, some errors of tens of k€ may mar the results because A POGEE
does not exactly calculate die cost from the actual dispatch (see section 3.3.5.1).
In addition, A POGEE does not converge exacdy, as shown by the frequency

distributions of Figure 3.7, which represent the difference between the target value for a
given time step and the value actually dispatched by APOGEE. Note that the convergence
is better for reserves than for energy because the penalties are higher for the former than
for the latter. Since APOGEE does not exacdy converge to its target value, a money penalty
has to be attributed to the gap to be able to compare costs of dispatches. This penalty is
obtained here by multiplying for each time step the gap with the corresponding marginal
cost. This calculation is shown in relations (3.6) and (3.7), where X'p and X'R represent
the different marginal costs, P the power dispatched during the time step (in MW), Rsei. the
secondary reserves (in MW), Rpn the primary reserves (in MW) and D the day o f dispatch.
c D = C + c f + ^ (pena/ty'p + Penalty^ + PenaltyR j (3.6)

;=l
With:
Penaltyp —Xp(p/f/W P ^ u/j)

PenaltyR —max
PenaltyR —max 0 ; X'R (r '
V ” ' P demand
- R Pnj dispatch,)'i
These relations are worth some comments and some remarks from section 3.3.5.1
have to be highlighted. First, these costs and penalties are calculated for the day D because
uncertainty is too high for D + l to consider gaps for D + 1 as wrong. Second, tertiary reserve
is not included in (3.6) because it is currendy not optimised by A PO G EE two days
(whereas it could do so). Indeed, ED F Producers and RTE agreed that specific units (hydro
units and gas turbines) provide this reserve. Third, the penalty for reserves is zero when
APOGEE provides more reserves than its target. Lastly, the hydro cost may be negative if
reservoirs have higher levels of water at the end of the day than they had at the beginning.
In addition, the hydro cost does not represent the actual cost (which is close to zero for
hydro), but instead values the loss of water.
0 2t
0.18-
0 1&
0 14
Distribution
0.12
0 1
00 8
00&
0 04
0 02
-Soo" -250 -200 -150 -100 -50

Gap (MW)
(a) Power
0 2t
0.18t
0 16
0 14
Distribution
01
0 08
0 06
0 04
002
-300 -250 -200 -150 ^100 -50

Gap (MW)
(b) Secondary reserves
0 18
0 16
0 14
0 12
Distribution
0 1
0 08
006
0 04
0.02 J"
-& T -250 -200 -150 -100 -50

Gap (MW)
(c) Primary reserves
Figure 3.7: Frequency distributions o f the gap b etw een A P O G E E dispatch and A P O G E E dem and
for all the studied tim e steps b etw een 1 and 48
3.4 DAY-AHEAD DE-OPTIMISATION COST FOR A PRODUCER 165
3 .4 Day-Ahead D e-O ptim isation C ost for a Producer
While the previous section presents the methodology adopted in this study, as well as the
hypotheses made, this section gives the results obtained concerning the de-optimisation
cost. For obvious confidentiality reasons, only data that do not reveal EDF costs, while
providing essential results are given here. These results can be applied to some extent to
other systems. However, die best would be to apply the same methodology with the actual
portfolio and optimisation algorithm of each system.
3.4.1 De-optimisation cost over two and a half years

Figure 3.8 shows the evolution of the relative de-optimisation cost from 01/01/2005 to
30/08/2007 for all the studied days (day 1 represents the 1st of January, while day 365 is the
31st of December). The relative de-optimisation cost ik_optbllhlirim is obtained by dividing
the de-optimisation cost Cft_ by the dispatch cost with reserves as shown
in (3.8). The missing days have been extrapolated with a 7-day moving average (see
section 3.3.4.3). Extrapolated points have been marked by crosses in Figure 3.8.
de-optimisation
relathx de—optimisation
(3.8)
irith resen ys
0.08
0.07
2005 2006
0.06
.2 0.05
0.04
2007
% 0.03
0 .0 ’
120 150 180 210 240 270 300 330 360

Day
-v 2005 x 2005 >v 2006 x 2006 x 2007 -v 2007
Figure 3.8: E volu tion o f the relative d e-op tim isation c o st from 0 1 /0 1 /2 0 0 5 to 3 0 /0 8 /2 0 0 7
First, the de-optimisation cost tends to be relatively highfrom the beginning ofA pril (around day
90) to the end of October (around day 300), with the most expensive days in April 2005 and
August 2006. The high de-optimisation cost during summer can be explained by the
environmental constraints, which lead to less hydro power being available. Nevertheless,
2007 sees a low cost during the summer, which may be explained by both a better
modelling of the capabilities of hydro units and a better water availability. Second, the
relatively low cost during the whiter hides a high absolute de-optimisation cost. Indeed, the dispatches
are expensive during this period because the portfolio is constrained by the high demand
for energy. The parameters influencing the costs will be discussed in more details in
section 3.4.3.
Besides giving some hints on the parameters that may influence the de-optimisation
cost, Figure 3.8 gives two important results. First, the day-ahead de-optimisation cost is not
negligible, since it represents up to 7.8 % of the initial dispatch cost. In particular, Figure 3.9
shows the frequency distribution of the relative de-optimisation cost Cfela(i!Kde_optimjmtion. The
average of the relative de-optimisation cost calculated by APO G EE two days was equal to
2.2 % over the studied period. Therefore, this cost cannot be ignored, since it represents
hundreds of millions of Euros every year. In addition, note that this result is in line with
Proctor (2001), who estimates the spinning reserve costs around 2 % of the total operating
expenditure in a 16-unit system.
Second, the day-ahead de-optimisation cost exhibits strong daily variations. To evaluate the
importance of these daily variations, we define A%C^_optim0alioll as in (3.9). This indicator
gives the relative variation of the de-optimisation cost for D in comparison to D -l. The
frequency distribution of this indicator is given in Figure 3.10. It is clear that the de
optimisation cost has a strong variation (greater than ±20 % in more than 54 % of the
days). The design of the market should therefore take this variation into account to help
participants seize the opportunities (see Chapter 4).
1
D dr- optimisation de—optimisation
A%C de—optimisation D-l (3.9)
C de—optimisation
0.15 t
0.14
0.13
0.12
0.11
>,0.09
| 0.08
£ 0.07
£ 0.06
0.05
0 04
0.03
0.02
0.01
- . 1 1 1 l l .l 1 1 1 «=w =
0.01 0.02 0.03 0 04 0.05 0.06 0 07 0.08
Relative deoptimisation cost
F ig u re 3.9: F re q u e n c y d is trib u tio n o f th e relative d e -o p tim isa tio n c o st for all th e stu d ie d d a y s8
89 T he term “probability” is used here instead o f “relative frequency” by misuse o f language (see
A ppendix A.1.1).
0.13r
0.12
0.11
0.1
0.09
>. 0 08
1 0.07
| 0.06
it 0.05
0.04
003
0.02
001
-150 -100 -50 100 150 200 250

Variation of the de-optimisation cost over two days (in %)
Figure 3.10: Frequency distribution o f the variation o f d e-op tim isation c o st over tw o days for all the
studied days
3.4.2 Seasonality of the de-optimisation cost

Producers are interested in any information that can help them to estimate their future
costs. Therefore, a study of the seasonality of the de-optimisation cost has been performed.
In addition to being useful for the producer, understanding the seasonality of the de
optimisation cost is also very useful for the policy makers (see Chapter 4). Lasdy, spotting
the seasonality can reduce considerably the need for further studies by concentrating
calculation on specific periods. The absolute de-optimisation cost is used here instead of the
relative de-optimisation cost, since this parameter is the most interesting for the producer.
In addition, the seasonality results obtained with the relative de-optimisation cost turned
out to be of little interest.
To investigate the seasonality of the de-optimisation cost, two complete one-year

periods*’ are considered here. Figure 3.11 shows that the autocorrelation functions (ACF)
for these two periods have relatively high correlation coefficients for lags of less than 30
days. In other words, the de-optimisation cost is somewhat correlated to the 30 previous
days. To confirm this trend, the partial autocorrelation functions (PACF) have been drawn
for various periods (see Figure 3.12 to Figure 3.15). Figure 3.12 exhibits a relatively low
seasonality effect for periods greater than one week, except two peaks around 30 and 90
90 From 0 1 /0 1 /2 0 0 5 to 3 1 /1 2 /2 0 0 5 and from 0 1 /0 8 /2 0 0 6 to 3 1 /0 7 /2 0 0 7 (see section 3.3.4.2).

Missing data have been extrapolated with a 7-dav window according to the m ethod described in
section 3.3.4.3. As the num ber o f extrapolated points is low (around 3 %), it is adm itted that the missing data
do not bring a strong bias in this study.
days. Figure 3.15 shows a clear weekly seasonality, while Figure 3.13 and Figure 3.14 analyse
in more details die two peaks around 30 and 90 days. The partial autocorrelation
coefficients are however not large enough for these two peaks to be really significant.
Lastly, there is no seasonality effect over longer periods, such as several months.
In conclusion, the de-optimisation cost exhibits mainly a weekly seasonality. Therefore, a

week is likely to be the best timeframe to help estimate the future de-optimisation cost.
However, it does not seem possible to select a particular period and then to generalise the
result of this period over the whole timeframe. Nevertheless, the studied timeframe is
probably too small to detect clear long-term seasonally effects. Hence, this study should be
completed in the future by using data over a longer period.
0.5
0.4
c 0 ,3
.9
go 0-2
oo
3 0.1
<
100
- 0 .2 J
w From 01/01/2005 to 31/12/2005 -v From 01/08/2006 to 31/07/2007
F igure 3.11: A utocorrelation function o f the d e-op tim isation c o st up to 1-year period
0.4
0.2 -
LL.
o<
CL
- 0.1 -
Period
-0.2 J
— From 01/01/2005 to 31/12/2005 — From 01/08/2006 to 31/07/2007
F igure 3.12: Partial autocorrelation fu n ction o f the d e-op tim isation c o st up to 1-year period
0.1
0.05
91 92 96
-0.05
- 0.1
Period
-0.15
■ From 01/01/2005 to 31/12/2005 ® From 01/08/2006 to 31/07/2007
Figure 3.13: Partial autocorrelation fu n ction o f the d e-op tim isation co st around a 3-m on th period
0.2
0.15
0.1
0.05
20 21 22 27 28 U 35
-0.05
- 0.1
Period
-0.15
■ From 01/01/2005 to 31/12/2005 P From 01/08/2006 to 31/07/2007
Figure 3.14: Partial autocorrelation fu n ction o f the d e-op tim isation c o st around a 1-m onth period
0.7 -|
0.4 -
u.
O
<
CL
0.2 -
Period
■ From 01/01/2005 to 31/12/2005 EFrom 01/08/2006 to 31/07/2007
F igure 3.15: Partial autocorrelation fu n ction o f the d e-op tim isation c o st around a 1-w eek period
3.4.3 Param eters affecting the de-optimisation cost

From the method presented in section A. 1.4, the following sections investigate the link
between the de-optimisation cost series smoothed over a week according to (A. 1.13) and
other parameters, which have also been smoothed over a 7-day window and normalised
according to (A. 1.14). The 879 studied days are considered (see section 3.3.4.2). The three
selected parameters amongst the parameters described in section A.2.4 are: (a) the marginal
costs of reserves; (b) the demand for reserves; and (c) the thermal/hydro mix for the
provision o f reserve. These parameters have been selected because they influence most the
de-optimisation cost according to the performed multidimensional data analysis, i.e. these
parameters have high correlation coefficients with the de-optimisation cost (e.g., see the
screenshot in Figure A.2.10). The results for the other parameters are thus of less interest.
3 . 4 . 3 . 1 I m p a c t o f t h e m a r g in a l c o s t s
Figure 3.16 shows the normalized de-optimisation cost as a function of the weighted
marginal cost of reserves 4^ , as defined in (3.10) and where the notations are similar to the
ones used in section 3.3.6. Since is defined for each time step of the day, the data
plotted here is the average over the 48 time steps of a day. It is clear that the de
optimisation cost and the weighted marginal cost of reserves exhibit a strong linear relation,
confirmed by a correlation coefficient91 r of 0.94. Figure 3.17 shows the de-optimisation
cost as a function of the primary and secondary reserve marginal costs, and a similar linear
91 Values o f basic statistics are defined in A ppendix A .I.

relation is found. Therefore, the expected de-optimisation cost clearly increases ivith the daily average
marginal costs of reserves,
Xn
Kfri
x R.Pn ■
dispatch
T A.1 x R ,
K*r a t dispatch
(3.10)
RpP 1dispatch + R ;^tspaicb
R 2.5
w
nx
1 1.5 2 2.5 3 3.5 4 4.5
E -0.5 Weighted marginal cost of reserves (average)
Figure 3.16: N orm a lized d e-op tim isation c o st as a fu n ction o f the average n orm alized w eig h ted
m argin al c o st o f reserves for all th e stu d ied days (rxy = 0.94)
.3 1.5
2.5 3 1.5 2 2.5 3 3.5

6 -0 . 5- prmary reserve nrargrrial cos! (mnm) eeonflaryrK^&~ma?pi3'3»r6iieaiiY
(a) Primary reserve (% = 0.92) (h) Secondary reserve (r*y = 0.92)
F igure 3.17: N orm a lized d e-o p tim isa tio n c o st as a fu n ction o f th e average norm alized m arginal co st
o f reserves for all the stu d ied days
3 .4 .3 .2 I m p a c t o f t h e d e m a n d for r e s e r v e s
Figure 3.18 shows the relation between the de-optimisation cost and the maximum primary
reserve demand over the day (i.e., max^R^/ rfj). The correlation between these two
variables is not very clear, but the period spanning from 01/08/2006 to 31/07/2007
exhibits a more interesting relation (see Figure 3.19). Indeed, when the maximum demand
increases, the de-optimisation cost becomes more spread with a trend for high expected
values, whereas the expected de-optimisation cost is quite low when the maximum primary
reserve demand over the day is low. This relation is however not true for 2005, as shown in
Figure 3.20, which explains the poor correlation over the whole study period (i.e.,
Figure 3.18). The de-optimisation cost is dispersed for a given reserve amount since the portfolio
conditions may be very different for the same reserve demand. In addition, requiring high
volume of reserve increases the constraints on die dispatch, so the de-optimisation cost tends to
increasefor high reserve demands.
The daily average of primary reserve demand shows a similar relation as the
maximum demand for primary reserve, as shown in Figure 3.21. On the other hand, the
impact of secondaty reserve on the de-optimisation cost seems to be Io n since the correlation coefficient
is equal to -0.18 and since Figure 3.22 does not show any clear pattern. This is probably due
to the fact that the French generating units’ capabilities are twice as large for secondaty
reserves than they are for primary reserves (e.g., most of the French nuclear power plants
allocate around 2 % of their nominal power to primary reserve and 5 % to secondaty
reserve92). Therefore, dispatching primary reserve is more constraining than dispatching
secondaty res ewe for the French portfolio.
In conclusion, primary res ewe demand seems to have an impact on the de
optimisation cost, whereas the influence of secondary reserve has not been clearly
identified. Therefore, increasing the primary reserve provision capabilities may be a good
way to reduce the constraint due to primary resewe demand and thus to decrease the de
optimisation cost. In addition, an accurate generic de-optimisation cost function depending on the
de?nand for reserves cannot be built because the dispersion is high. This latter subject will be
investigated in more details in section 3.4.4.
92 Some nuclear pow er units, such as 900-MW plants, may be dispatched how ever w ith a 7-%
provision o f primary frequency control reserve.
4.5r-
4
3 .5 -
* X
X
X K X ' ....
» 1.5
“ * * v ’ . jj t f *** *
t) -0.5
-2.5 —
F igure 3.18: N orm a lized d e-op tim isation c o st as a fu n ction o f the m axim u m n orm alized primary
reserve d em an d for all the stu d ied days (r*y = 0.39)
o
2 0.51
Normali
F igure 3.19: N orm alized d e-op tim isation c o st as a fu n ction o f the m ax im um n orm alized primary
reserve dem and from 0 1 /0 8 /2 0 0 6 to 3 1 /0 7 /2 0 0 7 (rxy = 0.71)
x x
.... . . x . .
..
x x x n
'H 15 * -
? 15 *2 iS’.S 3 M r
-0.5 naliî prir^ryTesjtn/e ^lepiandf(maiimum)
* -* H f— ~
x * ....
F igure 3.20: N orm a lized d e-o p tim isa tio n c o st as a fu n ction o f the m axim u m norm alized primary
reserve d em an d from 0 1 /0 1 /2 0 0 5 to 3 1 /1 2 /2 0 0 5 (% = 0.13)
# m
a 1.5 % ■*
d (r»M n)
**
Figure 3.21: N o rm a lized d e-op tim isation c o st as a function o f the m ea n norm alized primary reserve
dem and for all the studied days (rX7 = 0.38)
u 2.5
» 1.5
£ 0.5
«2.5 3 3.5 4
.M -0.5 ^(m ean)
F igure 3.22: N o rm a lized d e-op tim isation cost as a function o f the m ea n norm alized secondary
reserve dem and for all the studied days (rxy = -0.18)
3.4.3.3 Im pact o f the therm al/hydro mix
Figure 3.23 shows the relation between the de-optimisation cost and the minimum thermal
share for reserve rnin(share'/t/) over the day. The thermal reserve share sharetb for a time step
i is calculated as shown in (3.11). When the minimum thermal share is high, the de
optimisation cost tends to be high as well. In other words, when hydro cannot he used extensively
to provide reserves for some critical time steps, the de-optimisation cost tends to be high. As a corollary,
thermal participation to reserves tends to be expensive for some critical time steps. This link
between hydro participation and de-optimisation cost is explained in section 3.4.4.
R P” dispatch
J tb + Riff ‘dispafrb lb
share[, (3.11)
RP***dispateb + Rsec’dispatch
4 ---------
jrvqg (minimum)
Figure 3.23: N orm a lized d e-op tim isation co st as a fu n ction o f the m in im u m reserve share provided
by therm al units for all the stu d ied days (rxy = 0.44)
3.4.4 De-optimisation cost and demand for reserves
As shown in section 3.4.3.2, the de-optimisation cost is linked to the daily average reserve
demand, but the relation is not straightforward. Therefore, this section attempts to provide
an insight on this relation by analysing the evolution of the de-optimisation cost for a given
day as a function of the reserve demand. Thursday 24 May 2007 is considered here, because
this is a recent day with an average de-optimisation cost. Indeed, the relative de
optimisation cost for this day is around 1.9 %, whereas the average is 2.2 % for all the
studied days (see section 3.4.1). The initial demand for reserves has been multiplied by a
coefficient ranging from 0 % to 100 %, with a step of 1 %. As 100 % correspond to the
simulation with the initial demand and 0 % to the simulation without reserves, 98 additional
simulations have been launched for this study.
Figure 3.24 shows the relative de-optimisation cost C ^ t!v(d(_opiimisallm{X ) as a function
of the average reserve demand over the day. This de-optimisation cost is redefined as
shown in (3.12), where C ®lhX»/0^ A/mnr( ^ ) corresponds to the dispatch cost when the
demand for reserves is equal to X % of the initial demand, with X ranging from 0 to 100.
The average amount of reserves along the horizontal axis is obtained by summing the
demand for primary and secondary reserves over the 48 steps of the day, and then by
dividing this sum by 48. Below an average demand around 500 MW (i.e., one third o f the
total reserve demand), reserves do not cause any supplemental cost for APOGEE
according to our hypotheses (see section 3.3.3) and the model’s limits (see section 3.3.5.1).
In fact,the de-optimisation cost <k_oplw,iu:lmi(X ) calculated by APO G EE is constantfor an average
demandof reserves between 0 and 500 MW, In other words,the dispatch cost
x %nj inina îe same whatever is the provision of reserves below 500 MW.
However, beyond this point, the de-optimisation cost depends almost linearly on the demandfor reserves,
with a slope around 2.7x1 (T3 % /MW/day, i.e. C fT hx%nf hMajirJMf(X ) increases almost linearly
beyond 500 MW of reserve. Figure 3.25 shows the schematic cost curve superimposed on
the actual calculation.
,y \ C uith
D resort —C nitb
D X % of initial n s m v(PC)
\ )
rrla/iiv de—optimisation \ / y' •D \ ' )
uitb re.te/rr
(3.i3)
C relative
D de—optimisation fOl
\ ) = CD
rclaliif de-optimisation
(3141
V' >
By analysing the dispatches proposed by APOGEE, it can be seen that a large

amount of reserve is still provided despite a null demand. In other words, APOGEE
considers that some reserve provided has no cost, and even may help to save money
because this reserve may be useful to meet the dispatch constraints. This “no-cost reserve”
is mainly provided by hydro units. In fact, 92 % of the primary reserve and 77 % of the
secondary reserve initially dispatched on hydro units were still allocated by APOGEE when
the demand for reserves was null (for the period from 01/08/2006 to 31/07/2007). The
remainder is supplied by thermal units due to the initial constraints imposed by the
operator93. This choice by APO G EE in the allocation of reserves is due to the modelling
and the operating constraints of hydro units. First, the participation in secondary frequency
control is decided prior to APOGEE for several hydro units. Therefore, APOGEE cannot
remove the reserve provision of these units, even if the demand for secondary frequency
control reserves is zero93. Second, since the maximum efficiency for hydro units is at 80 %
of nominal power, it is usually more efficient to set a hydro group at such a level (the group
thus provides reserves with the remaining capacity). Therefore, the operating procedures
specify that a hydro unit should always provide reserve when producing energy. These
93 R em em ber that the choice has been m ade in this stud}7to keep ail die initial operational constraints
to evaluate the de-optim isation cost (see section 3.3.4.1).
procedures thus do not acknowledge that it may be more efficient to set hydro units at
100 % for some expensive time steps, though at degraded efficiency. As a consequence of
these two operating procedures (i.e., pre-scheduled reserves and compulsory participation
when providing energy)94, providing reserves with hydro cause litde de-optimisation cost for
APOGEE, whereas there is an actual de-optimisation cost. It is important to note that the
second part of the curve of Figure 3.25 is likely to give an order of magnitude of the actual
constraints due to both hydro and thermal reserves. Therefore, the actual de-optimisation
cost could be extrapolated by using the second part of the curve for the whole reserve
demand. Furthermore, it is also important to quote that the “free hydro reserve” for the
model (but which has a cost in the real world) is only available during the day, i.e. when
hydro units are synchronised. At night, this is not true because French hydro units are not
synchronised except some units that are able to pump water or those that cannot store
water (run-of-water units). However, this type of units cannot provide reserves while
pumping.
This threshold effect that leads to a non-linearity may thus explain the poor direct relation
between the average reserve demand and the de-optimisation cost (see section 3.4.3.2). This relation is
also in line with the conclusions of section 3.4.3.3, i.e. the de-optimisation cost calculated by
APOGEE tends to increase when the hydro participation in reserve decreases.
is 1-6
S °-8
0>) 0.6
§ 0.4
* 0.2
100 200 300 400 500 600 700 800 900 1000 1100 1200 1300
Average amount of reserves (MW)
•v De-optimisation cost for the 24/05/2007 S' Smoothed de-optimisation cost
Figure 3.24: R elative d e-op tim isation c o st as a fu n ction o f the n ew d em an d for reserve for 2 4 /0 5 /2 0 0 7
9,1 T hese operational limitations will be rem oved in the short term , b u t will require a m ore complex
optim isation algorithm.
3.5 MARGINAL COSTS OF FREQUENCY CONTROL FOR A PRODUCER 179
1.8
Supplemental cost:
1.6 2.7x10-3 %/d ay/MW
Average cost: \ Total cost:

1.5x10‘3 %/day/MW 1.9%
a.
0.8
0.6
0.4
0.2
100 200 300 400 500 600 700 800 900 1000 1100 1200 1300
Average amount of reserves (MW)
F igure 3.25: Sim plified relative d e-op tim isation cost as a function o f th e n ew dem and for reserve for
2 4 /0 5 /2 0 0 7
3 .5 Marginal C o sts of F requency Control for a

Producer
Section 3.4.3.1 has shown that the marginal costs impacts the final de-optimisation cost.
Indeed, the marginal costs of reserves reflect the constraints due to frequency control
provided by a producer. Therefore, this section proposes to study these marginal costs. For
confidentiality reasons, the absolute values are not given. Instead, comparisons between
marginal costs are preferred in order to identify the binding and non-binding constraints.
3.5.1 Study of the binding constraints

The marginal costs help determine which parameter is binding in the dispatch. Indeed, the
marginal cost of a product represents the cost to provide an additional unit of this product
(e.g., a MW). Figure 3.26 gives the percentage of time when marginal costs of energy are
higher than marginal costs of reserves as a function of the time for the 879 days studied (see
section 3.3.4.2). Time step 1 spans from 00:00 to 00:30, while time step 48 spans from 23:30
to 24:00. It is clear that the demandfor energy is most of the time the most binding constraint in the
dispatch. Nevertheless, the reserves are the most binding constraint for some time steps (e.g.,
time step 44, i.e. from 21:30 to 22:00). Flence, the demand for one additional MW of
reserve is likely to cost more than the demand for one additional MW for energy for these
time steps. It might therefore be cost effective to develop storage or load control strategies that could be used
180 CHAPTER. 3 COST OF ANCILLARY SERVICES
during critical time steps in order to allow a conversion of the cheaper energy into reserves. However, such
arrangements would be profitable only if the difference between the marginal costs of
reserves and energy is large enough. For confidentiality reasons, this data cannot be
published here.
To assess the influence of the type of frequency control, Figure 3.27 shows the
percentage of time when marginal costs for primary reserve are higher than marginal costs
for secondary reserve as a function of the time of the day for all the studied time steps. It is
clear that primary reserves are usually more binding than secondary reserves, which is in line with the
results of section 3.4.3.2. Indeed, this section shows that the de-optimisation cost is more
sensitive to the demand for primary reserve than the demand for secondary reserve.
100 ,
90 —
£ 70 \
! 60 |
O 50
E 40
§ 30
a>
a 20
10
Time step
F igure 3.26: P ercentage o f tim e w h en m arginal costs o f energy are h igh er th an m arginal costs o f
reserves as a fu n ction o f the tim e step for all the stu d ied days
£ 30
30
Time step
F igure 3.27: P ercentage o f tim e w h en m arginal costs o f primary reserve are h igh er th an m arginal
c o sts o f secondary reserve as a fu n ction o f the tim e step for all the stu d ied days
3.5 MARGINAL COSTS OF FREQUENCY CONTROL FOR A PRODUCER 181
3.5.2 Study of the non-binding constraints

A null marginal cost for frequency control reserves implies that the demand for reserves is
not a binding constraint. In other words, when the marginal cost for frequency control
reserves is null, reserves do not de-op timize the dispatch of the producer. Figure 3.28
represents the percentage of time when the marginal costs of reserves are zero95 as a
function of the time of day for all the days in the study. In the middle of the night, i.e.
between 02:00 and 06:30, the marginal costs of reserves are quite often equal to zero.
Therefore, it. would bepossible to increase the security of the system at least cost by increasing the demand
for reserves during thisperiod.
20r
18
16
g 14
0
E 12
20 30
Time step
-v Null primary marginal cost v Null secondary marginal cost
Figure 3.28: P ercen tage o f tim e w h en m arginal costs o f reserves are null as a fu n ction o f th e tim e o f
the day for all the studied tim e steps
95 M ore precisely, inferior to 1 ( € /MW) / h.

3 .6 C ost of T im e Control in F rance
RTE modulates its demand for primary frequency reserves by ±150 MW because of the
time control. In fact, the time control modifies the system frequency, which leads to the
deployment of primary frequency control power (see section 2.3.4.3). As the demand for
reserves impacts die de-optimisation cost (see section 3.4.3.2), this section estimates the de
optimisation cost because of the modulation of the demand for reserves. Therefore, new
simulations were launched with the following changes in the initial datasets:
S Targetfrequency at 49,99 ITy. the initial primary reserve demand is decreased by 150 MW for
each time step of the day;
V Targetfrequency at 50.01 Hy: the initial primary reserve demand is increased by 150 MW for
each time step of the day.
Note that these values correspond to the modulation of the demand for the whole
of France. Therefore, by applying these values to EDF Producer’s portfolio, the de
optimisation for the whole France can be estimated. It is thus assumed that the French cost
structure is equivalent to the one of EDF Producer.
Figure 3.29 and Figure 3.30 give the frequency distribution of the relative de
optimisation cost due to frequency control for all the studied days from 01/01/200596. On
average, the time control cost (because of an increased reserve demand) 0.46 % of the initial
dispatch cost at 49.99 Hz, and saved (because of a decreased reserve demand) 0.57 % of the
initial dispatch cost at 50.01 Hz. The average cost obtained with the target frequency set to
50.01 FIz has to be taken with caution, since the number of days is not sufficient to provide
very significant results, and because the scattering of data is high (27 days at 50.01 Hz,
whereas there were 175 days at 49.99 Hz). The de-optimisation cost due to time control thus
represents tens of millions of euros over oneyear. One can thus wonder whether this cost is justified.
Lasdy, it is important to note that the de-optimisation cost due to time control is included
in the de-optimisation cost calculated in section 3.4, since the initial demand for reserves
takes into account the time control.
% T he time control study is perform ed from 0 1 /0 1 /2 0 0 5 because available data on target frequency
span from 0 1 /0 1 /2 0 0 5 to 3 1 /0 8 /2 0 0 7 (see section 3.3.4.4).
3.7 CONCLUSION 183
0.16,
0.14
0.12
0 . 1*
| 0.08
o
£ 0 06
0 04
0 02
Hi
02 04 0.6 0.8 1 1.2
Relative de-optimisation cost (in %)
Figure 3.29: Frequency distribution o f the relative de-op tim isation cost due to tim e control w ith
f, = 49.99 H z for all the studied days from 0 1 /0 1 /2 0 0 5
0.16,
0.14
0.12
>. 0.1
fg 0.08
.a
o r- rn r~ n
£ 0 06
0.04 _ jjfc
0 02
° l8 7T6 -1.4 -1.2 1 -0.8 -0.6 -0.4 -0.2 0

Relative de-optimisation cost (in %)
Figure 3.30: Frequency distribution o f the relative d e-op tim isation co st due to tim e control
f, = 50.01 H z for all the studied days from 0 1 /0 1 /2 0 0 5
3.7 C onclusion
This chapter has reviewed the main cost components of ancillary sendees and has proposed
a methodology to estimate the day-ahead de-optimisation cost due to frequency control
incurred by a producer. This methodology has been successfully applied to EDF Producer,
and has provided several interesting results. Therefore, this methodology could be applied
to other producers in order to perform a similar economic assessment. The main findings
of this chapter are listed below. Policy makers could use these findings to assess their
current policies.
S Estimating the full real cost of ancillary services is impossible for a producer.
■ Ancillary services exhibit both fixed and variable costs that have each several cost
components such as over-investment, capacity reservation or utilisation;
■ Different inter-dependent time horizons have to be considered, from tens of years ahead
to real-time. For example, the day-ahead de-optimisation cost is linked to the over
investment that has been done several years before;
■ It may be difficult to separate the cost due to ancillary servicesfrom the cost due to otherproducts
such as the provision of energy. A classical example is the separation of the cost of
reactive and active power provisions with an alternator;
■ The cost of ancillary services depends on the portfolio considered and it is affected by many
uncertainties, especially for utilisation costs.
V One main cost component of frequency control is the day-ahead de-optimisation cost due to
the reservation of the capacity to provide reserves. A. methodology has been proposed to
estimate this cost and a specific tool, named OTESS, has been developed to apply this
methodology to any producer;
V This methodology has been successfully applied to ED F Producers portfolio (i.e., around 150
thermal units and 50 hydro valleys) over 879 days with actual operational data and
ED F’s actual optimisation algorithm;
V The results obtained are very helpful to assess the frequency control provision in France. In
addition, they give some hints to improve the system regarding both the technical
aspects and the market design. In particular:
* The de-optimisation cost is not negligible, since it represents up to 7.8 % of the initial
dispatch cost. In addition, this result is an underestimate because of the modelling
limitations (numerous inflexible constraints for hydro units) and the methodology
(short-term approach and unchanged initial constraints). Therefore, an efficient
provision of frequency control reserve is essential to reduce this cost as much as
possible;
■ The de-optimisation cost exhibits strong variations (greater than ±20 % in more than 54 %
of the days). Consequently, the market design should take this variation into
account to help participants reap opportunities;
■ The seasonality of the de-optimisation cost is relatively poor; even if the auto-correlation of
the de-optimisation cost is quite good over one week. Therefore, it does not seem
possible from the timeframe considered to select a particular period and then to
generalise the result of this period over the whole timeframe. In particular, results
3.7 CONCLUSION 185
cannot be generalized from one year to another. Hence, future studies on the de-
optimisation costs have to consider a large number of days to be accurate. In
addition, these further results may help to identify potential seasonalities;
■ For EDF Producer, the de-optimisation cost is sensitive to three majorparameters'.
n The expected marginal costs, there is a strong link between the marginal costs of
reserves and the de-optimisation cost. Therefore, the marginal costs can be used
as a proxy to estimate the de-optimisation cost;
n The primaty reserve demand: primary reserve demand has an impact on the de
optimisation cost, whereas the influence of secondary reserve has not been
clearly identified. Therefore, increasing the primary reserve provision capabilities
may be a good way to reduce the constraint due to primary reserve demand and
thus to decrease the de-optimisation cost. Unfortunately, it is impossible to
build a generic de-optimisation cost function depending on the demand for
reserves because the dispersion is high;
n The quantity of reserves provided by hydro units, when hydro cannot be used
extensively to provide reserves for some critical time steps, the de-optimisation
cost tends to be high. As a corollary, thermal participation to reserves tends to
be expensive for some critical time steps. This trend is due to a somewhat
simplistic modelling of hydro units in the optimisation algorithm and should be
weaken in the future with more complex constraints taken into account.
■ The demandfor reserves can be the most binding constraint. Therefore, the demand for one
additional MW of reserve can cost more than the demand for one additional MW of
energy. It may thus be useful to develop storage or load control strategies for these
critical time steps in order to allow a conversion of the cheaper energy into reserves;
■ The reserves may not be binding at all (from the model’s perspective), so the security of
the system can be increased at very low cost by increasing the demand for reserves
during periods when the marginal costs of reserves are low. This result is in line
with the conclusion of Chapter 2 arguing the advantages of an elastic demand for
reserve.
■ The de-optimisation cost due to time control represents tens of millions of eurosperyear in France.
The value of such an expenditure needs to be proven. ■
CHAPTER 4
PROCUREMENT OF SYSTEM
SERVICES
It is often the solution directly set aside that mould have been the tight one.
Edouard Michelin (1859 - 1940)
4.1 Introduction
NCILLARY sendees and energy are most of the time provided by the same
A generating units, and the provision of AS generally reduce the amount of energy that
a generating unit can produce, and vice versa. Ancillary sendees can thus be viewed as an
externality of the energy market. To manage externalities, economists recommend using
property rights [Boucher et al. (2006)]. Property rights are the rights of an individual to own
property and keep the income earned from it. If these property rights are well defined, trade
between agents would result in an efficient allocation of the externality [Varian (1999)]. In
other words, the externality becomes a constraint or a trade-off opportunity. In addition,
Chapter 3 has shown that die cost of providing ancillary sendees is not negligible. These
two reasons motivate the creation of a market for ancillary sendees to procure system
services.
A Comprehensive Âssessment of Marketsfor Freqtten/y and Voltage Control Ancillary Services, Yann Reboucs, 2008 187
188 CHAPTER 4 PROCUREMENT OF SYSTEM SERVICES
As capacities for energy and AS are physically linked and since AS often involve the
production or consumption of energy, markets for energy and ancillary services present
numerous similarities, while trying to take into account AS specificities described in
Chapter 2. A lot of attention has thus been concentrated on defining pricing mechanisms to
coordinate ancillary services with the markets for electrical energy in a centralised unit
commitment structure. However, by focusing on centralised pricing and assuming that a
spot market combined with bilateral contracts is the best model, one misses some aspects
of the AS markets, which are not only crucial but also very different from what one finds in
energy markets. Difficult issues that have not received enough attention include for example
the procurement of AS in the long-term, a coordinated clearing of markets for energy and
reserves with a decentralised unit commitment, or an allocation of the cost of AS that gives
proper incentives.
This chapter identifies and discusses eight fundamental issues in the design of
markets for ancillary services, namely: (1) choosing the entity responsible for AS
procurement (section 4.2); (2) matching demand and supply (section 4.3); (3) choosing the
relevant procurement methods (section 4.4); (4) defining the structures of offers and
payments (section 4.5); (5) organizing the market clearing procedure (section 4.6); (6)
avoiding price caps (section 4.7); (7) providing efficient incentives (section 4.8) and (8)
assessing the procurement method (section 4.9). Addressing these eight issues is essential to
the development of electricity markets that are not only efficient but also durable. This
chapter does not pretend to propose a standard market design, since all market
architectures should be tailor-made and evolve over time in relation to the market
structure97. Instead, it discusses current issues and proposes solutions for most of the
features that a market has. In addition, it surveys actual practices in order to illustrate the
discussion. This survey complements previous works by, amongst others, Kirby and Hirst
(1996), Cali et al. (2001), Arnott et al. (2003), Zhong (2003), Eurelectric (2004) and Raineri et
ah (2006).
It is assumed in this chapter that the required quantity and location of each quality
of system services have been defined for the various time horizons. The discussion in this
chapter is focused on the issues related to the procurement of the required system services.
97 T he market architecture describes how the m arket works (e.g., defining the auction process), while
the m arket structure refers to w hat is traded and by w h o m (e.g., a lo t o f small companies).
4.2 NOM INATING TH E ENTITY RESPONSIBLE OF PROCUREMENT 189
Procurement is the act of ensuring the presence in the system of the necessary SS (provided
in part -with the help of some AS). As general rules, the procurement of system services
should [Cali et al. (2001)]: (a) permit participants to recover the investments with enough
return; (b) reduce the total cost to consumers of SS; (c) give an incentive to improve the
sendee level; (d) find a price between the value and the cost of the sendee; (e) have a price
setting mechanism with a reasonable transaction cost98. The present chapter will mainly
discuss the short-term procurement (say up to two or three years), even if most of the
concepts are also applicable to long-term procurement.
4 .2 N om inating th e Entity R esp o n sib le o f P rocurem ent
The definition and the deployment of system sendees is a responsibility that falls naturally on
the system operator, even if it does not always control directly some of these sendees, such
as the primary frequency control ancillary sendee (see section 2.3.3.1). On the other hand,
the Entity Responsible for Ancillary services Procurement (ERAP) can be either the system
operator or some users of the system.
A tempting solution is to put several users of the power system into competition for the
procurement of AS. For example, the load-serving entities (LSEs) of PJM must provide
frequency control sendees such as regulation and synchronised resenê in pro rata of their
load share [PJM (2008)]. This approach creates some competition between AS buyers and
providers, which should result in lower prices. It also implicitly allocates the cost of ancillary
sendees amongst LSEs. However, this approach has three weaknesses. First, the LSEs
themselves are not users of AS and therefore do not know the true value of AS. Therefore,
this results in an inelastic demand, which may jeopardise the match between supply and
demand (see section 4.3). Second, the quantity, quality and location of system sendees are
set by an entity that has no incentive to reduce the demand, because it does not pay for
then procurement. Last, imposing such requirement on users can be a barrier to the entry
of new competitors if the level of AS required from each participant is too high.
98 This list is n o t exhaustive. F o r example, increase innovation may also be an objective o f the
procurem ent.
Another popular approach is to designate the system operator as the single buyer'9. This
solution solves the problems mentioned above but raises others issues, common to most of
regulated activities. First, the SO needs to recover somehow the costs incurred, which is an
issue by itself addressed in section 4.8.2. Second, the management rules of the SO have to
be transparent in order to ensure a fair treatment of all SS users, whatever their position in
the industry (e.g., incumbent or new entrant). Third, as discussed in section 4.8.5, the SO
should have enough incentives to optimise its procurement process. In particular, if the TO
and the SO are the same entity, this entity should have enough incentives to choose the best
procurement processes and not to select only the self-procurement (see section 4.4.1). In
conclusion, designating the system operator as the ERAP is probably the most desirable
solution if proper incentives are put in place by the regulator.
4 .3 M atching Supply and Dem and

If supply and demand do not match in the long run as well as in the short run, the market
will fail. As shown in Chapter 1, the supply of ancillary services is mostly ensured by
generating units. Marginally, other participants also provide ancillary services, such as means
of storage that smooth either consumption or generation; consumers that can modulate
their consumption upon request or automatically; or transmission owners that help to
control voltage with purpose-built equipments. The demand for ancillary services is defined
by the system operator and depends on the power system structure (see Chapter 2).
4.3.1 Long-term matching

Ideally, AS markets should provide sufficiently strong signals to trigger investments in new AS
capacities and thus ensure in the long mu: (a) enough AS to meet demand; (b) a liquid
market; and (c) innovative AS resources that are competitive. However, because revenues
from AS are highly uncertain and because of the long life of generating assets, no specific
incentive scheme has been found yet to encourage efficient investments in AS capacity.
Therefore, the long-term supply of AS is currently ensured through the imposition

of connecting conditions on generating units. Still, since AS capabilities deeply rely on the
99 A single buyer can force th e providers to supply AS (i.e., the com pulsory procurem ent described in
section 4.4).
4.3 MATCHING SUPPLY AND DEMAND 191
adequacy of investments in energy generating capacity, which remains an open question in

deregulated electricity markets [e.g., see Baldick et al. (2005) or Cramton and Stoft (2005)],
the long-term procurement of AS remains an important issue. In addition, connecting
conditions are likely to be non-optimal since they somewhat arbitrarily increase energy
prices (because they raise the cost of building generating units) and decrease payments for
ancillary sendees (because they artificially increase the supply of AS), Since SOs are
generally responsible for both the definition of the connecting conditions100 and the
procurement of ancillary sendees, they are tempted to ask for strict requirements. Indeed,
strict requirements are better for the security of the system and do not cost anything to the
SO. Lastly, compulsory connecting conditions increase the barrier to entry of new
competitors 101 .
Some solutions alternative to the compulsory connecting conditions have been

developed to give specific signal for long-term supply. For example, in France, some
geographical areas are declared as voltage-sensitive zones. If a reactive power AS provider
supplies its service in one of these areas, it receives an additional remuneration [RTE
(2007b)], which thus gives an incentive to invest in reactive power support. In addition, if a
SO agrees on a long-term bilateral contract with an AS provider (e.g., following a tendering
process), such an agreement gives the AS provider sufficient incentive to invest in new
capacities.
It would also be worthwhile to explore the feasibility and desirability of incentives

and connecting conditions that would reduce the demand for A S . For example, when
intermittent generation such as wind farms represent a significant proportion of the energy
mix, more stringent constraints on the output of intermittent generation could reduce the
amount of frequency control that would be required otherwise.
4.3.2 Short-term matching

An adequate balance between supply and demand of AS in the long-term does not
guarantee that the supply and demand of AS will match in the short-term. To cover this
risk, the ERAP may rely on self-procurement or on long-term contracts. If hedging is not
100 M ost o f the time, the legislator defines the connecting conditions. H owever, it usually follows the
technical advice o f the system operator.
101 See section 4.9.3 for a short discussion on barriers to com petition.
possible, enough incentives should be given to AS providers to offer their capacity in the short
term to make enough AS available (e.g., compulsory minimum participation may be required
for a market with an intrinsic concentration problem).
In addition, an elastic A.S demand102 is essential (which is possible only when the
ERAP is the SO). Indeed, Figure 4.1 illustrates the problem that appears when the supply is
insufficient to meet the demand (e.g., when the system is stressed). This figure also shows
that some responsiveness, even a small amount, can be enough to match supply and
demand, especially when the system is constrained. In practice, demand for AS is usually
very inelastic because the criteria used to define the amount of SS usually ignore the cost
and the value of AS provision. However, all the system operators that buy AS have implicit
non-vertical demand curves because, at some point, they will not accept to pay for
additional AS103, even if their explicit demand is most of the time totally inelastic (as shown
in section 2.4). Anyway, some SOs have an explicit demand curve. For example, the New
York Independent System Operator (NYISO) has a demand curve for secondary frequency
control as shown in Figure 4.2 [NYISO (2008)]. In this case, if the supply is larger than the
ISO’s target, supply and demand will surely meet. However, this demand curve is very
simple and does not reflect the true value of system services. In addition, one could argue
that secondary frequency control reserve above its target is worth something (black-outs
happen also at night, when frequency control reserves tend to be cheaper, as shown in
section 3.5.2). Therefore, this curve could be refined.
supply
supply
demand demand
quantity o f AS quantity of AS
(a) m arket failure (b) market clearing is possible
Figure 4.1: Im p act o f th e d em an d resp on siven ess on m arket clearing
102 A dem and is said to be elastic w hen it is sensitive to the price o f the comm odity. N o te th at there
w ould probably be a cross-elasticity (which represents the link betw een the prices o f two commodities)
betw een active pow er AS and load reduction, supposing that load reduction can be priced through dem and-
side offers. See K irschen and Strbac (2004a) for m ore details about elasticity.
103 T his limitation on dem and is tightly related to the price caps discussed in section 4.7.
4.4 CHOOSING TH E RELEVANT PROCUREMENT METHODS 193
Price
($/iviwm)
300
250
Q u an tity o f se c o n d a ry
fre q u e n c y co n tro l (MW)
target - 25 MW target
F ig u re 4.2: N Y IS O ’s d e m a n d cu rv e for se c o n d a ry freq u en cy c o n tro l. D a ta b a s e d o n N Y IS O (2008)
4 .4 C h oosing th e R elevant P rocurem ent M ethods
The entities responsible for AS procurement can use various methods to obtain ancillary
sendees, namely: (a) compulsory provision, (b) self-procurement, (c) bilateral contracts, (d)
tendering and (e) spot market. Various factors influence the choice of one of these methods
over the others. These factors include risk aversion, market concentration, mode of energy
and transmission rights trading, costs recovery method, centralised or decentralised AS
control (see section 2.3.2), and the type of providers (e.g., loads are likely to prefer bilateral
contracts).
4.4.1 Identified procurem ent m ethods
Compulsory provision refers to the obligation of a certain class of AS actuators connected to

the network (typically large generators) to provide upon request from the TSO up to a
certain amount of a given ancillary service. Compulsory provision is “fair” because all the
users belonging to a certain class must provide the same absolute or relative amount of
ancillary services. However, for the sake of fairness and transparency, the requirements for
compulsory provision are often expressed in a manner that does not catch all the
complexity of the issue. This simplification has two main consequences. First, the volume
of a n c illa ry sendees provided may exceed what is needed, imposing unnecessary costs on
the providers. Second, compulsory provision does not necessarily minimize costs because
potentially low-cost providers are treated on the same basis as more expensive ones. Lastly,
note that this kind of procurement is possible only when the system operator is the unique
ERAP.
Self-procurement makes reference to an ERAP that owns assets susceptible to provide

AS. For example, a TSO can install static voltage compensators in its network to provide
voltage control (see section 1.6.3).
When an ERAP procures an ancillary service using bilateral contracts, it negotiates

with some providers the quantity, quality, price and delivery conditions of the service to be
provided. Such negotiations could remove the two problems associated with compulsory
provision because only the needed amount is procured and the ERAP can deal only with
the cheaper providers. Bilateral contracts for AS are generally traded over-the-counter
(OTC), which leads to forward contracts104. However, bilateral contracts have
disadvantages. First, since their terms are usually not disclosed to third parties, this form of
procurement lacks the transparency that is desirable when one of the parties is a monopoly
(e.g., if the SO is the unique ERAP). Second, bilateral negotiations can be long, complex
and costly. Third, because of the high transaction cost of bilateral contracts, price and
volume are often fixed for a long time, especially for OTC transactions. This will inevitably
be detrimental to one of the parties if market conditions change, which is the case for AS.
Section 3,4.1 shows that the cost of frequency control AS varies a lot over two successive
days. However, bilateral contracts are very important in electricity markets because they
allow participants to hedge against risk105, especially with option contracts, and the cost of
transaction may be offset by this mean. Nevertheless, if too large a part of the market is
based on long-term bilateral contracts, it may be a barrier to the entry o f new participants.
The fourth and fifth procurement methods involve the development of a tendering
process or the creation of a spot market. Drawing a line between these two methods is not
always easy. Here, the term spot market is used to denote a market where standardised
products with a short delivery (i.e., one week or less) are exchanged (see section 1.3.1). A
tendering process involves less standardised products with a longer duration. Both methods
enhance transparency and foster competition. On the down side, they have high data
management costs and may facilitate the manipulation o f the market by some participants.
Furthermore, a pure spot market for AS is unlikely to be implemented because markets
respond too slowly. In other words, the ERAP will never be able to buy AS instantly.
104 Bilateral contracts can also be traded in an organised m arket, w here contracts are standardised and
called future contracts. N o te that a m arket trading short-delivery futures is close to the sp o t m arket (see
section 1.3.1).
i°5 N ote that w hen risk diminishes, investm ents are eased.
4.4 CHOOSING TH E RELEVANT PROCUREMENT METHODS 195
Therefore, an AS spot market is in reality a future market with products with short
deliveries and short durations. AS markets are thus incomplete106, like electrical energy-only
markets [Wilson (2002), Boucher et al (2006)].
Table 4.1 summarizes for each procurement method some of the perceived
advantages (+) and disadvantages (—). This grading is subjective and thus may change from
a market participant to another. Moreover, advantages and disadvantages will be affected by
the duration of the contracts. Table 4.1 shows that no method is superior to all the others.
Furthermore, the importance of each parameter varies across jurisdictions (e.g., a market
designer may give more importance to a procurement method that facilitates entrance of
new participants, whereas another designer may prioritise market transparency). In
particular, a market design relying on bilateral contracts associated with a spot market is not
necessarily the best solution. As various methods are complementary, market designers are
likely to choose a mix of methods, as shown in section 4.4.2.
T ab le 4.1: Param eters in flu en cin g the ch o ice o f A S-procurem ent m eth od
Compulsory Self Bilateral Tendering Spot

provision procurement contracts process market
Mitigate the influence ------
+++ +++ + ----
of dominant players
Facilitate entrance
into the market of + ------ -/+ ++ +++
new AS providers
Hedge against risk ++ ++ +++ + ------
Lower transaction
++ + - - -
costs
Secure enough AS +++ +++ +++ +++ +
Increase the global ------- +
---- ++ +++
welfare
Increase market
+++ ---- — +
transparency
Recognize the ------ - +++ +++ +++
externality of AS
Integrate demand +++
— +++ ++ +
response as an AS
106 A complete m arket is a m arket in which ever}7 contingency can be negotiated. In other w ords, “there
is a forward m arket for ever}7 com m odity at ever}7 date”, as form ulated by Mas-Colell et ai (1995), p. 745.
4.4.2 Procurement m ethods in practice

Table 4.2 summarizes the procurement methods chosen for the different types of ancillary
services in various systems as of October 2006107. As shown in section 2.3.4.1, the services
used for tertiary control not related to the balancing mechanism are numerous and differ
significantly from system to system. Useful comparisons are therefore not possible, so these
ancillary services have not been included in Table 4.2. Primary frequency control is the
ancillary service for which the widest choice of procurement methods exists. A compulsory
provision for this service has the advantage of fulfilling an intrinsically homogeneous
geographical repartition. Secondary frequency control is never compulsory and only France
uses bilateral contracts. Sweden and Great Britain do not use this service, while other
countries rely on more competitive procurement methods. By definition, the basic voltage
control service is always compulsory. Bilateral contracts and tendering are the preferred
trading methods for enhanced voltage control. No spot market for basic or enhanced
voltage control has been put in place yet because these services are very local and therefore
highly susceptible to the exercise of market power.
T able 4.2: P rocurem ent m eth o d s c h o sen across various system s as o f O ctober 2006
Compulsory Bilateral Tendering Spot

provision contracts process market
Primary frequency
ES, PJM AU, FR, N Z DE, GB, N Z, SE AU, N Z
control
Secondary AU, ES,
- FR DE, N Z
frequency control PJM
B asic voltage AU, ES, DE, FR,
FR, NZ GB -
control GB, NZ, PJM, SE
Enhanced voltage
- FR, DE AU, ES, GB -
control
107 See Table 2.3 in section 2.3.4 for basic inform ation on systems and the signification o f the
abbreviations. In addition to personal com m unications and the websites o f the m arket and system operators,
the following references were particularly used: AU: N EM M C O (2001), N EM M C O (2004); D E : E .O N
(2005), B D E W (2007a); ES: Spanish industry and energy departm ent (1998), Spanish industry and energy
departm ent (2000), O M E L (2005); FR: R T E (2004), French econom ic and industry departm ent (2005), R T E
(2007b); GB: N ational G rid (2008b), N ational G rid (2008c); N Z : Electricity Com m ission o f N ew Zealand
(2004), N Z IE R (2005), Electricity C om m ission o f N ew Zealand (2005); PJM: PJM (2005a), PJM (2005b), PJM
(2008); SE: SvK (2005).
4.5 D EFIN IN G TH E STRUCTURES OF OFFERS AND PAYMENTS 197
4 .5 Defining th e S tru ctu res of O ffers and P aym en ts
The structures of bids are related to the scoring and clearing processes, while the structures of
payments are related to the settlement process. A good price structure helps send correct
price signals to the participants. It also ensures that costs are recovered efficiently. As
guidelines, an efficient structure of payment should reflect as closely as possible the various
costs that a provider of sendees may incur, while avoiding double accounting [Cali et al
(2001)], and staying simple enough to be practical. Indeed, if the structure of remuneration
increases in complexity, recovery of cost would be a priori more efficient, but it would also
be more difficult to choose the best offer because the theory of multi-dimensional auctions
is complicated, even for two-part offers [Chao and Wilson (2002)]. As usual, the principal
issue is to find the most appropriate balance. For example, paying separately the meals in a
plane and the cost of the flight may be more efficient than paying an all-in-one flight ticket.
Indeed, budget travellers may prefer to buy a cheaper ticket and to bring their own
sandwich rather than to purchase a more expensive bundle ticket. On the other hand,
separating the costs of the crew, the fuel and the plane does not bring a lot of advantages to
consumers.
4.5.1 Identified structures of offers and payments
Ancillary sendees cannot be remunerated in the same manner as the classical electrical
energy product (i.e., MWh). For example, a generator that is part-loaded in order to provide
frequency control expects to get a payment for the sendee, even if it is not actually called
during the period considered. In the scientific literature, most of the discussions on
structures of offers and payment of AS are concentrated on three components, namely
capacity, utilisation and opportunity cost [e.g., Hao and Papalexopoulos (1997), Singh
(1999), Oren (2001), Chao and Wilson (2002), Chicco and Gross (2004) or Galiana et al.
(2005)]. The literature also proposes innovative components [e.g., Doherty et al. (2005)]. In
addition, some systems like Great Britain have adopted additional components, as shown in
section 4.5.2. This section discusses six identified structures for offers and payments that
are used in practice or have been proposed in the literature, namely: (a) a fixed allowance,
(b) an availability price, (c) an utilisation payment, (d) a payment for the opportunity cost,
(e) an utilisation frequency payment, and (f) a price for kinetic energy.
A fixed allowance is paid to the provider in every instance. For example, a

synchronous condenser is generally remunerated even if it is not synchronised. An
availability price is paid only when the unit is in a “ready-to-provide” state (e.g., a part-loaded
generator that provides secondary frequency control). Therefore, the purpose o f the fixed
allowance is to compensate the provider for the fixed costs (e.g., investment in equipment),
while the availability price compensates for the variable cost to maintain the unit in a
“ready-to-provide” state (e.g., additional staff or energy costs). The separation between
fixed allowance and availability payment may be blurred easily, so the fixed allowance is
often included in die availability payment in the literature. However, the separation may be
useful, especially for AS with a low operating cost. For example, the industry uses fixed
allowance payments for voltage control service (see section 4.5.2). On the other hand, using
bodi a fixed allowance and an availability payment introduces more easily a double
accounting than a single payment does. Furthermore, it is sometimes complex to separate
an availability price for two different ancillary services. For example, it is hard to split the
cost of a generator exciter between active and reactive power AS, as shown in section 3.2.2.
An utilisation payment remunerates the actual delivery of the service. To use economic
language, availability and utilisation payments can be considered as an option fee and a
strike price, respectively. More specifically, an AS with this payment scheme is equivalent to
a swing option because the SO can usually exercise it several times (see section 1.3.1).
However, ancillary services with availability payments are different from standard financial
options, because their goal is not to provide a financial hedge, but to fulfil preset standards
for a “security hedge” [Singh (1999)]. Secondary frequency control is a very good example
of AS with both availability (e.g., in €/M W /h) and utilisation payments (e.g., in €/MWh),
because the TSO can exert or not its option on the available capacity through the regulation
signal sent to the generating unit (see section 2.3.3.1). An utilisation payment should
consider the amplitude of delivery of the service. For example, an utilisation spike has often
a stronger impact on the provider than a continuous utilisation does. Figure 4.3 shows two
different utilisations, which are however usually entitled to the same utilisation payment
because both energy integrals are equal.
4.5 D EFININ G THE STRUCTURES OF OFFERS AND PAYMENTS 199
output
(a)
output
F ig u re 4.3: S am e u tilisa tio n p a y m e n ts for m o d iffe re n t u tilisa tio n s
In addition to availability and utilisation payments, the payment of the opportunity cost
has been identified for a long time by the community as an important allowance [Hirst and
Kirby (1997)]. As presented in section 3.2.2, the opportunity cost represents the profit that
the provider would have made if it had sold other products (e.g., energy) instead of
supplying the service. Note that to compute the opportunity cost, the supply curves of units
have to be provided, which is not popular in markets with decentralised unit commitment.
In addition, it is usually assumed that the energy’ price would not change if the unit were
economically dispatched without providing any AS. This assumption is a good
approximation only when the liquidity of the market is sufficient.
An utilisation frequency payment is based on the number of calls to provide a sendee

over a given period of time. It thus reflects the extra costs that may be incurred each time
the sendee is called upon. For example, an event fee is used in New Zealand (in
NZ$/event). In case of a large disturbance, the user that triggers the disturbance pays an
event fee, w7hich is thus a penalty (see section 4.8.5). This fee is then split amongst the
participants that need to be rewarded [Electricity Commission of New^ Zealand (2005)].
A price for kinetic energy (e.g., in €/MWs) remunerates the quantity of kinetic energy’
made available to the system [Doherty et al. (2005)]. This innovative approach recognizes
that machines wdth high inertia (and thus with a high kinetic energy) reduce the rate of
change of frequency'"8 (but still cannot stabilize it durably). In fact, a power system must
have a maximum rate of change of frequency (in Hz/s) following a contingency, so the
participants reducing this rate of change of frequency should be remunerated. However,
108 H igh inertia is characterised by a large / and thus a large d~wm/dr in equation (1.2) o f section 1.4.1.
although interesting, this approach is probably practical only in small interconnected

systems that are much more sensitive to the stored kinetic energy (e.g., Ireland) than large
interconnected systems are (e.g., the UCTE).
4.5.2 Structures of offers and paym ents in practice

In practice, only a few payment components are actually chosen, depending on the ancillary
service considered, because the costs incurred are different (see Chapter 3). The timing of
markets (see section 4.6.6) and the information available also enter into account (e.g., the
opportunity cost is difficult to compute in a market with decentralised unit commitment).
Table 4.3 shows the structures of payment chosen across various systems as of October
2006 (see 107 p. 195). For primary frequency control, an availability payment is in most cases
the only type of remuneration because it operates continuously and the additional amount
of energy it requires is considered negligible. The opportunity cost is not considered either,
whereas the primary frequency control may cause such a cost. In three of the six systems
that use secondary frequency control, both availability and utilization are remunerated.
However, energy savings obtained when the output o f a generator is reduced are subtracted
from the additional energy used when this generator increases its output (see section 4.5.3).
Fixed and availability remunerations are favoured for basic and enhanced voltage
controls. Except in Great Britain, New Zealand and Spain, the actual use of reactive power
through the voltage control is not remunerated, so the utilisation cost of reactive power
does not seems to be considered as significant. Voltage control is not remunerated at all in
Sweden, The active power used when generators work in synchronous compensator mode
is remunerated in Australia, France, PJM and Spain. Opportunity costs are not widely used,
probably because they are more difficult to compute and they may be included in availability
payments when a long period is considered (e.g., a week or more).
The structure of the offers is often identical to the structure of the payments, but
some differences can appear. First, a payment component may be indexed on another
market instead of being explicitly defined in the bidding process. For example, the
utilisation payment of an active power AS can be based on the energy-only spot market.
Second, the structure of an offer may be more complex than the structure of a payment.
For example, it can be desirable to have different offer prices for different conditions of
utilisation of an AS provider. However, the price of payment is unique for this provider. In
4.5 D EFIN IN G TH E STRUCTURES OF OFFERS AND PAYMENTS 201
practice, providers often provide an offer curve. For example, Australian generators can
offer up to ten frequency control AS price bands depending on the output of the given
generator (see Figure 4.4) [NEMMCO (2001)]. In Great Britain, an equivalent system is
used for reactive power AS, where up to six price bands are possible (see
Figure 4.5) [National Grid (2008c)].
T a b le 4.3: S tru c tu re s o f p a y m e n t c h o se n acro ss v ario u s sy ste m s as o f O c to b e r 2006
Utilisation Opportunity
Fixed Availability Utilisation
frequency cost
Primary frequency AU, DE, FR,
GB GB GB, NZ -
control GB, NZ, SE
Secondary AU, DE, FR,
- DE, ES, FR - PJM
frequency control PJM, ES
Basic voltage FR, GB, NZ,
FR, GB, NZ GB, NZ - PJM
control PJM
Enhanced voltage AU, ES, FR,
FR, GB ES , GB - AU, DE
control GB
R esp o n se (MW)
R esponse max
(MW) x
Band 10 (MW) P r lc e T ^ c i Ho; {W .b^VV/B):
Dispatched
output (MW)
Band 1 (MW) F jPrice ;band A Âpp/MW/h]
Enablement min Low breakpoint High breakpoint Enablement max

(MW) (MW) (MW) (MW)
F ig u re 4.4: R e p re se n ta tio n o f th e g e n e ric fre q u e n c y c o n tro l an c illa ry service tra p e z iu m u s e d in

A u stralia. B a se d o n N E M M C O (2001)
Obligatory Reactive
Active Power
Obligatory Reactive
Power Service Power Service
(Grid Code) - 0.95p1 (Grid Code) - 0 85pf
Leading
Full Output
Rated MW
T01 TQ2 TQ3

L ead TQ3 TQ2 TQ1 Lag Reactive
Power
Figure 4.5: R eactive pow er capability curve in Great Britain [N ation al Grid (2008c)]
4.5.3 Price sign and symmetry
Beyond the number of parameters to consider in the price structure, another interesting
issue is the price sign (i.e., positive or negative price) and the price symmetry (i.e., down and up
sendees with same prices). For example, if a unit provides secondary frequency control,
shall it pay or be remunerated for reducing its output when the regulating signal requires it?
Indeed, the SO may argue that in such cases this unit saves money by not producing power
(so the unit should pay back), whereas the producer may argue that this reduction helps
maintain the system in a secure state (so the unit should be paid). In the case of reactive
power, the question is even harder, since both production and absorption can be paid or
charged, depending on the state of the network (i.e., whether the voltage is higher or lower
than desired). To achieve maximum flexibility in offers and hence enhance competition, it is
probably best to avoid placing any constraint on the sign of the price or requiring pricing
symmetry.
4 .6 Organizing th e Market Clearing P rocedure
The market clearing process matches supply and demand by setting the prices and the
amounts traded. Self-scheduled AS providers are not considered here, since they do not
participate directly in the price discovery process resulting from the market clearing.
4.6 ORGANIZING THE MARKET CLEARING PROCEDURE 203
This section discusses the main features that a clearing procedure must have: (a) the
structure of the market can be either centralised or decentralised (section 4.6.1); (b) various types
of auctions are possible. Auctions define how the offers must be made (section 4.6.2); (c)
scoring, i.e. ranking offers in a way that encourages truthful bidding and an efficient choice
(section 4.6.3); (d) because the various AS are correlated with each other and with energy,
coordinating the various markets is necessary (section 4.6.4); (e) there are still differences of
opinion about the best settlement rule, in particular on the choice between discriminatory and
uniform prices across AS products, and between the zonal and nodal approaches
(section 4.6.5); lastly, (f) timing of the market events such as clearing, closure of demand and
offers, as well as duration of contracts have to be chosen carefully (section 4.6.6).
4.6.1 Structural arrangem ent
With a centralised unit commitment, the market (usually called "pool”) is cleared by a unique
entity. In pool energy-only markets, which are popular across the United States, participants
provide bids and offers. Then, the system operator directly commits and dispatches the
producers. Therefore, this approach favours links between AS and other products such as
energy. For example, the opportunity cost can be calculated easily. However, markets with a
centralised unit commitment are deemed to be opaque because the clearing process is quite
complex. In addition, bidders have to provide a lot of information and this system hardly
takes into account all the variables of the system.
On the other hand, a decentralised unit commitment may be adopted. In this system,
which is popular across Europe, participants propose bids and select offers directly in the
market. Therefore, a global co-optimisation is difficult since participants buy and sell
independently from each other. Instead, each participant does its own co-optimisation with
its assets. Therefore, the dispatching is a priori less efficient, but participants in the market
are able to take more parameters into account. Note that the decentralised arrangement
described here can use centralised markets (usually called “exchanges”) as described in
section 1.3.1.
To probe further, CER (2003) discusses in more details advantages and drawbacks
of pool and power-exchange markets.
4.6.2 Types of auction
By setting the rules for submitting and matching bids and offers, an auction helps determine
the value of a commodity that has an undetermined or variable price. More precisely, in the
case of ancillary services, the auctioneer wants to reveal the marginal cost of AS providers
for each cost component. In economics, various auctions are available, depending on the
problem considered. Section 4.6.2.1 discusses the general categories o f auctions, while
section 4.6.2.2 introduces the usual auctions and section 4.6.2.3 examines the most
appropriate arrangement for AS procurement.
4 .6 .2 .1 P rin cip a l c a t e g o r i e s o f a u c t io n s
First, auctions can be private-valued or common-valued. In a private-value auction, the good

(e.g., a painting) has a different value for every participant. In a common-value auction, the
good (e.g., off-shore drilling rights to extract oil) has an identical value for all the
participants, even if the estimate of each participant may vary. In this latter type of auctions,
the optimal strategy for participants is to bid less than their estimate value in order to avoid
an overvaluation that would lead to a financial loss. The difference between the amount that
the winner paid and the next lower bid of an auctioned good is often referred as the
Winner’s curse [Varian (1999)]. Auctions for ancillary services can be either private-valued
or common-valued depending on the size of the market considered. Indeed, the value of
frequency control services would be higher in a part of the system with high AS
requirements than in a less demanding part. However, a perimeter can be found where AS
have a common value.
Second, an auction can be single-sided or double-sided. In the former type, only one
buyer or one seller wants to acquire or sell the good. In the latter type of auction, several
buyers and sellers participate in the auction. As highlighted in section 4.2, the AS buyer(s)
for a given zone can be either unique (e.g., the SO) or several (e.g., users with AS
requirements). Therefore, single-sided and double-sided auctions are both possible for AS.
Even if solutions based on the unique buyer are popular, SOs may compete against each
other to buy AS across interconnected systems, as discussed in section 2.5. Therefore, the
procurement of ancillary services is likely to move towards a double-sided auction in the
long-term.
Third, single-unit and multiple-unit auctions are possible. Single-unit auctions deal
solely with identical goods (e.g., buying 1 000 iPods). This type of auctions is the simplest
and benefits from a lot of analysis [Varian (1999)]. On the other hand, differentiate goods
are traded in multiple-unit auctions (e.g., buying 100 second-hand cars). Unfortunately,
properties of single-unit auctions cannot be extended to multiple-unit auctions [Staropoli
and Jullien (2006)]. Therefore, multiple-unit auctions are much harder to deal with and no
hasty conclusion should be drawn when dealing with such auctions. AS auctions can be
either single- or multiple-unit, depending on the differentiation possible between AS
providers (see section 2,3).
Lastly, auctions can also be categorised by their settlement rule. In particular,

discriminatory auction prices and uniform prices are two important settlement rules that
differentiate auctions. Section 4.6.5 discusses in details settlement rules.
4.6.2.2 Usual auction m ethods
Commenting on auction design, Klemperer (2002) states that “one size does not fit all”,
because an auction helps to answer to a specific need. For example, an auction that is used
to trade private-valued multiple-unit goods is not necessarily efficient for common-valued
single-unit goods. Iti practice, some types of auction are often used. The usual types of
auction presented in this section are English, Dutch, sealed-bid first-price and Vickrey.
From these usual auctions, different variants are possible.
In an English auction, participants increase their price gradually from a minimum

price, called the reserve price. The highest-price bidder takes the good. English auctions can
be used for both single-unit and multiple-unit auctions, and are Pareto efficient as soon as
the reserve price is below the clearing price. The Japanese auction is a variant of the English
auction. In such a system, participants are able to enter or to leave the auction at any time,
in a similar way as in a poker game. For a single-sided AS auction (e.g., the ERAP is
unique), an English auction would not be practical, since this type of auction is designed for
several buyers. However, this auction may be of interest in the case of a double-sided
auction. For example, an AS provider would propose a given quality of sendee with a
reserve price. The different ERAPs would then raise their biddings up to no more than the
value that they attach to this sendee.
In a Dutch auction, a price is given by the auctioneer and then the price is decreased
by steps until a buyer is interested. Dutch auctions have the advantage of being very fast109.
An AS auction based on a modified Dutch auction may be imagined as follows. The ERAP
would propose to buy a given quality of ancillary service from AS providers. The ERAP
would start with the highest price that it would be ready to pay for and then AS providers
would manifest their interest. As long there would be an AS provider interested, the price of
the contract would decrease by step. The last AS provider to accept would win the AS
contract. However, such a system may not be Pareto efficient because it is not guaranteed
that the latest accepted price be the lowest price that the AS provider was ready to accept.
In a sealed-bid auction (also called Yankee auction or sealed-bid first-price auction), all
the participants make a bid (or an offer) at the same time. The best bid takes the good and
pays as bid110. This type of auction is commonly used in tendering processes in the
construction in industry. Lastly, this auction is applicable to both multiple-unit and single
unit procurements, but it is often not Pareto efficient. In fact, “no buyer knows other
buyers’ valuations, so the highest valuation buyer may bid too low and lose to another
bidder” [Bergstrom (1999)]. A system for ancillary services is easily imagined in which the
ERAP would call to tender the AS providers. The AS providers would submit their offers,
then the ERAP would select the best ones.
Lastly, a Vickrej auction (also called philatelist auction or sealed-bid second-price

auction) is similar to a sealed-bid auction except that the winner pays the second highest
bid. These auctions, which theoretically lead to the same result as English auctions but in a
one-round auction [Varian (1999)], are related to single-unit auctions. In the case of
multiple-unit auctions, the Vickrey-Clarke-Groves (VCG) auction may be used instead. In a
VCG auction, each bidder divulges the value of each combination of items (in the particular
case of products of the same quality, the bid is a demand curve). The bids are then selected
by the auctioneer in order to maximize the global value. Then, the winner of each bid pays
the value of its bid minus the added value o f its bid. The added value of a bid is calculated
by subtracting the maximized global value without the bidder from the actual maximized
global value [Pekec and Rothkopf (2004)].
109 T he D utch auction holds its nam e from the D u tch m arkets w here this auction is used to trade
cheese and flowers [Varian (1999)].
1!0 A variant o f this auction is possible with a uniform pricing, e.g., based o n the last accepted offer
o r die first rejected offer.
An illustrative example of a VCG auction is given in Table 4.4. In this example, by

granting the product {a} to bidder 1 and the product {b} to bidder 2, the global value is
maximised (30€). Table 4.5 gives a second example, where bidder 1 evaluates highly
product {a} at 1 000 €. However, the product allocation and bidder l 5s payment are the
same as in example 1. On the other hand, bidder 2 will receive money in this case, whereas
the seller is unlikely to pay for selling its good111. In practice, such a case can happen if the
two bidders collude and share the benefits of this strategy. This simple example thus shows
the main flaw of VCG auctions: they are revenue deficient. Therefore, VCG auctions are
rarely used in practice [Pekec and Rothkopf (2004)]. Furthermore, Rothkopf (2007) lists
thirteen reasons why VCG auctions are not practical. Note however that the VCG auction
differs from the Groves-Clarke tax described in section 2.4.4 m .
T ab le 4.4: E xam p le 1 o f a Vickrey-Clarke-Groves au ction
Bids Settlement
{a} {b} {a,b} Gets Pays
Bidder 1 12 € 6€ 20 € {a } 12-(30-25) = 7 €
Bidder 2 3€ 18 € 25 € {b} 18-(30-20) = 8 €
T ab le 4.5: ExampLe 2 o f a Vickrey-Clarke-Groves auction
Bids Settlement
{a} {b} {a,b} Gets Pays
Bidder 1 1 000 € 6€ 20 € {a } 1 000-(l 018-25) = 7 €
Bidder 2 3€ 18 € 25 € {b} 18-(1 018-20) = -980 €
4 . 6 . 2 . 3 S e l e c t i n g t h e a p p r o p r ia te a u c tio n m e th o d
Table 4.6 summarizes some features of the described auction methods. English and Dutch
auctions are multiple-round auctions, while sealed-bid first-price and Vickrey are single
round auctions. All auctions are Pareto efficient under particular conditions, but only the
Vickrey auction gives incentives to buyers to reveal their true value for a good, even in an
111 N o te th at sellers are som etimes ready to pay to get rid o f their goods, especially w hen the goods
cannot be stored anymore (e.g., electricity o r perishable products).
112 F o r example, a V C G auction helps determine the price o f each product, whereas the cost
allocation is pre-determ ined in the Clarke-Groves tax system.
imperfectly competitive market [Chao and Wilson (2002)]. However, all auctions are
vulnerable to collusion or strategic behaviour [Varian (1999)]113.
The desirable features for an AS auction are highlighted in bold. In fact, AS auctions
are multiple-unit, as explained in section 4.6.2.1. In addition, it is desirable to have an
auction both Pareto- and revenue-efficient and that helps to reveal the true value of the
good. Lasdy, single-round auctions are easier to manage in real-time than multiple-round
auctions, so they should be preferred. It is clear from Table 4.6 that none of the proposed
auction methods fulfils all the requirements. In practice, sealed-bid auctions (first-price or
uniform-price) are very popular across systems to procure ancillary services, because they
are convenient for the procurement methods preferred by participants, i.e. tendering
processes and spot markets (see section 4.4). However, despite their interesting features, the
Vickrey and VCG auctions do not seem to be used because they are unpractical.
Nevertheless, Chao and Wilson (2002) propose a market design for spinning reserves based
on Vickrey and VCG auctions.
In conclusion, sealed-bid first-price auctions are probably the best trade-off so far.
However, stakeholders have to keep in mind that such auctions do not guarantee the Pareto
efficiency and do not reveal the true value of AS, so they are likely to send incorrect signals
to the market. The possibility of non-efficient prices is even higher because markets for
ancillary services usually involve an oligopoly and repetitive auctions. Participants have thus
time to understand the market and the gaming possibilities [Staropoli and Jullien (2006)].
T ab le 4.6: S ynopsis o f the u su al auction m eth od s
S in g le / H elp to S in g le /
A uction S in g le /D o u b le Pareto R evenue
M ultiple- reveal true m ultiple-
m ethod sid e efficient efficien t
un it value round
English Both Both Yes Yes No Multiple
Dutch Both Both Y es114 Yes No Multiple
Sealed-bid
Both Both Y es114 Yes No Single
first-price
Vickrey Single Both Yes Yes Yes Single
VCG M ultiple Both Yes No Yes Single
113 T o probe further into auction theory, Varian (1999) gives a com prehensive introduction, while
K lem perer (1999), K lem perer (2002), and Pekec and R o th k o p f (2004) discuss auctions in m uch m ore details.
Lasdy, Staropoli and Jullien (2006) show the interest o f laboratory experim ents, w hich help m arket designers
to study and assess the perform ance o f m arket rules in an environm ent simpler than reality.
114 Pareto efficiency can actually be hard to reach in practice.

4.6.3 The scoring problem

To select an offer over another, both have to be classified, e.g. by giving them a score115
(e.g., expressed in monetary value). This score should be calculated in a way that induces
truthful bidding by participants [Bushnell and Oren (1994)] and, obviously, that leads to the
appropriate result (i.e., the scoring should lead to the optimum solution according to the
selected objective function116). The barriers to appropriately score ancillary services are
namely: (a) the different technical qualities of products because it makes the products
harder to compare (see Chapter 2); (b) the composite structures of offer and payment,
which make offers dependent of the utilisation and also make offers harder to compare (see
section 4.5); and (c) the strongly variable non-linear costs of ancillary sendees as function of
time and demand, which makes the scoring of offers sensitive to time and demand (see
Chapter 3). As usual, a trade-off has thus to be found between an overly simple solution,
which can be totally inefficient [Wilson (2002)], and a non-practical mechanism.
To score offers while trying to minimity its procurement cost, the ERAP must take into
account both the price and the probability of AS utilisation. This presents two difficulties.
First, the optimisation problem introduces AS prices (i.e., the dual variables) into the primal
problem, which makes the optimisation problem much harder to solve [O’Neill et al
(2006)]. Though, Lull et al (2006) have proposed recently an optimisation algorithm that
overcomes this difficulty by updating the Lagrangian multiplier in an iterative process, i.e. in
a similar way as die price decomposition117. Second, taking into account the probability of
AS utilisation may favour gaming. For example, some units may lower their utilisation
offers and increase their availability payments to increase their profits [Singh (1999)]. To
avoid such gaming, Chao and Wilson (2002) have proposed a new scoring method for
active power AS. The idea is to force providers to reveal their true energy marginal cost
through a Vickrey auction (see section 4.6.2.2). For the settlement, the opportunity cost of
die various energy offers constitute die availability payment, and energy is paid at the lowest
unused energy offer.
115 N o te that the scoring should also include a tie-breaking feature th at separates two bids with the
same score. D efining this rule is a relatively easy task if bids can be split.
116 As explained in section 2.2.3, the ERA P can either minimize the procurem ent cost (m ethod also
know n as the rational buyer approach) or maximize the global welfare. B oth objective functions are dependent
on the settlem ent m ethod chosen. Various settlem ent m ethods are described in section 4.6.5.
117 See section 3.3.5.1 for an introduction to the primal and dual problem s and to the price
decom position m ethod.
The main down side of scoring AS in the perspective of minimizing the

procurement cost is to ignore the value that AS (and thus SS) have for system users. If the
perspective of global welfare maximisation is adopted, the ERAP has to take into account the
value of ancillary services in addition to their procurement cost. However, including the
value in the scoring is complex, as shown in section 2.2.1. Therefore, most of the practical
methods are limited to the procurement cost minimization.
4.6.4 Coordination of the different markets

Once a type of auction and a scoring method have been defined, it is possible to collect the
AS offers, rank them and then to select the best offers. However, AS are an externality of
the electrical energy-only market. For example, if a generator provides reserves for
frequency control, it cannot sell all its capacity on the electrical energy-only market. On the
other hand, if a generator is committed through energy dispatch, it is then able to provide
reactive power support. A direct consequence of this feature is that AS prices and energy
prices will interact. This statement can even go further concerning transmission: financial
transmission rights and AS prices are likely to interact because active power ancillary
sendees have similar geographic constraints as energy products do118. In fact, the ERAP
may want to secure enough transmission capacity in order to have an efficient AS
provisioning. More specifically, AS are a production externality of the energy-only market
because the volume or the price of AS will not change the amount o f electrical energy
bought on the market [Varian (1999)].
Since AS are an externality of other products such as energy and because several AS
products exist (see section 2.3), AS auctions have to be coordinated. In independent merit-order
auctions, each ancillary service auction is cleared independently. However, because an AS
provider can generally provide several AS products119, this approach may lead to infeasible
solutions. Therefore, sequential auctions (also called cascading clearing) have been defended in
the early designs of market for frequency control AS [Hirst and Kirby (1997), Singh (1999)].
With this method of coordination, AS are provided in a prioritised manner. Once an AS is
bought, the capacity of the successful provider of this service for the following AS auction
118 H owever, because o f their local effect, reactive pow er AS have stronger geographical constraints
than active pow er AS do. See section 1.5 for m ore details.
m See for example Baldick et al (2005) about p ro d u ct substitution.

4.6 ORGANIZING TH E MARKET CLEARING PROCEDURE 211
is reduced. However, this approach suffers from some flaws, in particular price reversals,
illustrated by die infamous failure of the Californian AS market at the end of the
1990s [Brien (1999), Oren (2001)]120. Since then, it has been recognised that simultaneous
auctions have to be in place for AS and energy-only markets [e.g., Yan and Stern (2002), Ma
et al (2003), Baldick et al (2005) or Galiana et al (2005)].
To illustrate the flaws of independent and sequential clearing, consider a simple

example. Two ancillary sendees called AS 1 and AS 2 are considered. The SO is the only
ERAP and has to buy 100 units of each to meet its reliability criterion through a classical
sealed-bid auction, as illustrated in Table 4.7 (i.e., inelastic demand). Two AS providers
propose their capacities as shown in Table 4.8. In order to select the providers, the ERAP
performs three auctions. The first auction is based on an independent merit-order, the
second auction is sequential121 (starting with AS 1) and the last auction is simultaneous. In
all auctions, the objective function of the ERAP is to minimize the procurement cost,
subject to the constraint of meeting the demand (i.e., the rational buyer behaviour). The
results of the three auctions are given in Table 4.9. It is clear that the simultaneous clearing
gives the best results. Indeed, the sequential market clearing, although giving a feasible
solution, depends on the clearing order. An optimal solution is thus not guaranteed.
However, simultaneous auctions between AS and energy-only markets raise some

practical difficulties. First, the feasibility of AS simultaneous auctions depends on the
choices for the electrical energy-only market design. In particular, the choice between a pool
and a power exchange market is very important122. In fact, simultaneous auctions are easier
to implement in pool systems than in exchanges, since a unique entity controls the whole
system. Second, from a technical point of view, the optimisation problem has to be feasible.
Simultaneous auctions require a large amount of data, which may be too large for an
efficient clearing. For example, finding the optimum between AS offers and the active and
reactive power losses in real-time is computationally demanding. Third, simultaneous
120 A price reversal is said to appear w hen an ancillary sendee o f lower quality is m ore expensive than
another o f higher quality. F urther in this section, the example o f Table 4.9 illustrates such a price reversal. As a
practical example, during the Californian crisis, the price o f non-spinning reserve was m ore expensive than the
price o f spinning reserve, whereas spinning reserve should be m ore expensive to provide than non-spinning
(see section 2.3.4.1 for the vocabulary o f reserves).
121 W e consider here a supply (or product) substitution, i.e. the supply for a lower quality AS can be
substituted by a higher quality AS. In die case o f a dem and substitution, the dem and for a lower quality AS
can be substituted by a higher quality AS [Oren (2001)].
122 See section 4.6.1 for a short discussion between pools and exchanges.
auctions are much less transparent than independent or sequential auctions, whereas
participants usually prefer transparent processes. Fourth, it could be argued that it would be
better to co-optimize the products over long periods of time to obtain better results.
Indeed, transmission and generation capacity investments have an impact of the cost of
ancillary services. However, such a long-term optimization is likely to be difficult to
implement in a liberalized environment.
In practice, short-term simultaneous auctions are performed in PJM (secondary

frequency control, synchronised reserves and energy together) and Australia (primary
frequency control, secondary frequency control and energy together). The other surveyed
systems have adopted independent or sequential auctions and each participant co-optimizes
individually its portfolio.
T ab le 4.7: Param eters o f the d em an d in the consid ered exam p les o f m arket coordination
Product Volum e
AS 1 100 u
AS 2 100 u
Total 200 u
T able 4.8: Param eters o f the offers in the con sid ered exam p les o f m arket coordination
Bidder Capacity Product Volum e Price

Bidder 1 100 u AS 1 100 u 5 € /u
AS 2 100 u 15 € /u
Bidder 2 100 u AS 1 100 u 100 € /u
AS 2 100 u 300 € /u
T ab le 4.9: R esu lt o f the clearing p rocesses in the exam p les o f m arket coordination
Auction Product Provider Volum e Price Cost

Independent AS 1 Bidder 1 100 u 5 € /u 500 €
AS 2 Bidder 1 100 u 15 € /u 1 500 €
Total - Infeasible solution
Sequential AS 1 Bidder 1 100 u 5 € /u 500 €
AS 2 Bidder 2 100 u 300 € /u 30 000 €
Total - 200 u 152.5 € / u 30 500 €
Simultaneous AS 1 Bidder 2 100 u 100 € /u 10 000 €
AS 2 Bidder 1 100 u 15 € /u 1 500 €
Total - 200 u 57.5 € / u 11 500 €
4.6 ORGANIZING TH E MARKET CLEARING PROCEDURE 213
4.6.5 The settlem ent rule

The settlement rule organises die payment of the AS providers selected in the auction. Such
a rule should give enough incentives to invest in AS capacities and to offer AS at their
marginal costs if there is enough competition. A settlement can be either ex-ante (i.e., the
price is discovered before the actual operating of the system) or ex-post (i.e., the price is
discovered after the actual operation of the system). Ancillary sendees can be either non-
remunerated, or paid according to one of three types of price: an administrative price, a
discriminatory auction price or a uniform price.
While a non-remnnerated system is very convenient for the ERAP, it is unlikely to be

economically optimal because the costs that the providers incur end up bundled in the price
of other products such as electrical energy. In addition, a non-remunerated policy is most of
the time contrary to the concept of property rights, which is important for AS since they are
an externality of the electrical energy-only market (see section 4.1).
An administrative price is set by the regulator or the SO and can be either nodal or
zonal. This form of remuneration, also called regulated pricing, can take several forms: a
reimbursement (i.e., an ex-post payment that compensates the AS providers for the cost of
sendees), a value-based payment (i.e., a payment based on the estimated value of AS) or a
tariff (i.e., an administrative rule of payment) [Cali et al. (2001)]. In practice, since the value
of AS may be hard to calculate and tends to be high or very high (see section 2.2.1),
reimbursements and tariff payment are preferred in practice. An administrative price is
particularly justified when dominance of some participants is an issue. In general, however,
a regulated price is not desirable as it reflects very imperfectly the actual cost of providing
an ancillary sendee, particularly when this cost changes with time or circumstances, as it is
the case of frequency control ancillary services (see section 3.4.1).
With discriminatory auction prices, winners of the auction pay/receive different prices
for the same good. Discriminatory auction prices are suitable when the quality of the
ancillary services offered is highly differentiated, so the offers are not easily comparable.
However, a discriminatory auction price does not give providers an incentive to offer their
marginal cost, except when market concentration is low [Stoft (2002)]. In the particular pay-
as-bid discriminatory system, the supplier receives the price of its accepted offer. Variants of
tire pay-as-bid system are possible, as proposed by Bushnell and Oren (1994). However,
Staropoli and Jullien (2006) argue that discriminatory auctions generally perform poorly in
electric power markets. Nevertheless, numerous systems have actually adopted

discriminatory auction prices, as shown in Table 4.10.
In a uniform price system, all the winners of the auction pay/receive the same price.
This form of pricing, defended for example by Chao and Wilson (2002) for frequency
control AS, is deemed to give suppliers a real incentive to offer their marginal cost. O n the
other hand, it is not adapted to differentiated products because all the offers have to be
comparable. Various settlement rules are possible in the case of a uniform price system and
are described below [Oren (2001), Shahidehpour et al. (2002)]. We consider here only the
settlement rules in case of a simultaneous auction because it has been shown in section 4.6.4
that sequential auctions do not guarantee efficiency, though being simpler than
simultaneous auctions.
Link between uniform prices of different AS of same nature (e.g., all frequency control
ancillary services):
■ Common price between AS of same nature: this approach, defended by Galiana et
al. (2005), is based on the fact that all the active power AS have an equal role in the
security. In addition, the philosophy of common pricing between AS can be
justified by the fact that some AS cost components may be common (e.g., an energy
payment);
" Separate prices between AS of same nature: the classical approach, in which each
AS has its own market clearing price.
Z Methods to calculate the market clearing price:
■ Marginal value pricing: the price is equal to the most expensive of all the accepted
AS offers of equal or lower quality;
■ Last accepted bid: the price is equal to the highest accepted bid for the AS
considered;
■ First rejected bid: the price is equal to the lowest rejected bid for the AS considered.
Note that Bernard et al. (1998) and Staropoli and Jullien (2006) report that the last
accepted offer auction performs slightly better in energy markets than the first
rejected offer mechanism does. However, this conclusion cannot be directly
extended to AS.
V Perimeter of the uniform pricing:
■ Zonal prices: they may be preferred for frequency control AS because the value of
frequency control AS is usually not geographically dependent (see section 1.4).
However, defining the zone remains a problem because congestions may prevent
the supply of the AS. In practice, the area of the TSO is often considered as the
price zone [O’Neill et al. (2006)];
■ Nodal prices: they may be favoured for voltage control sendees because their effect,
and so their value, Is local (see section 1.5). Some system operators, such as in
Texas, argue that nodal payment of AS gives an incentive for relieving congestion
similar to energy [Mickey (2006)], In addition, Galiana et al. (2005) argue that it
gives more signals to compute the marginal value of security in a nodal fashion than
in a zonal manner. On the other hand, nodal pricing is expensive to implement,
increase the complexity of bilateral contracts123 and may increase the market power
of some participants. More broadly, discussions on the relative advantages of nodal
and zonal pricing are still ongoing in many countries [Sioshansi and Pfaffenberger
(2006)].
Table 4.10 summarises the settlement rules chosen across various systems as of
October 2006 (see 107 p.195). PJM and Spain are the only systems where primary frequency
control is not remunerated, so these systems do not recognize the property rights of
generating units over their governor response. When this sendee is remunerated, a
discriminatory auction price is preferred because primary frequency control is a
differentiated product. Secondary frequency control services are remunerated in all the
systems that use it. Since secondary frequency control is managed directly by the TSO and
uses energy, it is obviously a service that providers deliver to the TSO and should therefore
be remunerated. For basic voltage control there is as yet no consensus on whether this
sendee should or should not be remunerated. Finally, discriminatory auction price is the
preferred option for enhanced voltage control where this sendee is defined.
123 Bilateral contracting is eased in energy-only market with the help o f “hub prices” that reflect the
average price in a given zone [O’Neill et al (2006)].
T able 4.10: Settlem ent rules ch o sen across various sy stem s as o f O ctober 2006
Regulated Discrim inatory T,

N one . J Uniform price
price auction price
Primary frequency AU, FR, D E, GB,
PJM, ES NZ
control NZ, SE
Secondary frequency
- - PJM g £ n J s! D E ’ A U .E S. PJM
control
AU, DE,
Basic voltage control GB, PJM FR, GB, N Z
ES, SE
Enhanced voltage
- ES FR, AU, DE, GB
control
4.6.6 The timing of markets

A market can be viewed as a game in which all the players try to maximize their benefits
while respecting the rules. The timing of the various events (e.g., clearing, closure or review
of the AS needs) is an important factor of the game. Indeed, the market events give
information to the players that will then adapt their strategies in real-time until an
equilibrium region is attained (see section 1.3.1). In addition, the timing impacts practical
constraints. For example, scoring offers every 5 min is a different challenge from doing it
on a day-ahead basis (see section 4.6.3).
One of the most obvious parameters in the market timing is the frequency of market
clearing. The frequency of market clearing can vary from several years to a continuous
quotation. The more often the market is cleared, the closer the market should be to the
actual cost of supply and value of demand, the better are the signals given to the
participants, and the more efficient the market should be. However, the feasibility and
practicality of the market decreases when the frequency of clearing increases. It is not the
purpose here to describe in details all the associated engineering issues (e.g., computing
capacity, means of communication, data format, speed of transmission), but, as usual, a
trade-off has to be found between the operational requirements and the possible economic
benefits. In general, the lower the frequency of clearing is, the easier it is to use complex
scoring methods. In addition, a high frequency market clearing helps players understand the
behaviour of the market, and therefore to game it, as happened in California, where some
capacities were withheld in order to raise prices [Brien (1999)]. Therefore, a low frequency
market clearing has the advantage of reducing somewhat the influence o f dominant players.
Table 4.11 gives the frequencies of market clearing across various systems as of
October 2006 (see 107 p.195). In practice, the active power AS can benefit from high
frequency market clearing (e.g., every five minutes in Australia). However, even if real-time
reactive power markets have been proposed, for example by Zhong et al. (2004), none has
been implemented yet. In fact, the variable cost of reactive power is low [Hao and
Papalexopoulos (1997)], so benefits from a high frequency clearing are likely to be low as
well. In addition, reactive power is local, so gaming is increased, as illustrated by the simple
example of Figure 4.6. By increasing the reactive power absorption of generator B, the value
of reactive power production is increased as well. Therefore, load L is giving more money
to generator A, even if load L’s consumption has not changed. In conclusion, with a simple
collusion between generators A and B, it is easy for them to make money from L.
T ab le 4.11: F req u en cies o f m arket clearing across various system s as o f O ctober 2006
AU DE ES FR GB PJM NZ SE
Primary Positive: Ever)7

Every 5 Every six Every two or 30 min Every week
frequency - three years
Every month - Negative: every and ever)’ hour
minutes months
control year
Secondary
Every 5 Every six Ever)7two or
frequency minutes months
Ever)7day
three years - Ever)’ hour Ever)7year
control
B asic
Ever)7two or Every six
voltage - - - tliree years months
Every year No rec.
control
E n h an ced
Ever)7two Ever)7two or Every six
voltage years
No rec. Ever)7year
three years months
. -
control
Gets p a id Production of
€X+ Y reactive power
> Pays € Y
Pays € X JAbsorption of
reactive power
F igure 4.6: Illustration o f a g a m in g possib ility w ith a real-tim e reactive pow er m arket
A parameter related to the frequency of market clearing is the opening duration of the
market. For example, a day-ahead market has typically an opening duration of less than one
day. In addition, note that the instant of the gate closure is important as it has to be
coordinated with the other markets (see section 4.6.4).
The frequency of offer submissions (i.e., supply side) should not be confused with the
frequency of market clearing. Indeed, it represents how often an AS provider can modify its
offers. In order to reflect as closely as possible the actual conditions of the market, it may
be desirable in general to have a high frequency offer submission system. However, close to
the actual operation of the system, real-time offer submission may favour gaming and be
unpractical as well. Therefore, most of the systems have adopted a market where AS offers
are frozen on a day-ahead or longer basis before the actual operation o f the system. For
example, in PJM, secondary frequency control offers are partly frozen after 6 pm on the day
ahead (new offers and price changes are not possible), whereas the market is actually cleared
every hour of the operating day [PJM (2008)].
The frequency of reviews of the needs (i.e., demand side) in terms of ancillary services is
also an important parameter of the market timing. By reviewing on a frequent basis the
amount of ancillary services that it needs, an ERAP can ensure that it only purchases what it
really needs to meet its objective function (see Chapter 2). Table 4.12 gives the frequencies
of the reviews of the needs across various systems as of October 2006 (see 107 p. 195).
Practices are thus very different from a system to another.
T able 4.12: F req u en cies o f review s o f the n eed s across various system s as o f O ctober 2006
Every month,
Primary or on the day,
Every 5 Every' half-
frequency Twice the year - Every' day if applicable - Every year
minutes hour
control (revision
contract)
Secondary
Every 5 Every six
frequency Every day Every' day - Twice a day Every year -
minutes months
control
B asic As soon as the
Every six Every' half-
voltage - - - generating unit Unknown -
months hour
control is connected
E nhanced Every half- As soon die

Every six
voltage hour (a trading Unknown Every day unit is - - -
months
control interval) connected
The date of settlement is a minor factor, but has also to be taken into account,
especially if complex calculations to estimate the AS utilisation are needed.
Lastly, the duration of contracts is also an important parameter in the market timing.
Note that the duration of the opening of the market and the duration of the contracts are
not necessarily linked, even if they are often the same in an energy-only spot market. As
soon as a long-term contract (e.g., several years) is in place, there is no more competition
until it comes up for renewal. This can be either a drawback or an advantage for the ERAP
if the contract is respectively worse or better than the other offers on the market. In
addition, the value of long-term contracts resides in the fact that the price is fixed over
time [Kirschen and Strbac (2004a)] and allows suppliers to include maintenance and semi
fixed costs [DePillis (2006)]. Short-term contracts (e.g., five minutes) increase competition
and opportunities, but also risks. It should be noted that even if the needs are constant, the
available resources can vary during the day, which can brings some opportunities to the
ERAP.
To illustrate some advantages and drawbacks from the various durations of

contracts, let us consider the AS availability payments. In the short run, generators with
some free AS capacity will propose their remaining capacity for almost nothing, since the
unit is already committed. However, if the ERAP wants more capacity than is readily
available, it would have to pay for the start-up costs, which would lead to a very high
availability price. On the other hand, the generator can include in the long run its fixed and
opportunity costs over the whole AS supply curve and can anticipate a potential high
demand for AS. Figure 4.7 illustrates this discussion with supply curves for three different
terms (short, medium and long). For the sake of simplicity, they are represented as
continuous. However, they could have break points without changing the essence of the
argument. The demand is represented in a schematic manner. Demand curve 1 cuts the
short-term supply curve (i.e., representing only the short-term variable costs) at a very low
price, whereas demand curve 2 and the short-term supply curve intersect only at a high
price. O n the other hand, the long-term supply curve includes all the costs and tends to
have a more constant marginal cost. Therefore, prices obtained for the demand curves 1 or
2 are quite similar. In conclusion, it is clear from this curve that the short-term market is
risky, but a large gain can be achieved, whereas the long-term market is less risky, but gains
are likely to be smaller. In practice, small volumes are generally bought in short-term
markets, while long-term contracts are preferred for larger volumes, with the help of
bilateral contracts124.
A v a ila b ility Short-term

Medium-term
supply curve
P ^ce Demand curve 1 supply curve
Long-term
supply curve
Demand curve 2
Quantity of AS
F igure 4.7: Supply curves for different due dates
4 .7 Avoiding P rice C aps
Price caps appear to be good remedies against abuses o f market power. However, they
should be avoided in markets for ancillary services. Figure 4.8 illustrates the effects of an
offer cap (i.e., that specifies the maximum offer price that a supplier can propose) and a
purchase cap (i.e., that limits the buyer’s demand price). A unique cost component is
considered for each quantity of ancillary service (e.g., in €/MW), the demand curve is
assumed to be vertical (inelastic demand, e.g. for a volume R^.) and the supply to be
sufficient to meet demand and supply. In both cases, the price cap PC is the same for the
given period. Lastly, q* represents the amount of AS finally bought.
The graph on the left shows that with an offer cap the initial demand will be met but
will cause a revenue shortage for the AS providers and will discourage them from investing
in new capacity. On the other hand, a purchase cap avoids this revenue shortage, but leads to
a lower amount of AS {q* < R^). In a sense, a purchase cap is only a simplistic demand
curve. Unfortunately, the value of AS is not a linear function of the quantity of AS because
one supplemental MW of AS may be enough to avoid a load-shedding of several MW.
m See section 4.4 for a discussion on procurem ent m ethods.

4.7 AVOIDING PRICE CAPS 221
Therefore, an appropriate demand curve is much more desirable than price caps. In other
words, AS have a value, but dais value is difficult to determine because, as discussed in
section 2.2, the needs of users are hard to define.
Price of AS Demand curve Price of AS

Demand curve
PC PC
Supply curve Supply curve
(a) Offer cap

q * - R se c Quantity
(b) Purchase cap
q* r :
Z
Quantity
of AS of AS
F igu re 4.8: E ffect o f an offer cap (a) and a purchase cap (b)
Despite their drawbacks, if one of the two proposed price caps is adopted, finding
the appropriate price is an issue. Theoretically, the value of AS has to be taken into account
at some point. In practice, Table 4.13 shows the price caps adopted across various systems
as of October 2006 (see 107 p.195). Australia has implemented a purchase cap at the VOLL
(i.e. AUD 10,000 in August 2001) for active power AS, whereas PJM has adopted offer caps
at 100 $/M W /h for its secondary frequency control AS market. Most of the European
countries have not adopted any price cap yet for these products. In addition, price caps are
constant over time, whereas it would be better to have a different value for each market
clearing [Singh (1999)].
T ab le 4.13: Price caps across various system s as o f O ctober 2006
Primary Purchase cap at the

None - None None - None None
frequency control VOLL’25
Secondary Purchase cap at the Offer cap at

None None None - Unknown -
frequency control VOLL'25 100 5/MW /h
Fixed: offer cap at

999.999 £/M vai/h
B asic voltage Availability: offer cap at
- - - None None None -
control 999.999 jC/Mvar/h
Utilization: offer cap at
999.999 £/Mvarh
Fixed and availability: offer
E n h an ced voltage cap at 999.999 £/Mvar/h
None N o rcc. None None - - -
control Utilization: offer cap at
999.999 y/Mvarb
4 .8 Providing A ppropriate In cen tiv es
Since system services are public goods, free riding is a strong temptation for participants in
markets for AS. Therefore, participants should have enough incentives for behaving
properly (section 4.8.1). Besides competition, several tools are available to the market
designers to give incentives to participants. The typical solution is to introduce penalties and
rewards (section 4.8.5), assuming that sufficient A.S monitoring is in place (section 4.8.4). The
method chosen to allocate the costs of system services has also a deep impact on the behaviour of
the participants (section 4.8.2). Furthermore, an appropriate information system helps users
respond to incentives (section 4.8.3). Finally, it is important to remember that intangfble
incentives, such as the fear of blackouts, also influence participants because they can damage a
company’s image even if it was not directly responsible.
4.8.1 Stakeholders that should have incentives

First, the system operator settles the security level and therefore the associated system services.
The SO should thus have enough incentives in order to define appropriately the optimal
quantity, quality and location of system services as described in section 2.2. These
specifications should be in accordance with all the other interconnected systems. In
125 T he Value O f L ost Load (VOLL) is equal to A U D 10,000 in Australia (August 2001) [N EM M CO
(2001)].
4.8 PROVIDING APPROPRIATE INCENTIVES 223
addition, in the case of AS trading across systems and with the SO as ERAP, the SO should
not have a right of pre-emption over the cheap AS providers within its territory.
Second, the entities responsible for A S procurement should have a procurement cost in
accordance with the objective function of the SS needs. If the ERAPs are users of the
system (e.g., the PJM’s LSE described in section 4.2), this pressure will come up naturally
with the competition against other ERAPs, except in the case of the use of a dominant
position. If the ERAP is the SO, the regulator should intervene to control the cost with the
classical tool of regulated activities. For example, Weisman (2005) compares two types of
regulated price: proftt-share penalties and revenue-share penalties. Note that the profit-
share penalties seem to be more efficient than the latter one. However, these regulated
prices do not guarantee that AS expenses are optimal.
Third, the A S providers should not deviate from what they are contractually obliged
to deliver and they should declare any deviation from such obligations. Indeed, not
delivering the contracted AS when requested could jeopardize the security of the system. In
addition, they should have enough incentive to offer their services at their marginal long
term cost.
Lastly, it is desirable to give some incentives to all the users of the network to improve
their behaviour in order to reduce the need for system services.
4.8.2 Allocation of system services costs

Since system sendees are public goods, determining their cost is a joint cost allocation
problem126. The cost allocation has to be fair and revenue-adequate (i.e., that all the SS cost
should be recovered). The Oxford Dictionary (2001) defines the adjective “fair” as: (a)
“treating people equally” or (b) “just or appropriate in the circumstances”. Therefore, a
“fair” cost allocation is likely to be subjective. In this section, it is considered that the
allocation of SS costs should give incentives to maximise the benefit to users in the long
run. However, such an allocation may go against the uniformity of some transmission tariffs
(e.g., the postage stamp tariff, see section 1.3.3) in force in many European countries. The
126 Because o f the public good characteristic o f system services, n ote that the allocation o f SS costs is
som ew hat similar to the transm ission pricing problem .
choice between socialisation and differentiation of some cost components of SS is a

political decision and the legislator has thus an important role to play.
A first method to allocate the SS costs is to distribute the amount of AS to procure

between users of the networks, who thus become ERAPs. However, section 4.2 has argued
that such a solution should be avoided. Nevertheless, the SO has always some SS costs to
recover, even if the procurement of AS is delegated to some users of the system. A second
method is to have a unique ERAP (the system operator), which then allocates the cost of SS
to users of the system. The cost allocation should depend on the type of cost considered,
namely: (a) the fixed costs (section 4.8.2.1); (b) the potential SS utilisation cost
(section 4.8.2.2); and (c) the SS actual utilisation cost (section 4.8.2.3).
4 . 8 . 2 . 1 A l l o c a t i o n o f t h e fi x e d c o s t s
To fairly allocate SS fixed costs127 a first solution is to share them in proportion to the use
of the system. Indeed, SS fixed costs, besides being sometimes difficult to quantify, are
usually difficult to separate between users. For example, the cost can be split amongst
different classes of users. Then, the cost for each class is divided amongst the users in the
class on the basis of their energy consumption/production only. Obviously, this cost
allocation can have different levels of complexity.
The easiest solution to implement is to include directly the SS fixed costs in the
transmission tariff, as recommended by a Cigre Task Force (2001). This approach seems to be
the fairest and most of the European systems have adopted it [ETSO tariffs task force
(2006)]. Consequently, the SS fixed costs are allocated in function of the various parameters
of the tariff (e.g., voltage level at the connection point, maximum power subscribed, or
quantity of energy injected/withdrawn).
4 . 8 . 2 . 2 A l lo c a tio n o f t h e p o te n tia l u tilisation c o s t
Even if no user actually uses any SS, a quantity of SS is always reserved as a function of the
particular configuration of the system, such as the type o f users connected or the power
transits (see section 2.4). Important parameters in the assessment of the SS needs are the
reliability and the size of users connected. For example, Kirby and Hirst (2000) have shown
that industrial consumers account for 93 % of secondary frequency control. On the other
127 SS fixed costs are for example T O ’s capacitor banks, the inform ation system o r the staff necessary
to provide the system services (see section 3.2.1).
hand, large generating units increase the requirement for tertiary frequency reserve. In
addition, an unreliable large generator increases risks more than a small reliable one. Lastly,
loads are the principal consumers of reactive power.
The tariff proposed in section 4.8.2.1 may be inefficient to allocate the SS potential
utilisation cost, because such a tariff does not give any incentive to the entities that increase
the SS potential utilisation cost to reduce it. Therefore, it would be interesting to distinguish
between classes of users in order to allocate the cost of ancillary in a fairer manner. For example,
in Australia, the cost of the “lower frequency control contingency” ancillary sendees is
charged to loads, while generators pay for the “raise frequency control contingency”
ancillary sendees [NEMMCO (2001)]. However, despite a better allocation, such an
arrangement is not likely to give enough incentives to individual users of the network to
improve their behaviour. Indeed, it still consists of only a single price for a whole category
of users and therefore encourages free riding.
Strbac and Kirschen (2000) have proposed to allocate the cost of frequency control
SS to producers as afunction of their siê and their availability. This method thus allocates the SS
costs to generators that are likely to use them (i.e., differentiation between “bad” and
“good” producers). The proposed method could be extended to consumers. However,
three issues have to be addressed for such a cost allocation. First, no perverse incentive
should be given to users. For example, an AS provider should not have any incentive to
increase the SS needs in order to benefit from the sale of additional ancillary services.
Second, the period considered to calculate the availability of generators has to be chosen
carefully, since the availability of a unit may vary a lot, depending on the operating
conditions and the maintenance. Last, the capacity of a unit can be hard to define, since a
generator is unlikely to produce at its maximal output all the time. And if the payment is
based on the declared capacity for, say, the next hour, generators may be tempted to under
schedule, as happened in the early designs of the Californian AS market [Singh (1999)].
4 . 8 . 2 . 3 A llo c a t io n o f t h e a c t u a l utilisation c o s t
The method described in section 4.8.2.2 allocates only the potential SS utilisation cost.
However, the SS actual utilisation cost varies in real-time, depending on the availability of
cheap AS providers. Therefore, two users of the system can have the same potential
utilisation, but one can use SS when they are expensive (e.g., during a consumption peak),
while another can use them when there are cheap (e.g., at night, see section 3.5.2). Note that
only a fraction of the reserved SS capacity is actually used in real-time operation128.

Nevertheless, sufficient active and reactive power reserves are necessary at all times because
system events are by nature unpredictable. Moreover, it is important to note that the
utilisation of active power and reactive power SS differ. In fact, the arithmetic mean of
frequency control utilisation is close to zero, while the arithmetic mean of voltage control
utilisation is generally non-zero129.
The Cigre Task Force (2001) proposed three methods to allocate the cost of actual
SS utilisation, namely: (a) marginal cost allocation, (b) Shapley-value pricing; and (c)
Aumann-Shapley pricing130. It is assumed here that the SS cost function is known, which is
not necessarily easy, as shown in Chapter 3. First, the marginal cost allocation method charges
users in proportion to the marginal cost they induce because of their SS use. This method
usually gives a higher amount recovered than the actual cost because the SS cost function is
usually convex (the marginal cost is thus higher than the average cost). If a revenue
adjustment is done, revenue and actual cost would match but the cost allocation would not
be efficient.
Second, the Shapley-value method calculates the cost induced by each user as a function
of their entrance order. All the entrance combinations are tried and then an average is
performed. This scheme, in addition to its high computational requirements, is not neutral
with respect to the size of similar users (two users with the same impact on the cost
function but with different sizes would have different rates). This latter drawback can be
resolved by splitting the users into equal sizes (agent splitting method). However, this
increases the computational efforts required.
us p o r example, frequency variations are generally o f low am plitude in large interconnected systems
(e.g., the U C T E o r the N orth-A m erican systems), so solicitations for providers o f primary frequency control
are m ost o f the time relatively low.
129 T he frequency o f an interconnected system is kept centred around its nom inal frequency, so the
deployed quantities o f "u p active pow er balancing” and “ dow n active pow er balancing” (see section 2.3.2) are
relatively similar. O n the other hand, users cannot easily procure reactive p ow er by themselves and the
netw ork naturally uses reactive pow er. Therefore, the deployed quantities o f “up reactive pow er balancing”
and "dow n reactive pow er balancing” are generally very different.
130 Y oung (1994) describes in m ore details and from an econom ic p o in t o f view these cost allocation
methods.
Third, the Ajtmann-Shapky method is the limit case of the agent splitting method
where the size of agents tends to zero. The price n} charged to each user i is given by (4.1).
j 3 c y r ) dt (4l)
o %
where q* represent the SS actually used (it is a vector that represents the SS utilisation at
every bus of the network), C{g) is the SS cost in function of the SS used and y is the SS
used by the user i. The main drawback of the Aumann-Shapley pricing is that it requires the
SS cost curve. This curve can only be calculated through a centralised algorithm and only if
providers give a complete offer curve. For example, Lin et al (2005) have applied this
method to the reactive power SS in an eight-bus system. In this case, the price of voltage
control SS is computed for each bus and only the opportunity cost of the reactive power AS
is considered.
Another possibility is to charge the utilisation of SS through other-related mechanisms

such as the balancing mechanism for frequency control ancillary sendees.
In conclusion, a cost allocation method based on the actual SS utilisation is

theoretically possible and results in different prices for different users. However, a practical
realisation may be easier for active power SS than for reactive power SS, since the former
sendees already benefit from an information system.
4.8.3 Transmission of data

Transmission of data is crucial for a market, as sufficient and balanced information has to
be provided to market participants in order to foster competition while avoiding collusion.
First, signals may help to take the appropriate decisions. For example, high reactive power
supply prices will provide incentive to: (a) reduce the demand for reactive power in this
area; (b) increase cheaper supply; or (c) find alternative solutions (e.g., building new
transmission lines). Second, if the signals sent are not symmetrical (i.e., they are not the
same for every stakeholder), the participants who benefit from the most relevant
information have a definite advantage over the others. Lastly, too much transparency is
likely to remove competition by introducing tacit collusion. Therefore, the transmission of
data has to be regulated carefully.
As highlighted by Hirst and Kirby (1997), three questions have to be answered

concerning data publicity. First, which data} As a general rule, data should cover [Newbery et
al, (2003)]: (a) the trading mechanism, (b) the needs in terms of SS, (c) factual data on the
structure of the power system, such as generation capacity and reliability131, transmission
grids and interconnections, reserves and imports and exports of electricity; and (d) the
relevant economic signals. In the case of uniform pricing (see section 4.6.5), publishing the
clearing price is the most natural choice, since it gives an incentive to AS providers to offer
under this price in order for their offer to be accepted. However, the question is more
difficult for a discriminatory auction pricing. For example, publishing the highest offer in a
pay-as-bid system might encourage participants to raise the price of the lower offers in
subsequent auctions. Therefore, weighted average prices for discriminatory auction pricing
are more desirable, as it is done in Germany for the frequency control AS tendering
processes. It is also desirable that some elements of penalties and rewards be communicated
Second, who should have access to the data} The same data should be available to all
participants of a class (e.g., AS providers) to avoid any discrimination. However, because of
their specific position, the SOs and the market monitoring entity should be allowed to
gather more information than other participants (which thus creates asymmetrical
information). In the case of AS trading across systems, the transmission of data between
SOs is getting very sensitive.
Third, when should be the data published? They can be disclosed either ex-ante or ex-post,
as discussed in section 4.6.5.
4.8.4 Monitoring
Monitoring of ancillary services covers both technical and economic aspects. Indeed, on the
one hand, AS providers have to actually provide the contracted ancillary services. The goal
of this technical monitoring is to detect non-complying AS providers that might endanger the
security of the system. This technical monitoring is thus more important than the one for
energy, because AS are directly related to the security o f the network. On the other hand,
participants should behave fairly by not abusing their dominant position. In addition,
131 H ow ever, generating companies are likely to be reticent about providing the reliability inform ation
on their units.
market flaws have to be detected as soon as possible. Note that such an economic monitoring is
efficient only when it is associated with a system of penalties and rewards, which is
discussed in section 4.8.5.
4 . 8 . 4 . 1 T e c h n i c a l m o n ito rin g
As shown in section 2.3, the definition of ancillary sendees includes numerous parameters.
Therefore, AS monitoring requires specific equipments such as Phasor Measurement Units
(PMU) to assess the efficiency of the AS provider132. Since the SO is in charge of the
security of the system and is the only entity to have sufficient technical means to monitor
the AS delivery, it should be in charge of the AS technical monitoring. Note that in case of
cross-border trading (see section 2.5), monitoring may be difficult and does require a strong
cooperation between the TSOs involved.
The SO’s technical monitoring may be split into three components. First, the SO
should cany out a preliminary monitoring in order to test the actual capability of the AS
provider [e.g., Robert and Prestat (2004)]. Alternatively, the AS provider may give enough
proof of its capability to fulfil the various criteria. In every case, this preliminary monitoring
should be compulsory for any new provider in the market.
Second, the SO should enforce a continuous monitoring in normal operation. This

monitoring may be costly, especially when the number of providers increases. To reduce the
number of controls, the monitoring can be split into two levels. A global monitoring may
help detect areas that do not behave as expected. A more precise monitoring of the AS
providers within this area is then possible. Another solution to reduce the difficulty of
collecting measurement from many sources is to choose only a few AS providers to supply
AS for the whole system and to monitor them carefully, as is done in the Netherlands. Such
a method allows an easy and low-cost monitoring, but may not favour security. Indeed, it is
a priori better to have numerous small providers rather than a few large ones. A third
solution to reduce the monitoring burden is to select randomly the AS providers being
monitored carefully. Lastly, to monitor the contributions of distributed generation, a good
solution is to delegate the task to DSOs and then to monitor the contribution of DSOs.
132 F o r detailed explanations on the actual im plem entation o f the technical m onitoring o f AS, see for
example Sterpu (2005) or M argotin ct at (2006).
Third, because AS are principally used when the system is stressed, AS monitoring
is effective only during system events such as the sudden loss of a large generating unit. As
such an event is unpredictable, a SO has to wait for an event of this kind to actually
monitor an AS provider. Therefore, the SO may carry on instead an extraordinary monitoring
by both stressing the AS provider (e.g., through the variation of the regulation signal sent to
the units from the secondary frequency control) and monitoring its output. This type of
monitoring should be exceptional because it stresses the system.
Lastly, since SOs cannot monitor everything, the proposed scheme should be
complemented by incentives for self-improvement and transparency {self-monitoring). In fact,
AS providers should be encouraged to declare any problem to the SO. However, AS
providers would need sufficient means to monitor their service, which is not always the
case, especially for small providers. Note that this lack of self-monitoring also means that
the AS provider is not in a position to able to argue against the SO’s figures.
4 . 8 . 4 . 2 E c o n o m i c m o n ito r in g
The economic monitoring is usually the task of the regulator or/and the anti-trust bodies
(see section 1.2). The SO can also be involved, but obviously only if it is independent. The
economic monitoring helps both give an incentive to participants and improve the
procurement method. It can be split into four activities [Perrot and Molinier (2005)]: (a)
detecting anti-competitive behaviours; (b) investigating in case of suspicion; (c) imposing
sanctions against proven anti-competitive behaviours (e.g., by using penalties described in
section 4.8.5); and (d) correcting the market design (see section 4.9).
Detecting an anti-competitive behaviour by a participant can be either done

continuously in real-time (intensive method), as in the USA for energy spot markets, or
upon ex-post request (non-intensive method), as in Europe. Discussing the advantages and
the drawbacks of the available methods is not within the scope of this thesis. However, two
comments should be highlighted. First, economic monitoring is intrinsically very difficult.
Second, since monitoring rules are different across jurisdictions, economic monitoring of
cross-border trading is likely to be even more difficult. Lastly, note that section 4.9.3
discusses in more details the assessment of the global economic efficiency of the
procurement process.
4.8.5 Penalties and rewards

Monitoring without penalties or rewards does not give any incentive to participants to
improve their behaviour. Penalizing AS providers should be the SO’s task, since it is
independent and has the means of control (see section 4.8.4). For example, the contract in
France between the AS providers and RTE defines different levels of alert, with particular
penalties [RTE (2007b)]. The penalties and rewards can then have different nature,
intensity, structure and volatility.
First, the nature of the incentives has to be decided. In fact, penalties and rewards
may be either financial or in other terms. For example, a unit which falls to provide a
contracted sendee may have to provide free sendee (or receive a reduced payment) for a
given period of time or until its next good response [Proctor (2001)]. In New Zealand, the
party responsible for a disturbance pays some event charges determined annually by the
TSO. Then, this fee is redistributed amongst the providers of the frequency control
ancillary sendees during the two months prior the event [Electricity Commission of New
Zealand (2005)]. Other schemes may be imagined, such as a redistribution of the fee
amongst the “good” users during the event. Unfortunately, the literature does not seem to
provide any discussion on this particular issue.
Second, an appropriate structure of the remuneration or the fee should be defined.

Section 4.5 provides a discussion on the various structures possible.
Third, the intensity of the incentives varies with the measured performance. Keller
and Franken (2006) have proposed four different incentive schemes, as shown in
Figure 4.9.
Lastly, the volatility of the incentives should be determined. Indeed, penalties and
rewards can be either fixed or variable over time. For example, they can be high when AS
resources are scarce and low when the resources are abundant [DePillis (2006)], such as
balancing mechanisms that provide volatile penalties (the price to pay for an imbalance
depends on the market conditions).
2. Continuous
1. Standard 4. Dead Band
3. Capped
Measured
performance
High U E Low
■E
(0
©
a:
Figure 4,9: Four in ten sities o f an incentive sch em e [Keller and Franken (2006)]
4 .9 A s s e s s in g th e P rocurem en t M ethod
Once a complete procurement method has been set up, it should be assessed against the
following three goals: effectiveness (section 4.9.1), minimal running cost (section 4.9.2) and
economic efficiency (section 4.9.3). It is important to perform this assessment regularly
since the various parameters o f the market evolve over time, such as power system
operating conditions, the number and types of AS providers, or the behaviour of
participants. This assessment should be followed by an improvement of the market design.
4.9.1 An effective procurem ent process

Effectiveness measures the extent to which all the system services that are needed are made
available when required. So far, most of the systems in place are effective since ERAPs
achieve to procure the required amount of ancillary services to fulfil the SS needs. Note that
an effective procurement is not necessarily efficient, because the cost of SS procurement is
not taken into account when assessing effectiveness.
4.9.2 Low running cost

The cost of running theprocurement system should be minimized because it is subtracted from the
global welfare. While this cost is sometimes forgotten in the liberalization process, setting
up the metering, monitoring, billing and processes required can be expensive. The possible
4.9 ASSESSING TH E PROCUREMENT METHOD 233
gains from setting up a complex market for ancillary sendees rather than keeping a simple
framework should therefore be considered carefully. To give an idea, the British balancing
market is said to cost ^80m/year to run [Thomas (2003)]
4.9.3 Economic efficiency

Economic efficiency depends on the behaviour of the participants. In theory, in a perfectly
competitive market, all participants reveal the true value or true cost for a good. In practice,
as shown in section 2.2, revealing the true value of system sendees is very difficult.
Furthermore, markets for ancillary sendees suffer from high market concentration and poor
liquidity because of the technical characteristics of ancillary sendees, as only a few providers
are able to meet the technical criteria and there may be location constraints. Lastly, if the AS
market is poorly designed and the demand is not very responsive, there is a strong chance
that competition might be weak. Therefore, before moving towards the spot market
paradigm for ancillary sendees, it may be wise to assess carefully and honestly the context
and current markets and decide whether such a move is justified.
4 .9 .3 .1 M e a su r in g th e e c o n o m ic efficien cy
To measure the economic efficiency, tire deviation from the simulated competitive price can
be assessed, with, for example, the Lerner index. The Lerner index L I is calculated as in (4.2),
where rt* is the actual price and 7t the simulated competitive price [Liu el ai (2006)].
Sometimes, the competitive price may be hard to determine, so benchmark prices can be used
instead. For instance, generator’s reactive power offers in Australia are compared to
benchmark prices based on the cost of SVCs [Cigre Task Force (2001)]. Deviations of the
actual price from the simulated or benchmarked competitive price can be due to the abuse
of a dominant position. However, some participants may have a dominant position but still
can behave acceptably by helping to reach the competitive price. The parameters affecting a
dominant position are discussed in section 4.9.3.2.
* > *
n
It may be useful for SOs, regulators and AS providers to be able to benchmark their
own markets for ancillary sendees against others and over time. By rising questions and thus
initiating debates amongst participants, such a benchmark may improve the efficiency of
markets. The proposed Cost Indicator (Cl or C IK

AS) for a given zone and a given ancillary
service is calculated as in (2.4).
a ^ = |f- (4-3>
tmrgf
where C Z
AS (in € /year) is the annual ancillary service cost and C* (in € /year) is the annual
wholesale energy cost. CIJ is obtained by multiplying the average wholesale market price
in the same year as CI'AS by the energy consumption of the country in 2003133 PEA (2005)].
The wholesale energy cost is taken instead of the cost for end users in order to avoid price
distortion due to taxes or specific agreements. CAS is either estimated by multiplying the
average price by the average volume of ancillary services or obtained directly from reports.
Figure 4.10 shows the cost indicators in 2004-5 for primary frequency, secondary
frequency and voltage control ancillary services for some systems considered in this thesis.
The results are of the same magnitude as those reported in the study by Hirst and Kirby
(1996) on the AS cost for twelve US utilities. The cost indicators for primary frequency
control are fairly similar across all systems. This Cl is comprised between 0.5 and 0.7 % of
the energy cost, except in Australia where it is 0.35 %. This can be explained by a low price
and a low negative reserve volume. Because of the lack of data on the new British market
for ancillary services, the average cost of the pre-October 2005 system is used.
Secondary frequency control costs are rather high in Spain (more than 4 % of the
energy cost). Indeed, the TSO bought large volumes at high prices in 2005. These high
prices may be explained by the drought that affected Spain during that year. The French
cost indicator is one of the lowest (0.5 %) mainly because of a volume that is more
optimized than the average (see section 2.4.5). Cost incurred in Germany, PJM and New
Zealand are equivalent (from 1 to 2 %). Lastly, the Australian cost indicator is the lowest
because the amount of secondary frequency control reserve is very low.
Australia is the country with the highest voltage control cost. New Zealand cost is
very low, even if the basic voltage control is remunerated, probably because of the large
133 More recent data were not available at the time o f the study.
4.9 ASSESSING THE PROCUREMENT METHOD 235
proportion of cheap hydro units. PJM and France have a similar and intermediate position,
while Great Britain is among the cheapest. Lastly, since the Swedish TSO does not
remunerate voltage control, no cost indicator can be calculated.
4.5 n
Primary frequency control
S econ d ary frequency control

3.5
| : : : :| V oltage control
= 2.5
o 2 .0
« 0.5 ■■
FR DE* ES* GB $
*: V oltage control co st unknown
F ig u re 4.10: A ncillary serv ices c o st in d ic a to rs acro ss sy ste m s su rv ey ed in 2004-5
4.9.3.2 S tructural param eters influencing the econom ic efficiency
The parameters given in this section influence the dominance of some participants and are
intrinsic to the market. First, the market concentration quantifies the diversity of possible AS
providers and is usually measured using indicators such as the Herfindahl-Hirschman
Index1'4 (HHI) or the Residual Supply Index1'0 (RSI). A high market concentration is often
mentioned as the main reason of dominance. This lack of competitors is said to lead to an
inefficient market.
Market concentration can be reduced by several means. First, large suppliers may be
split into smaller ones. Besides having vert7strong political consequences, such a policy may
be unfavourable for investments, which are often large in the electricity sector, rely
economies of scales and depend on the control of technology. Second, less restrictive
134 The H erfindahl-H irschm an Index o f a market is calculated by summing the squares o f the
individual m arket shares (in %) o f all participants. It is thus com prised betw een 0 and 10,000. The
correspondance betw een values o f the H H I and the generally agreed concentration o f the market is as follows:
0-999: low; 1,000-1,800: m oderate; 1,801-10,000: high [Newberv et al (2004)].
135 The Residual Supply Index o f a company at a given time is calculated by subtracting the
com pany’s capacity’ (in MW) from the total capacity (imports included and in MW), and by dividing this
difference by the total dem and (in MW). Therefore, if the RSI is inferior to 1, the company is pivotal and has
thus a strong im pact on the market [Newberv et aL (2004)].
technical requirements may also decrease market concentration because they increase the
number of potential AS providers. As explained in section 2.3.1, a trade-off has to be found
between requirements that are too general or too restrictive. Third, barriers to entry (such as
compulsory connecting conditions that are too complex or long-term bilateral contracts)
decrease the number of potential AS providers. Therefore, removing these barriers may
help decrease market concentration. Lasdy, appropriate incentives, e.g. through a fair
remuneration, have to be given in order to foster competition in the markets for AS.
In practice, Thomas (2003) and Haas et al. (2006) highlight that the electricity sector
is strongly concentrated in Europe. The situation is worse in markets for ancillary services
because fewer units are able to provide AS. However, demand for energy is much more
important than for AS, which makes a comparison between markets for energy and AS
difficult. Nevertheless, actual markets for voltage control suffer from intrinsic high market
concentrations because of their local nature (see section 1.5). On the other hand, markets
for primary and secondary frequency controls are generally less concentrated because
participants located anywhere can provide this service as long as transmission capacities are
sufficient. However, primary frequency control has to be uniformly spread around the
network and technical requirements are more demanding for primary frequency control
than for secondary frequency control (see Chapter 2), so more units are able to provide this
latter sendee.
Second, the liquidity of the market also influences the dominance of participants.
Indeed, even small agents can easily change clearing prices of a non-liquid market by selling
or buying small quantities of AS. Good transmission and standardised products help
increase the liquidity of the market. Note that a market can be both liquid and concentrated.
Third, the topology of the network is an important parameter because the dominance of
an actor increases with network limitations (e.g., congestions). For instance, a non-meshed
network is more easily subject to strategic behaviours because it is easier to create
constraints in such a network. In addition, predictable network events such as the utilisation
of large amount of secondary frequency control at a given hour of the day are also likely to
increase the dominance of participants. Consequently, system operators have a key role to
play in the mitigation of dominant positions through a network design and operation that
foster competition. However, this mitigation of dominant positions has to be at any time in
concordance with the system security.
4.9 ASSESSING THE PROCUREMENT METHOD 237
Fourth, the characteristics of the demand are also very important. First, it is desirable to
have a demand for AS much lower than the supply in order to increase competition [Cali et
al. (2001)]t36. Second, an appropriate demand responsiveness reduces the incentives to
deviate from the competitive price. However, the demand is often inelastic in practice (see
section 2.4). Nevertheless, competition between ERAPs may help introduce more demand
elasticity. For instance, section 2.5 discusses the procurement of AS across different areas,
which implies a competition amongst TSOs. For a further discussion on demand curve for
reserves, Hogan (2006) discusses the definition of such a curve for operating reserves.
Fifth, along with the restructuring of electricity markets, the separation of responsibilities
is an important issue, in particular the unbundling of network and generation/supply
activities (see section 1.2). In the particular case of ancillary sendees, trading AS across
systems can raise similar questions. Indeed, the TSO usually manages the transmission
capacities at its borders. It is also the natural buyer of AS within its territory. Therefore, it
would be relatively easy for the TSO to have a right of pre-emption over the cheap AS
providers within its territory because the TSO would be both deciding the rules and playing
the game.
Lastly, the market design has an obvious impact on the influence of some actors. For
instance, the repetitiveness of the auction in similar conditions may favour strategic
behaviours (see section 4.6.6). As discussing the various issues related to the design of
markets for AS is the principal motivation of the whole chapter, no details are given here.
Broadly speaking, it is important that participants trust the market.
Consequently, it is very difficult to guarantee economic efficiency for AS markets.

However, despite all these practical issues, a competitive market for frequency control AS is
usually assumed in the literature. It is however generally recognized that markets for voltage
control are difficult to implement. A competitive market implies that no participant uses its
dominant position, which is hard to verify and may change over time. Therefore, it is
desirable to find practical solutions to mitigate the potential abuse of dominance as
described in section 4.8.
136 See section 4.3 for a discussion on how foster the supply o f AS.
4 .1 0 Sum m ary
This chapter has reviewed in a structured manner most of the issues and solutions related to
the design of the procurement of system services. It is clear that an ideal market design does
not exist, so tire choice of one method over another one deeply relies on the history, the
policy of the legislator, the design chosen for other related products (e.g., energy,
transmission or balancing markets) and the particularities of the power system considered.
To improve current market designs, a fully independent body (e.g., the regulator)
should describe and justify the methods adopted in its system area to solve every issue listed
below. Flaws will then become obvious and thus stakeholders will have strong incentives to
propose improvements.
V Nominating the entity responsible ofprocurement (ERAP): the SO is the best candidate since it
is a natural buyer of AS;
^ Matching supply and demand: long-term supply is achieved with compulsory connecting
conditions. Though it is not the optimal solution, this policy is currently the most
practical. However, reduction of the long-term demand is desirable but rarely
implemented in practice. Concerning short-term procurement, demand response is
compulsory if the ERAP has not covered its supply against risks. More generally,
demand for AS does not reflect the true value of SS in all the surveyed systems;
V Choosing the relevant procurement methods to contract AS provision: various methods are
possible for the ERAP, namely: (a) compulsory provision, (b) self procurement, (c)
bilateral contract, (d) tendering and (e) spot market. None of these solutions is perfect,
so it may be desirable to select a few of them and use them concurrently;
V Defining the structures of offers and payments to remove price distortions: six structures of
offer and payment have been proposed, namely: (a) a fixed allowance, (b) an availability
price, (c) an utilisation payment, (d) the opportunity cost, (e) an utilisation frequency
payment and (f) a price for kinetic energy. The choice amongst these components
depends on the cost structure of the ancillary services considered (e.g., high fixed costs
or high variable costs).
V Organising the market clearing procedure to fix the prices and amounts of transactions
between participants: In conclusion, because of the number of possible methods that
are still under debate, designing the most appropriate market clearing procedure is a
complex task.
4.10 SUMMARY 239
■ Two structural arrangements are possible: centralised unit commitment (pool) or

decentralised commitment (exchange). The choice is mainly a political matter,
though centralised structures are theoretically more efficient because a global
optimization can be performed;
■ The auction defines how the offer and the demand for AS are submitted and
matched. The Vickrey and the YCG auctions have nice features but are not very
practical. Therefore, classical sealed-bid auctions are preferred in practice, even if
such auctions are not Pareto efficient;
■ Scoring, which classifies AS offers, is very difficult because of the multi-dimension
feature of AS;
■ Coordination between the clearings of the various AS markets is essential. In particular,
simultaneous clearing is the most efficient solution, but it is harder to implement in
decentralised systems and is computationally very demanding;
■ The settlement rule organises the payment of the AS providers selected in the auction.
Non-remunerated polices remove the property rights of AS providers and thus have
to be avoided. Administrative prices are not efficient either, except in the presence
of dominant players. The best settlement rule is still under debate (e.g.,
discriminatory or uniform price, zonal or nodal price, etc...);
■ The timing of the market has to be chosen carefully. Indeed, higher frequencies are
likely to increase the efficiency of the market, but also the potential for abuse of
dominance and the operating cost.
V Price caps of any kind should be avoided as much as possible. An appropriate demand
curve is more desirable;
C Providing proper incentives to stimulate appropriate behaviours by participants, besides
intangible incentives (e.g., die fear of blackouts) and competition:
■ The allocation of VI costs should ideally be performed with three components: (a) the
fixed cost of SS through the transmission tariff, (b) the variable AS reservation costs
through an allocation based on the potential utilisation of the users and (c) the
actual utilisation of SS through to the users that use them;
■ Transmission of data: sufficient and balanced infonnation has to be given to the
market participants to foster competition. This information should cover the trading
mechanism, the needs in terms of SS, the structure of die power system and the
relevant economic signals;
■ Monitoring of the participants in markets for AS: the SO usually performs the
technical monitoring, which is divided into a preliminary, a continuous and an
extraordinary monitoring. Since it is very difficult to monitor all the AS providers,
incentives have to be given to encourage self-monitoring. The regulator and the
anti-trust body are usually in charge of the economic monitoring, which is split into
four functions: detecting, investigating, imposing sanctions and correcting;
■ Pena/ties and rewards-, to be effective, the monitoring should be complemented by

penalties and rewards of various natures, structures, intensities and volatilities.
S Assessing theprocurement method to improve it. The procurement method should fulfil three
goals; effectiveness, minimisation of running cost and economic efficiency. The
optimum between these three goals is currently searched through an iterative process.
However, assessing these three indicators of the procurement method is actually very
difficult and it is almost impossible to guarantee a competitive price for AS. Actual AS
prices are thus often obtained with the help of acceptable behaviours of the participants
or regulatory decision. ■
CHAPTER 5
CONCLUSIONS AND FUTURE
RESEARCH
Thefuture belongs to the one who has the longest tnemoty.

Friedrich Nietzsche (1844 - 1900)
5.1 R ation ale for th e T h e sis
From consumers to generating units, including system operators and transmission owners,
all users of an electricalpower system expect an appropriate standardforfrequency and voltage. The global
frequency and the local voltages can be controlled by acting on the balances between
consumption and generation of active and reactive powers across the power system. The
frequency and voltage control sendees are called system sendees (SS). In addition to
expecting a certain standard, the power system users affect the balances of reactive and
active powers. Therefore, any user can both perturb and help controlfrequency and voltages. When
users help control frequency and voltage, they provide the so-called ancillary sendees (AS).
While the responsibility of users to provide ancillary services was less clear twenty years ago,
die development of information technologies, power electronics and electricity markets
force now all users to face tiieir double responsibilities as consumers of system services and
A Comprehensive Assessment of Marketsfor Frequency and Voltage Control Ancillaty Services, Yann Reborns, 2008 241
242 CHAPTER 5 CONCLUSIONS AND FUTURE RESEARCH
providers of ancillary services. However, users need a consistent framework to manage ancillary
services in order tofully grasp opportunities and hence maximize the global welfare.
So far, academic and industrial research was concentrated on specific issues and
sometimes ignored the relations between technique, cost and market design. A global
framework was thus missing. To fill this gap, this thesis provides a comprehensive assessment of
markets for frequency and voltage control ancillary services along three axes: (a) defining the needs for
system services and the ancillary services that can fulfil these needs; (b) assessing the cost of
ancillary services; and (c) discussing the design of an efficient market-based procurement of
system services.
Such a framework exhibits several advantages', (a) stakeholders can grasp quickly the issues
related to ancillary services; (b) stakeholders benefit from a standardised method to assess their
system', (c) solutions are proposed to improve current arrangements', and (d) theoretical limitations
that need future work have been identified.
5 .2 C ontributions to K n o w led ge
First, this thesis has formulated the problem of the needs of users. To meet these needs,
given qualities and quantities ancillary services have to be provided at appropriate locations.
The issues raised by these technical specifications have been discussed and compared with
actual requirements across systems. In particular, the thesis has led to the following
conclusions:
^ Needsfor ystem sendees:

■ Expression of the needs', a cost-benefit optimisation is the best manner to define the
needs in terms of system services. Security, power quality and system utilisation
should all be taken into account in the optimisation process. However, practical
solutions tend to be based on obscure or simplified methods, past experience or
general recommendations;
■ Review of the needs', the needs of users and the system constraints evolve with time.
Therefore, the needs for system services should be reviewed frequently by system
operators. However, system operators have little incentive to change rules-of-thumb
that have performed reasonably well for years, even if these rules are sometimes
hard to justify.
5.2 CONTRIBUTIONS TO KNOW LEDGE 243
S Quality of ancillary services'.

■ Differentiation of ancillary services, it is not possible to find a perfect specification
between general and precise specifications of ancillary services;
■ Description of ancillary services, a framework to compare ancillary sendees has been
proposed and successfully applied to various power systems. In addition, a
standardised description has been proposed, as well as a standardised definition of
the term “spinning reserve”;
■ Standardisation of ancillary services, in practice, specifications of ancillary services differ
from one system to another. Therefore, efforts should be made to unify similar
ancillary sendees in order to foster short-term exchanges of services between
neighbouring systems.
K Quantity of ancillary services
■ Responsibility for quantity requirements, because system sendees are public goods, only
the system operator should define the requirements for ancillary sendees;
■ Definition of quantity requirementsfor ancillary services, the requirements should take into
account various features. However, current practices are usually opaque;
■ Elasticity of quantity requirements, the demand for ancillary sendees should be elastic;
■ Synopsis of quantity requirements, an indicator has been developed to compare quantity
requirements for ancillary sendees and thus gives incentives to system operators to
improve their practices.
K Eocation of ancillary sendees.
■ Importance of location', the location of reactive power providers is particularly
important. The actions of frequency control providers are global but still may be
hampered by a bad location (e.g., a congested area);
■ Mechanisms to exchange ancillary services between locations, technical mechanisms should
exist to allow participants exchange AS between neighbouring locations. However,
these mechanisms usually do not exist between neighbouring systems.
Second, this thesis has reviewed the main cost components of ancillary sendees and
has proposed a methodology to estimate the day-ahead de-optimisation cost incurred by a
producer because of frequency control. This methodology has been successfully applied to
EDF Producer, and has provided several interesting results. The main findings are:
K Estimating accurately the total cost of ancillary sendees is extremely difficultfor aproducer because:
■ The cost of ancillary services has several components:,

■ Different inter-dependent time horizons have to be considered, from tens of years ahead
to real-time;
■ It is often difficult to separate the cost o f ancillary services from the cost of other
products;
■ The cost of ancillary services depends on the portfolio considered and it is affected by many
uncertainties.
^ One main cost component of frequency control is the day-ahead de-optimisation cost
because of the need to set aside capacity to provide reserves. A. methodology has been
proposed to estimate this cost and a specific tool, named OTESS, has been developed to apply
tlais methodology to any producer;
S This methodology has been successfully applied to ED F Producer’s portfolio (i.e., around 150
thermal units and 50 hydro valleys) over 879 days with actual operational data and
ED F’s actual optimisation algorithm;
'T The results obtained are veiy helpful to assess thefrequency controlprovision. In particular:
■ The de-optimisation cost is not negligible, since it represents up to 7.8 % of the initial
dispatch cost as an underestimate. Therefore, an efficient provision of frequency
control reserve is essential to reduce this cost as much as possible;
■ The de-optimisation cost exhibits strong variations (greater than ±20 % for more than 54 %
of the days). Therefore, the market design should take this variation into account to
help participants reap opportunities (e.g., by allowing short-term exchanges);
■ The seasonality of the de-optimisation cost is relatively poor over the studied timeframe.
Therefore, it does not seem possible to select a particular period and then to
generalise the result of this period over the whole timeframe. Hence, future studies
on the de-optimisation cost have to consider a large number of days to both give
accurate results and help identify potential long-term seasonalities;
■ For ED F Producer, the de-optimisation cost is sensitive to three majorparameters'.
D The expected marginal costs of reserves, there is a strong link between the marginal
costs of reserves and the de-optimisation cost. Therefore, the marginal costs can
be used as a proxy to estimate the de-optimisation cost;
D The primary reserve demand: because of the capabilities of ED F Producer’s
portfolio, primary reserve demand clearly affects the de-optimisation cost,
whereas the influence of secondary reserve is not as marked;
5.2 CONTRIBUTIONS TO KNOW LEDGE 245
n The quantity of reserves provided by hydro units, when hydro cannot be used
extensively to provide reserves for some critical time steps, the de-optimisation
cost tends to be high. However, this trend is due to a somewhat simplistic
modelling of hydro units in the optimisation algorithm that does not reflect the
actual cost of hydro units.
■ The demandfor reserves can be the most binding constraint. It may thus be useful to develop
storage or load control strategies in order to allow a conversion of the cheaper
energy into reserves;
■ The reserve constraint may not be binding at all’ so the security of the system can be
increased at very low cost by increasing the demand for reserves during periods
when the marginal costs of reserves are low;
■ The de-optimisation cost due to time control represents tens of millions of euros peryear in France.
The value of such an expenditure needs to be proven.
Third, this thesis has proposed a comprehensive and structured analysis of the
features of a market for ancillary sendees. To improve current market designs and to
balance the information sent to participants, a fully independent body (e.g., the regulator)
should describe and justify the methods adopted to solve the design issues listed below:
>4 Nominating the entity responsible of procurement (ERAP): the system operator is the best
candidate since it is a natural buyer of AS;
'T Matching supply and demand: long-term supply should be ensured to have access in the
long run to sufficient AS capabilities at the appropriate locations. In addition, reduction
of demand for ancillary sendees should be pursued. Furthermore, an efficient short
term procurement mechanism is also essential;
T Choosing the relevantprocurement methods to contract AS provision: five methods have been
listed. Since none of them is perfect, it may be desirable that the ERAP selects a few of
them and uses them concurrently;
T Defining the structures of offers and payments to remove price distortions: six possible
structures of offer and payment have been proposed. The optimal choice amongst these
components depends on the cost structure of the ancillary sendees considered;
T Organising the market clearing procedure to fix the prices and amounts of AS transactions
between participants: Several choices have to be made:
■ Two structural arrangements are possible: centralised unit commitment (pool) or

decentralised commitment (exchange);
■ The auction defines how the offer and the demand for AS are submitted and
matched. Classical sealed-bid auctions are preferred in practice, even if such
auctions are not Pareto efficient;
* Scoring, which classifies AS offers, is very difficult because of the multi-dimension

feature of AS;
■ Coordination between the clearings of the various AS markets is essential. In particular,

simultaneous clearing is the most efficient solution, but it is harder to implement in
decentralised systems and is computationally very demanding;
■ The settlement rule organises the payment of the AS providers selected in the auction.
Non-remunerated procurement and administrative prices have to be avoided. The
best settlement rule is still under debate (e.g., discriminatory or uniform price, zonal
or nodal price, etc...);
■ The timing of the market has to be chosen carefully. Higher frequencies are likely to
increase the efficiency o f the market, but also the potential for abuse of dominance
and the operating cost.
"C Ptice caps of any kind should be avoided as much as possible. An appropriate demand
curve is much more desirable;
C Providing proper incentives to stimulate appropriate behaviours by participants, besides

intangible incentives (e.g., the fear of blackouts) and competition:
■ The allocation of SS costs should ideally be performed with three components: (a) the
fixed cost of SS through the transmission tariff, (b) the variable AS reservation costs
through an allocation based on the potential utilisation of the users, and (c) the
actual utilisation of SS through to the users that make the actual use o f these
services necessary;
■ Transmission of data', sufficient and balanced information has to be given to the

market participants to foster competition;
■ Monitoring of the participants in markets for AS: the SO usually performs the
technical monitoring and incentives have to be given to encourage self-monitoring
and transparency. The regulator and the anti-trust body are usually in charge of the
economic monitoring;
5.3 SHORT-TERM EVOLUTIONS DESIRABLE IN FRANCE 247
■ 'Penalties and rewards', to be effective, the monitoring should be complemented by

penalties and rewards of various natures, structures, intensities and volatilities.
C Assessing the procurement method to improve it. The procurement method should fulfil three
goals: effectiveness, minimisation of running cost and economic efficiency. Note that
current AS prices are often obtained with the help of acceptable behaviours of the
participants or regulatory decisions, but not from incentive to compete.
5 .3 Short-Term E volutions D esirable in F rance
By applying the assessment framework to the French power system, various short-term
evolutions become obvious candidates for improving current arrangements (the thesis
discusses in more details the issues related to the following propositions):
C Delivery of ancillary services'.

■ Needsfor system services', more incentives should be developed to reduce the long-term
needs for system sendees. Furthermore, the value of time control should be
explored in further details;
■ Quality of ancillary services', technical AS requirements should facilitate the entrance of
non-conventional participants, for example by allowing non-symmetrical
capabilities, which would help wind farms or loads participate in the market;
■ Quantity of andllay services: the French transmission system operator or the French
regulator should explain in details the method used to calculate the requirements for
frequency control res elves. In addition, an elastic demand would be desirable to buy
more ancillary sendees when they are cheaper;
■ Location of ancillary services', exchanges between neighbouring countries should be
implemented in order to increase the trade-off possibilities of participants.
However, such a system requires a co-optimisation with energy and transmission,
which can be complex in a decentralised trading mechanism such as the one used in
continental Europe.
C Cost of ancillary services', work on this topic should be performed individually by AS
providers;
Procurement ofsystem services'.
■ Procurement methods: since it has been proven that the cost of ancillary services has a
strong daily variation in France, short-term procurement methods are desirable to
complement the current system of long-term bilateral contracts. This would help
increase the global welfare;
■ Structures of offers and payments: symmetrical prices should be removed. The voltage
control availability payment (i.e., in €/M var/h) should be a function of the distance
from the unity power factor;
■ Settlement ride: the energy utilisation of the secondary frequency control should be
indexed on the price of balancing mechanism since the objectives of these two
controls are similar.
5 .4 S u g g e s tio n s for Future Work
This thesis has left open various issues. First, this thesis focused on the description of an
assessment method. Therefore, the developed framework will be particularly useful if it is
applied to actual systems (the thesis applied the framework only partially to actual systems).
Furthermore, the structures of systems vary over time. Therefore, the proposed assessment
should be applied regularly on power systems to improve current arrangements, balance information and
share goodpractices. Such an assessment would be particularly interesting for systems that are
likely to benefit from a common market for AS in the future, such as Belgium, France,
Germany, Luxembourg and The Netherlands. For example, a dedicated taskforce funded by
regulators could be set up. Note that particular attention should be brought to the operating
cost of markets, which is usually ignored in the restructuring process.
Second, defining appropriate needs for ancillary services is limited by a theoretical

barrier well-known since the early age of power system: the value of a continuous supply of
electricity. Furthermore, die value of ancillary services should also consider power quality
and the power system utilisation. It is thus essential to improve the knowledge of the value ofsystem
sendees to enhance markets for ancillary services by introducing elasticity.
Third, long-term procurement of ancillary services is currently achieved through

connecting conditions imposed on new generating units. A more efficient system could be
developed to send appropriate signals and trigger optimal investments in A S capabilities. In particular,
such signals are essential for AS providers to develop (or not) alternative means of AS
provision.
5.4 SUGGESTIONS FOR FUTURE WORK 249
Fourth, the limit between negotiable and non-negotiable products is not clear. For
example, stability is not remunerated in some systems, whereas some units are useful for
stability and others do not contribute to it. Therefore, it seems important to draw the limit
between negotiable and non-negotiableproducts and to evaluate the price of potential new sendees.
Fifth, the exchange of ancillary sendees across systems is an important feature to help
increase the global welfare. However, such an arrangement is challenging for several
reasons: (a) it necessitates product standardisation across systems; (b) it needs market
standardisation; and (c) it faces the strong engineering challenge to co-ordinate energy,
transmission and ancillary services markets in a decentralised manner. In addition, the best
organisation to tackle this issue is not clear. For example, should a dedicated entity or
informal networks perform such studies?
Sixth, markets for ancillary sendees suffer from structural dominant positions.
Therefore, economic research has to develop methods that guarantee an efficient procurement of
ancillaiy services despite structural issues.
Seventh, the cost of ylS should be better understood. In particular, OTESS could be used
to determine the annual supplemental cost of frequency control reserves (i.e., the cost to
provide one supplemental MW over one year). The study could also be completed with
additional data since August 2007 and improved by better taking into account hydro
constraints. Moreover, AS price forecasting should be improved to better negotiate long
term contracts.
Eight, to make the system as efficient as possible, it is essential to investigate all the
technicalpossibilities to controlfrequency and voltage. In particular, research should be performed to
increase contributions by intermittent generation, loads and storage.
Ninth, the diesis was concentrated on the markets for ancillaiy sendees. However,
other electricity markets, such as enetgy, transmission or C 02 could benefit from similar comprehensive
assessmentframeworks. ■
REFERENCES
Abbott (2001) M. Abbott, "Is the Security of Electricity Supply a Public Good?" The
Electricity Journal\ vol. 14 (7), pp. 31-33, August-September 2001.
Abiri-Jahromi et al, (2007) A. Ah.iri-Jahro.mi, M. Fotuhi-Firuzabad, and E. Abbasi,

"Optimal Scheduling of Spinning Reserve Based on Well-Being Model", IE E E
Transactions on Tower Systems, vol. 22 (4), pp. 2048-2057, November 2007.
Alvarado et al (2003) F. Alvarado, B. Borissov, and L. Kirsch, "Reactive power as an

identifiable ancillary sendee", Laurits R. Christensen Associates, Inc, Madison, WI,
USA, 18 March 2003. Available:
http: / / www.caenergy.com/ downloads / Alvarado_Kirsch_IdentifiableAncillarySer
vice.pdf
Anderson (1999) P. M. Anderson, Power System Protection, New York, NY: Wiley-IEEE
Press, 1999.
Arnott et al. (2003) I. Arnott, G. Chown, IC. Lindstrom, M. Power, A. Bose, O. Gjerde,
R. Morfill, and N. Singh, "Frequency control practices in market environments", in
Proc. 2003 C IG R E /IE E E PES International Symposium. Quality and Security of Electric
Power Delivety Systems, pp. 143-148.
251
252 REFERENCES
Arroyo and Galiana (2005) J. M, Arroyo and F. D. Galiana, "Energy and reserve pricing
in security and network-constrained electricity markets", IE E E Transactions on
Power Systems, vol. 20 (2), pp. 634-643, May 2005.
Bakken and Faanes (1997) B. Bakken and H. Faanes, "Technical and economic aspects of
using a long submarine HVDC connection for frequency control", IE E E
Transactions on Power Systems, vol. 12 (3), pp. 1252-1257, August 1997.
Baldick et al. (2005) R. Baldick, U. Helman, B. F. Hobbs, and R. P. O'Neill, "Design of

efficient generation markets", Proceedings of the IE E E , vol. 93 (11), pp. 1998-2012,
November 2005.
Bastard (1998) P. Bastard, "Fonctionnement et reglage des systemes de transport et de

distribution d’energie electrique", Lecture handout 06923/50T998, Supelec, Gif-
sur-Yvette, France, 1998.
Baughman et ai (1997a) M. L. Baughman, S. N. Siddiqi, and J. W. Zarnikau, "Advanced

pricing in electrical systems. I. Theory", IE E E Transactions on Power Systems, vol. 12
(1), pp. 489-495, February 1997.
Baughman et al (1997b) M. L. Baughman, S. N. Siddiqi, and J. W. Zarnikau, "Advanced

pricing in electrical systems. II. Implications", IE E E Transactions on Power Systems,
vol. 12 (1), pp. 496-502, February 1997.
BDEW (2007a) "TransmissionCode 2007: Netz- und Systemregeln der deutschen

Ubertragungsnetzbetreiber", Bundesverband der Energie- und Wasserwirtschaft
e.V, Berlin, Germany, 2007. Available:
http://www.bdew.de/bdew.nsf/id/DE_7B6ERD„NetzCodes_und_Richtlinien/$
file/TransmissionCode2007.pdf
BDEW (2007b) "TransmissionCode 2007 - Anhang D 1: Unterlagen zur Praqualifikation

fur die Erbringung von Primarregelleistung fur die UNB", Bundesverband der
REFERENCES 253
Energie- und Wasserwiitschaft e.V, Berlin, Germany, 2007. Available:

http ://www.bdew.de/bdew.nsf/id/DE_7B6ERD_NetzCodes_und_Richtlinien/$
file/Anhang_Dl_PRL.pdf
BDEW (2007c) "TransmissionCode 2007 - Anhang D 2: Unterlagen zur Praqualifikation

fur die Erbringung von Sekundarregelleistung fur die UNB", Bundesverband der
Energie- und Wasserwiitschaft e.V, Berlin, Germany, 2007. Available:
http://www.bdew.de/bdew.nsf/id/DE_7B6ERD_NetzCodes_und_Richdinien/$
file/Anhang_D2_SRL.pdf
Belgium economic department (2002) Ministere des affaires economiques, Arrete royal
etablissant nn reglement technique pour la gestion du reseau de transport de Peledricite et Pacces a
celui-ci, Moniteur Beige, 2002, pp. 58471-58541. Available:
http://www.ejustice.just.fgov.be/
Bergstrom (1999) T. Bergstrom, "Lecture on Economics. Chapter 17: Auctions", Lecture

slides, Department of Economics, University of California, Santa Barbara, CA,
1999. Available:
h ttp :// econ.ucsb.edu/~tedb/Courses/Ecl00BS06/PPSlides/Chl7.ppt
Bernard et al (1998) J. C. Bernard, T. Mount, and W. Schulze, "Alternative auction

institutions for electric power markets", Agricultural and Resources Economics Review,
vol. 27 (2), pp. 125-131, October 1998. Available: http://agecon.lib.umn.edu/cgi-
bin/pdf_view.pl?paperid:=5668&ftype=.pdf
Berry and Lindgren (1990) D. A. Berry and B. W. Lindgren, Statistics: tbeoiy and method.
Pacific Grove, Calif: Andover : Brooks/Cole ; International Thomson Pub.
Sendees, 1990.
Bertsimas and Tsitsiklis (1997) D. Bertsimas and J. N. Tsitsiklis, Introduction to Linear

Optimisation. Athena Scientific, 1997.
254 REFERENCES
Bhattacharya and Zhong (2001) K. Bhattacharya and J. Zhong, "Reactive power as an

ancillary sendee", IE E E Transactions on Power Systems, vol. 16 (2), pp. 294-300, May
2001 .
Billinton and Allan (1996) R. Billinton and R. N. Allan, Reliability evaluation ofpower systems.
New York; London: Plenum Press, 1996.
Bosch (2006) "Lave-linge Logixx 5 - Notice d'utilisation et d'installation", 0206 fr G1411-

01, Robert Bosch Hausgerate GmbH, Miinchen, Germany, 2006.
Boucher et al (2006) J. Boucher, O. Daxhelet, and Y. Smeers, "An alternative benchmark

of the Internal Electricity Market", in Proceedings of the Joumees de la section
electrotechnique dtt club E E A 2006, pp. 69-95.
Brien (1999) L. Brien, "Why the Ancillary Services Markets in California D on't Work and
What to Do About It", The Electricity Journal, vol. 12 (5), pp. 38-49, June 1999.
British Electricity International (1991) British Electricity International, Modem power station
practice : incorporating modernpower systempractice, 3rd ed. Oxford: Pergamon, 1991.
Bushnell and Oren (1994) J. B. Bushnell and S. S. Oren, "Bidder cost revelation in
electric power auctions", Journal of Regulatory Economics, vol. 6 (1), pp. 5-26,
February 1994.
CAISO (2005a) "Dispatch o f Operating Reserves", Procedure No M-430, Version No

2.1, CAISO, Folsom, CA, USA, 30 September 2005. Available:
http://www.caiso.com/docs/2002/07/30/2002073015073820275.pdf
CAISO (2005b) "Ancillary Service Procurement", Procedure No M-402, Version No 6.1,

CAISO, Folsom, CA, USA, 4 November 2005. Available:
http://www.caiso.com/docs/1998/12/02/1998120218202714536.pdf
REFERENCES 255
Cali et al (2001) A. Cali, S. Conti, and G. Tina, "Ancillary sendees in deregulated

electricity power industry: a structured comparison", in Proceeding of the Fifth
LASTED International Conference on Power and Energy Systems 2001, pp. 161-166.
CER (2003) "Self-Commitment versus Centralised Commitment", Commission for

Energy Regulation (CER), Dublin, Ireland, 22 December 2003. Available:
http://www.cer.ie/CERDocs/cer03299.pdf
Chao and Wilson (2002) H.-P. Chao and R. Wilson, "Multi-Dimensional Procurement
Auctions for Power Reserves: Robust Incentive-Compatible Scoring and
Settlement Rules", Journal of Kegulatoy Economics, vol. 22 (2), pp. 161-183,
September 2002.
Chicco and Gross (2004) G. Chicco and G. Gross, "Competitive acquisition of

prioritizable capacity-based ancillary sendees", IE E E Transactions on Power Systems,
vol. 19 (1), pp. 569-576, February 2004.
Cigre Task Force (2001) Task Force 38.05.07, "Methods and tools for costing ancillary
services", Report No 190, Conseil International des Grands Reseaux Electriques
(Cigre), Paris, France, June 2001.
Cohen (1978) G. Cohen, "Optimization by decomposition and coordination: A unified

approach", IE E E Transactions on Automatic Control, vol. 23 (2), pp. 222-232, April
1978.
Corsi et al (1995) S. Corsi, P. Marannino, N. Losignore, G. Moreschini, and G. Piccini,

"Coordination between the reactive power scheduling function and the
hierarchical voltage control of the EHV ENEL system", IE E E Transactions on
Power Systems, vol. 10 (2), pp. 686-694, May 1995.
Cramton and Stoft (2005) P. Cramton and S. Stoft, "A Capacity Market that Makes
Sense", The Electricity Journal, vol. 18 (7), pp. 43-54, August-September 2005.
256 REFERENCES
CRE (2008) "Observatoire des marches de l’electricite et du gaz - 4eme trimestre 2007",
Commission de Regulation de l'Energie (CRE), Paris, France, 29 February 2008.
Available:
http://w w w .cre.fr/fr/content/ download/5342/118109/file/dmspObsT407.pdf
Da Silva et al (2001) E. Da Silva, J. Hedgecock, J. C. Mello, and J. C. F. Da Luz,

"Practical Cost-Based Approach for the Voltage Ancillary Service", IE E E
Transactions on Power Systems, vol. 16 (4), pp. 806-812, November 2001.
DePillis (2006) M. S. DePillis, "The New Ancillary Services Markets of New England",
presented at the Midwest ISO Ancillary Services Roundtable, Carmel, IN, USA,
26-27 April 2006. Available:
http://www.midwestmarket.org/publish/Document/469a41_10a26fa6cle_-
74c30a48324a/7%20-
%20Midwest%20ISO%20Ancillary%20Services%20Roundtable_ISO-
NE.pdfPaction^downloadiS^property^ Attachment
D G EM P/D ID EM E (2003) "Les couts de reference de la production electrique",

D G EM P/D ID EM E/SD 6, Ministere de l'Economie, des Finances et de
l'lndustrie, Paris, France 2003. Available:
h ttp ://www.industrie.gouv.fr/energie/electric/se_ele_al0.htm
Doherty et al (2005) R. Doherty, G. Lalor, and M. O'Malley, "Frequency control in

competitive electricity market dispatch", IE E E Transactions on Power Systems, vol. 20
(3), pp. 1588-1596, August 2005.
DTe (2005) "Netcode", Directie Toezicht Energie, The Hague, The Netherlands, 1 July
2005. Available:
http://www.dte.nl/images/NETCODE%20per%2001072005_tcm7-74895.pdf
ED F (2004) "Conditions generales relatives a l'acces et a l'utilisation du reseau electrique

ainsi qua la fourniture d'electricite pour les clients soumis au code des marches
REFERENCES 257
publics ayant souscrit un contrat unique", Electricite de France (EDF), Paris,

France, 10 November 2004.
EDF (2006) Quly). "Une electricite de qualite". [Online]. Available:

http: / / collectivite.edf. fr/
Electricity Commission of New Zealand (2004) "Recommended Approach Regarding

Dispatch of Generators for Voltage Support", Electricity Commission, Wellington,
New Zealand, December 2004. Available: http:/ /
www.electricitycomtxiission.govt.nz/pdfs / opdev/comqual/consultationpdfs / volta
ge-support-consultation.pdf
Electricity Commission of New Zealand (2005) (1 September). "Electricity Governance

Rules". [Online]. Available:
http://www.electricitycommission.govt.nz/rulesandregs/rules/
Elizondo et al (2006) M. Elizondo, M. Ilic, and P. Mercado, "Determining the cost of

dynamic control capacity for improving system efficiency", presented at the IEEE
PES General Meeting, Montreal, Canada, 2006.
E.ON (2005) "Procurement of control power and energy in Germany", E.ON Netz
GmbH, Bayreuth, Germany, 07 July 2005. Available: http://www.eon-
netz.com/Ressources / downloads/procurement_controlpower_energy.pdf
ETSO tariffs task force (2006) "ETSO Overview of transmission tariffs in Europe:
Synthesis 2005", European Transmission System Operators (ETSO), Brussels,
Belgium, June 2006. Available: http://www.etso-
net.org/upload/documents/benchmarking2005.pdf
Ernu (2007) X. Ernu, "Introduction a la gestion optimisee de la production", presented at

EDF R&D seminar Gestion optimisee de la production, Clamart, France, 2007.
258 REFERENCES
Eurelectric (1997) Eurelectric, "Voltage control - Reactive energy", Report No

03005Ren9712, Union of the Electricity Industry - Eurelectric, Brussels, Belgium,
March 1997. Available:
http://w w w .eurelectric.org/Download/Download.aspxPDocumentID=5622
Eurelectric (2000) "Connection rules for generation and management of ancillary

services", Report No 2000-130-0003, Union of the Electricity Industry -
Eurelectric, Brussels, Belgium, May 2000. Available:
http://www.eurelectric.org/Download/Download.aspx?DocumentFileID=4604
Eurelectric (2004) "Ancillary Services Unbundling Electricity Products —An E m e r g in g
Market", Report No 2003-150-0007, Union of the Electricity Industry -

Eurelectric, Brussels, Belgium, February 2004. Available:
http://www.eurelectric.org/Download/Download.aspxPDocumentID-13934
Fourment et al. (1995) C. Fourment, H. Lefebvre, and J. M. Tesseron, "Power system

requirements and associated services at generator level in France", in Proceedings of
the IE E Colloquium on Frequency Control Capability of Generating Plant, London, UK, 8
February 1995.
French economic and industry department (2003) Ministere de Peconomie, des finances
et de l’industrie, Arrete du 4 juillet 2003 relatif auxprescriptions techniques de conception et
de fonctionnement pour le raccordement au reseau public de transport d’une installation de
production d’energie electrique, Journal Officiel de la Republique francaise n° 201, 2003,
pp. 14896-14902. Available:
http://www.legifrance.gouv.fr/WAspad/UnTexteDeJorf?numjo=INDI0301719A
French economic and industry department (2005) Ministere de Peconomie, des finances
et de Pindustrie, Decision du 23 septembre 2005 approuvant les tarifs d’utilisation des
reseauxpublics de transport et de distribution d''electricite, Journal Officiel de la Republique
frangaise n° 233, 2005, pp. 15888-15917. Available:
http://www.legifrance.gouv.fr/WAspad/UnTexteDeJorf?numjo=INDI0505749S
REFERENCES 259
Frontier Economics and Consentec (2005) "Benefits and practical steps towards the
integration of intraday electricity markets and balancing mechanisms", Frontier
Economics Ltd, London, U.K., December 2005. Available:
http://europa.eu.int/comm/energy/electricity/publications/doc/frontier_consen
tec_balancing_dec_2005.pdf
Fuller (1976) W. A. Fuller, Introduction to statistical time series. New York ; London: Wiley,
1976.
Galiana et al. (2005) F. D. Galiana, F. Bouffard, J. M. Arroyo, and J. F. Restrepo,

"Scheduling and pricing of coupled energy and primary, secondary, and tertiary
reserves", Proceedings of the IE E E , vol. 93 (11), pp. 1970-1983, November 2005.
Garcia (2005) M. P. Garcia-Alajarin, "Forecasting system imbalance volumes and analysis

of unusual events in competitive electricity markets", PhD thesis, School of
Electrical and Electronic Engineering, The University of Manchester, Manchester,
UK, 2005. Available:
http://w w w .eee.manchester.ac.uk/ research/groups/eeps/docs/garcia_JPhD_2005
.pdf
Gautier (1990) A. Gautier, "Le fonctionnement des centrales a eau sous pression -
Adaptation de la centrale aux besoins du reseau", Institut National des Sciences et
Techniques Nucleaires (INSTN), Saclay, France, 1990 (approximate date).
Goel et al. (2004) L. Goel, Z. Song, and P. Wang, "Impact of Transmission Line Failures
on the Cost/Risk-Based Spinning Reserve Allocation", International Journal of
Emerging Electric Power Systems, vol. 1 (1), May 2004.
Grainger and Stevenson (1994) J. J. Grainger and W. D. Stevenson, Power system analysis.
New York ; London: McGraw-Hill, 1994.
260 REFERENCES
Gross et al (2002) G. Gross, S. Tao, E. Bompard, and G. Chicco, "Unbundled Reactive

Support Service: Key Characteristics and Dominant Cost Component", IE E E
Transactions on Power Systems, vol. 17 (2), pp. 283-289, May 2002.
Haas et al (2006) R. Haas, J.-M. Glachant, N. Keseric, and Y. Perez, "Competition in the
Continental European Electricity Market: Despair or Work in Progress?" in
Electricity market reform: an international perspective, Elsevier global energy polity and
economics series, F. P. Sioshansi and W. Pfaffenberger, Amsterdam: Elsevier, 2006,
pp. 265-315.
Hamilton (1994) J. D. Hamilton, Time series analysis. Princeton, N.J.: Princeton University
Press, 1994.
Hao and Papalexopoulos (1997) S. Hao and A. Papalexopoulos, "Reactive power pricing
and management", IE E E Transactions on Power Systems, vol. 12 (1), pp. 95-104,
February 1997.
Hingorani and Gyugyi (2000) N. G. Hingorani and L. Gyugyi, Understanding FACTS:

concepts and technology offlexible A C transmission systems. New York, N.Y.: IEEE Press,
2000 .
Hirst and Kirby (1996) E. Hirst and B. Kirby, "Costs for electric-power ancillary
services", The Electricity Journal, vol. 9 (10), pp. 26-30, December 1996.
Hirst and Kirby (1997) E. Hirst and B. Kirby, "Creating competitive markets for ancillary
services", ORNL/CON-448, Oak Ridge National Laboratory, Oak Ridge, TN,
USA, October 1997. Available:
http://www.ornl.gov/sci/btc/apps/Restructuring/con448.pdf
Hirst and Kirby (1998) E. Hirst and B. Kirby, "Unbundling Generation and
Transmission Services for Competitive Electricity Markets: Ancillary Services",
ORNL/CON-454, National Regulatory Research Institute, Colombus, OH, USA,
REFERENCES 261
January 1998. Available:

http: / /\vww.ornl.gov/sci/btc/apps/Restructuring/ con454.pdf
Hogan (2006) W. W. Hogan, "Reliability and scarcity pricing: operating reserve demand
curves (continued)", Harvard University, Cambridge, MA, USA, 3 March 2006.
Available: h ttp ://www.hks.haivard.edu/hepg/Papers/Hogan_hepg_030206.pdf
Hoinkis and Lindner (2007) J. Hoinkis and E. Lindner, Chemie fur Ingenieure, 13 ed.
Weinheim: Wiley-VCH, 2007.
IEA (2005) (June). "IEA Energy Statistics". [Online]. Available:

h ttp ://www.iea.org/Textbase/stats/prodresult.asp?PRODUCT=Electricity/Heat
IEEE Std 1159-1995 (1995) IEEE Working Group on Monitoring Electric Power
Quality, IE E E Std 1159-1995 - IE E E Recommended Practice for Monitoring Electric
Power Quality. New York: The Institute of Electrical and Electronics Engineers,
Inc., 1995.
IMF (2006) "Financial Soundness Indicators: Compilation Guide", ISBN/ISSN 978-1-

58906-385-3, International Monetary Fund (IMF), Washington, DC, USA, 4 April
2006. Available: http://www.imf.org/external/pubs/ft/fsi/guide/2006/index.htm
Ingleson and Ellis (2005) J. W. Ingleson and D. M. Ellis, "Tracking the eastern
interconnection frequency governing characteristic", in Proceedings of the 2005 IE E E
Power Engineering Society General Meeting, pp. 2484 - 2489.
Jaggy and Longley (1991) E. Jaggy and A. P. Longley, "Implementation o f an AGC-

algorithm in different power systems and on different control levels", in Proceedings
of the 1991IE E International Conference on Power System Monitoring and Control, pp. 162-
167.
262 REFERENCES
Jaleeli et al. (1992) N. Jaleeli, L. S. VanSlyck, D. N. Ewart, L. H. Fink, and A. G.

Hoffmann, "Understanding automatic generation control", IE E E Transactions on
Power Systems, vol. 7 (3), pp. 1106-1122, August 1992.
Johannet (1997) P. Johannet, "Lignes aeriennes : chutes de tension", Techniques de

I'Ingenieur (D 4 438), February 1997.
Joskow (2006) P. Joskow, "Introduction to Electricity Sector Liberalization: Lessons

Learned from Cross-Country Studies", in Electricity market reform: an international
perspective, Elsevier global energy policy and economics series, F. P. Sioshansi and W.
Pfaffenberger, Amsterdam: Elsevier, 2006, pp. 1-32.
Joskow and Tirole (2004) P. joskow and J. Tirole, "Reliability and competitive electricity
markets", Report No CSEM WP 129, University of California Energy Institute
(UCEI), Berkeley, CA, USA, April 2004. Available:
http://repositories.cdlib.org/ ucei/csem/CSEMWP-129/
Kariuki and Allan (1996) K. K. Kariuki and R. N. Allan, "Evaluation of reliability worth
and value of lost load", IE E Proceedings: Generation, Transmission and Distribution, vol.
143 (2), pp. 171-180, March 1996.
Keller and Franken (2006) K. Keller and B. F. C. Franken, "Quality of supply and market
regulation; survey within Europe", 30630505-TDC 06-59095A, KEMA
Consulting, Arnhem, The Netherlands, 7 December 2006. Available:
h ttp ://www.leonardo-energy.org/drupal/files/2006/06-59095A.pdf?download
Kirby and Hirst (1996) B. Kirby and E. Hirst, "Ancillary-Service Costs for 12 U.S.
Electric Utilities", ORNL/CON-427, Oak Ridge National Laboratory, Oak Ridge,
TN, USA, April 1996. Available:
http://w w w .ornl.gov/ sci/btc / apps/Restructuring/con427.pdf
REFERENCES 263
Kirby and Hirst (2000) B. Kirby and E. Hirst, "Customer-specific metrics for the
regulation and load-following ancillary sendees", Report No ORNL/CON-474,
Oak Ridge National Laboratory, Oak Ridge, TN, USA, January 2000. Available:
h ttp ://certs.lbl.gov/pdf/ornl-con-474.pdf
Kirby and Hirst (2003) B. Kirby and E. Hirst, "New England Demand Response
Initiative: Technical issues related to retail-load provision of ancillary sendees", 10
February 2003. Available:
http://www.raponline.org/Pubs/NEDRI/AncillaiyBackground.pdf
Kirschen (2002) D. S. Kirschen, "Power system security", Power Engineering Journal, vol. 16
(5), pp. 241-248, October 2002.
Kirschen and Jayaweera (2007) D. S. Kirschen and D. Jayaweera, "A Comparison of

Risk-Based and Deterministic Security Assessments", PET Generation, Transmission
<&Distribution, vol. 1 (4), pp. 527-533, July 2007.
Kirschen and Strbac (2004a) D. S. Kirschen and G. Strbac, Fundamentals ofpower system
economics, 1st ed., Chichester: Wiley, 2004.
Kirschen and Strbac (2004b) D. S. Kirschen and G. Strbac, "Why Investments D o Not
Prevent Blackouts", The Electricity Journal, vol. 17 (2), pp. 29-36, March 2004.
Klemperer (1999) P. Klemperer, "Auction Theory: A Guide to the Literature", Journal of

Economic Surveys, vol. 13, pp. 227-286,1999.
Klemperer (2002) P. Klemperer, "What really matters in auction design", The Journal of
Economic Perspectives, vol. 16 (1), pp. 169-189, winter 2002.
Kluge (2006) T. Kluge, "Pricing Swing Options and other Electricity Derivatives", PhD
Thesis, Mathematical Institute, St Hugh's College, University of Oxford, Oxford,
UK, 2006. Available: http://eprints.maths.ox.ac.uk/246/01/kluge.pdf
264 REFERENCES
Kundur (1994) P. Kundur, Power system stability and control. New York; London: McGraw-
Hill, 1994.
Lasdon (1968) L. S. Las don, "Duality and Decomposition in Mathematical

Programming", IE E E Transactions on Systems Science and Cybernetics, vol. SSC-4 (2),
pp. 86-100, July 1968.
Lemarechal (2004) C. Lemarechal, "Saving Energy", in ERCIM News, vol. 56, January
2004, p. 43. Available:
http://www.ercim.org/publication/Ercim_News/enw56/EN56.pdf
Lin et al. (2005) X. J. Lin, C. W. Yu, and C. Y. Chung, "Pricing of reactive support
ancillary services", IE E Proceedings-Generation, Transmission and Distribution, vol. 152
(5), pp. 616-622, September 2005.
Liu et al (2006) J. D. Liu, T. T. De, and K. L. Lo, "An empirical method o f dynamic
oligopoly behavior analysis in electricity markets", IE E E Transactions on Power
Systems, vol. 21 (2), pp. 499-506, May 2006.
Luh et al. (2006) P. B. Luh, W. E. Blankson, Y. Chen, J. H. Yan, G. A. Stern, S.-C. Chang,
and F.Zhao, "Payment cost minimization auction for deregulated electricity
markets using surrogate optimisation", IE E E Transactions on Power Systems, vol. 21
(2), pp. 568-578, May 2006.
Ma et al. (2003) X. Ma, D. I. Sun, and K. W. Cheung, "Evolution toward standardised

market design", IEEE Transactions on Power Systems, vol. 18 (2), pp. 460-469,
May 2003.
Margotin et al. (2006) T. Margotin, P. Juston, M. P. Houry, G. Morales, P. Bertolini, and

D. Burnel, "Reactive power support on the French power system: generation units
performance monitoring", Proceedings of the 2006 IE E E Mediterranean Electrotechnical
Conference (MEEECON), pp. 906-909.
REFERENCES 265
Mas-Colell et al (1995) A. Mas-Colell, M. D. Whinston, and J. R. Green, Microeconomic

Theoiy. USA: Oxford University Press, 1995.
Mathworks (2007) "Online documentation - Percentiles". [Online] Available:

h ttp ://www.mathworks.com/access/helpdesk/help/toolbox/stats/index.html?/a
ccess/helpdesk/help/toolbox/stats/prctile.html
Meeus and Belmans (2008) L. Meeus and R. Belmans, "Electricity market integration in
Europe", Rome E tijdschrift, vol. 124 (1), March 2008.
Mickey (2006) J. Mickey, "Competitive Ancillary Sendees Market in ERCOT", presented

at the Midwest ISO Ancillary Sendees Roundtable, Carmel, IN, USA, 26-27 April
2006.
National Grid (2007) "Short-term operating reseive: general description of the sendee",
National Grid, Warwick, U.K., 14 December 2007. Available:
http: / / www. nationalgrid. com/ N R / rdonlyres / 72D4386B-2027-474C-B 281 -
2384F5B21A5E/22123/STOR_Sentice_Des crip tionJFinal.pdf
National Grid (2008a) "The Grid Code", Issue 3, Revision 25, National Grid, Warwick,
U.K., 1 February 2008. Available:
http:/ / www. nationalgrid. com /N R / rdonlyres /1403354E-3 6F2-4010-A7 2C-
F36862ElFF3E/23088/zFullGBGridCodeI3R25.pdf
National Grid (2008b) "Firm Frequency Response Market Information for Tenders for
April 2008", National Grid, Warwick, U.K., 2008. Available:
http://www.nationalgtid.eom/NR/rdonlyres/F4649A84-1580-4E15-B411-
0FAE9159608A/23984/FFR_Market_Info_Apr_2008_vl .pdf
National Grid (2008c) "Reactive Power Market. Obligatory and Enhanced Reactive
Power Sendees: Invitation to Tender and Guidance Notes for the Completion of
Tenders for Reactive Power Market Agreements Commencing 1 October 2008",
266 REFERENCES
National Grid, Warwick, U.K., March 2008. Available:

http://www.nationalgrid.com/NR/rdonlyres/lEBD3ClC-02Al-4124-B705-
FD7 660650BD7/24227/ Guidance_Notes.pdf
NEMMCO (2001) "Guide to Ancillary Services in the National Electricity Market",

Version 1.0, National Electricity Market Management Company Limited, Australia,
24 August 2001. Available: http://www.nemmco.com.au/ancillary_services/160-
0056.pdf
NEMMCO (2004) "Market Ancillary Service Specification", Version 1.5 Final, National
Electricity Market Management Company Limited, Australia, 27 February 2004.
Available: http://www.nemmco,com.au/anciUary_services/160-0163.pdf
NEMMCO (2005) "Operating procedure: glossary", SO_OP2000, National Electricity

Market Management Company Limited, Australia, 20 October 2005. Available:
http://www.nemmco.com.au/powersystemops/so_op2000v011.pdf
NERC (2004) "Operating manual", North American Electric Reliability Corporation,

Princeton, NJ, USA, 15 June 2004.
NERC (2005) "Frequency Response Standard Whitepaper", North American Electric

Reliability Corporation, Princeton, NJ, USA, 17 January 2005. Available:
ftp://www.nerc.com/pub/sys/all_updl/oc/rs/Frequency_Response_White_Pape
r.pdf
NERC (2006) "Operating manual", North American Electric Reliability Corporation,

Princeton, NJ, USA, 15 June 2006. Available:
ftp://www.nerc.com/pub/sys/all_updl/ o c/opm an/opman_061506.pdf
Newbery et al (2003) D. Newbery, N.-H. von der Fehr, and E. van Damme,
"Transparency and monitoring o f security and adequacy of supply", Directie
REFERENCES 267
Toezicht Energie (DTe), The Hague, The Netherlands, 25 August 2003. Available:
http://www.dte.nl/images/12_12720_tciTi7-3712.pdf
Newbery et al. (2004) D. Newbery, R. Green, K. Neuhoff, and P. Twomey, "A Review of
the Monitoring of Market Power: The Possible Roles of TSOs in Monitoring for
Market Power Issues in Congested Transmission Systems", European
Transmission System Operators (ETSO), Brussels, Belgium, November 2004.
Available: http://www.etso-
net.org/upload/documents/ETSO%20Market%20Power%20£inal.pdf
Nobile et al, (2001) E. Nobile, A. Bose, and K. Tomsovic, "Feasibility of a Bilateral

Market for Load Following", IE E E Transactions on Power Systems, vol. 16 (4), pp.
782-787, November 2001.
Nordel (2004) "Nordisk Regelsamling (Nordic Grid Code)", Nordel, Helsinki, Finland,
18 June 2004. Available:
http://www.statnett.no/Resources/Filer/Dokumenter/Div.%202004/Nordisk%
20regelsamling%20rev%20161104.pdf
Numerical Algorithms Group (2007) (October). "NAG Add-Ins for Excel". [Online]
Available: http://www.nag.co.uk/downloads/aedownloads.asp
NYISO (2008) NYISO Customer Technical Services, "Ancillary sendees manual",

Manual 2, Version 3.10, New York Independent System Operator (NYISO),
Schenectady, NY, USA, 6 March 2008. Available:
http://www.nyiso.com/public/webdocs/documents/manuals/operations/ancser
v.pdf
NZIER (2005) "Market Design Report: Initial Stock-take Paper Report to the Electricity
Commission", NZIER, Wellington, New Zealand, August 2005. Available:
http: / / www.electricitycommission.govt.nz/pdfs/opdev/wholesale/market-
design/Initial-Stocktake.pdf
268 REFERENCES
OMEL (2005) (September). "Market results". [Online]. Available:

http://www.om el.es/frames/en/resultados/resultados_index.htm
O ’Neill et al (2006) R. O'Neill, U. Helman, B. F. Hobbs, and R. Baldick, "Independent

System Operators in the USA: History, Lessons Learned, and Prospects", in
Electricity market reform: an international perspective, Elsevier global energy polity and
economics series, F. P. Sioshansi and W. Pfaffenberger, Amsterdam: Elsevier, 2006,
pp. 479-528.
Open University (1988) Open University, Applications o f Probability Course Team,

Applications ofprobability. Unit 15: Time series. Milton Keynes: Open University Press,
1988.
Oren (2001) S. Oren, "Design of ancillary service markets", presented at the 34th Annual
Hawaii International Conference on System Sciences, Maui, HI, USA, 3-6 January
2001.
Ortega-Vazquez and Kirschen (2007) M. A. Ortega-Vazquez and D. S, Kirschen,

"Optimizing the Spinning Reserve Requirements Using a Cost/Benefit Analysis",
IE E E Transactions on Power Systems, vol. 22 (1), pp. 24-33, February 2007.
Oxford Dictionary (2001) 0>ford Paperback Dictionary, Thesaurus and Wordpower Guide. New
York: Oxford University Press Inc., 2001.
Papadogiannis and Hatziargyriou (2004) K. A. Papadogiannis and N. D. Hatziargyriou,

"Optimal allocation o f primary reserve services in energy markets", IE E E
Transactions on Power Systems, vol. 19 (1), pp. 652-659, February 2004.
Paul et al. (1990) J. P. Paul, C. Corroyer, P. Jeannel, J. M. Tesseron, F. Maury, and A.

Torra, "Improvements in the organisation of secondary voltage control in France",
Proc. 33rd International Conference on Earge High Voltage Electric Systems (CIGRE)
Session, Paper N o 38/39-03, pp. 1-8, 26 August - 1 September 1990.
REFERENCES 269
Pekec and Rothkopf (2004) A. Pekec and M. H. Rothkopf, "Combinatorial Auction

Design", Management Science, vol. 49 (11), pp. 1485-1503, November 2003.
Peradotto (2001) A. Peradotto, "Data mining : etat de l’art", Internal report No HE-
72/2001/016/A, Electricite de France Research and Development (EDF R&D),
Clamart, France, 24 July 2001.
Perrot and Molinier (2005) D. Perrot and P. Molinier, "Surveillance des marches de
gros", Internal report No H-Rl 6-2005-01546-FR, Electricite de France Research
and Development (EDF R&D), Clamart, France, July 2005.
Pignon (2003) V. Pignon, IS harmonisation des methodes de tarification du transport de I'electricite:

une analyse economique du marche unique europeen, PhD Thesis, Sciences economiques,
Paris I Pantheon-Sorbonne, Paris, France, 2003. Available:
h ttp :// www.grjm.net/index.php?option=com_content&task—view&id=56&Itemi
d=65
PJM (2005a) "PJM eMKT Users Guide", PJM Interconnection, Norristown, PA, USA,
June 2005. Available:http://www.pjm.com/etools/downloads/emkt/ts-
userguide.pdf
PJM (2005b) "PJM Open Access Transmission Tariff', PJM Interconnection,

Norristown, PA, USA, 22 July 2005. Available:
http://www.pjm.com/documents/downloads/agreements/tariff.pdf
PJM (2006) "Manual 35: Definitions and Acronyms", Revision 13, PJM Interconnection,
Norristown, PA, USA, 1 November 2006. Available:
http://w w vrpjm .com /contributions/pjm-manuals/pdf/m35.pdf
PJM (2007) "Manual 12: Balancing Operations", Revision 16, PJM Interconnection,
Norristown, PA, USA, 01 November 2007. Available:
http: / / www.pjm.com/ contributions/pjm-manuals/pdf/ml 2.pdf
270 REFERENCES
PJM (2008) "Manual 11: Scheduling Operations", Revision 34, PJM Interconnection,
Norristown, PA, USA, 21 February 2008. Available:
http://www.pjm .com /contributions/pjm -m anuals/pdf/m ll.pdf
Powernext (2008) Powernext, "Powernext Futures - Monthly statistics", Powernext SA,

Paris, France, 30 June 2008. Available:
http://www.powernext.fr/modules/PwnDl/download/files/fra/PowernexpJFutu
res06-2008.pdf
Proctor (2001) R. L. Proctor, Management ofpower system reserve in a cotnpetitive environment,

PhD Thesis, School of Electrical and Electronic Engineering, The Queen's
University of Belfast, Belfast, U.K., 2001.
Quintana and Bautista (2006) V. Quintana and G. Bautista, "Financial Transmission

Rights in Electricity Markets (Invited Paper)", in Proceedings of the 2006 IE E E
Mediterranean Electrotechnical Conference (M ELECON), pp. 931- 934.
Raineri et al. (2006) R. Raineri, S. Rios, and D. Schiele, "Technical and economic aspects
of ancillary services markets in the electric power industry: an international
comparison", Energy Polity, vol. 34 (13), pp. 1540-1555, September 2006.
Rious (2007) V. Rious, Ee developpement du reseau de transport dans un systeme electrique liberalise,
un probleme de coordination avec la production, PhD Thesis, Faculte Jean Monnet, Paris-
Sud XI and Supelec, Paris, France, 2007. Available:
http://www.grjm.net/index.php?option=com_content&task=view&id=455&Ite
mid=65
Robert and Prestat (2004) G. Robert and B. Prestat, "Conformity evaluation method of
power plants performance for grid stability", in Proceedings of the 2004 IE E E PES
Power Systems Conference and Exposition (PSCE), vol. 2, pp. 1109-1112.
REFERENCES 271
Rothkopf (2007) M. H. Rothkopf, "Thirteen Reasons Why the Vickrey-Clarke-Groves

Process Is N ot Practical", Operations Research, vol. 55 (2), pp. 191—197, March—April
2007.
RTE (2004) "Memento de la surete du systeme electrique", Reseau de Transport

d'Electricite (RTE), La Defense, France, 2004. Available: http://www.rte-
france.com/htm/ fr/mediatheque/telecharge/memento_surete_2004/memento_s
urete_2004_compiet .pdf
RTE (2006) "Conditions generales relatives a l'acces au reseau public de transport.

Consommateur", Reseau de Transport d'Electricite (RTE), La Defense, France,
January 2006. Available: http://www.rte-
france.com /htm /fr / offre/telecharge/Conditions_generales_consommateurs_10_
01_06.pdf
RTE (2007a) (February). "RTE - Clients & acteurs du marche - Vie du systeme .
[Online]. Available: http://www.rte-
france.com /htm /fr/vie/ vie_mecanisme_historique.jsp
RTE (2007b) RTE, "Contrat type de participation aux services systeme", Reseau de
Transport d'Electricite (RTE), La Defense, France, 21 December 2007. Available:
http:/ / www. rte-
france.com /htm /fr/offr e/telecharge/Contrat_services_systeme__trame_type_2007
_12_21.pdf
RTE (2007c) "Dossier de presse : Le bilan electrique francais 2006", Reseau de Transport
de l'Electricite (RTE), La Defense, France, January 2007. Available:
http: / / www.rte-
france.com/htm/ fr/journalistes/telecharge/dossiers/DP_bilanElec_2006.pdf
Saluden (2006) Y. Saluden, "Etude des couts de maintenance de la sous-fonction

mecanique marche en compensateur pour les turbines Francis/Pelton", Internal
272 REFERENCES
report No IH-SSYST-00024-ABPE, Electricite de France Direction Production et

Ingenierie (EDF DPI), Le Bourget du Lac, France, 20 February 2006.
Saporta (1990) G. Saporta, Probability, analyse des donnees et statistique. Paris, France:
Editions Technip, 1990.
Shahidehpour et al. (2002) M. Shahidehpour, H. Yamin, and Z. Li, Market operations in

electricpower systems:forecasting, scheduling and risk management. New York: Institute of
Electrical and Electronics Engineers (IEEE) Wiley-Interscience, 2002.
Singh (1999) H. Singh, "Auctions for ancillary services", Decision Support Systems, vol. 24
(3-4), pp. 183-191, January 1999.
Sioshansi and Pfaffenberger (2006) F. P. Sioshansi and W. Pfaffenberger, Electricity market

reform: an internationalperspective. Amsterdam: Elsevier, 2006.
Spanish industry and energy department (1998) Ministerio de industria y energia,

Ilesolution de 30 dejulio de 1998, de laSecretaria de Estado de Energiay Recursos Minerales,
por la que se aprueba un conjunto de procedimientos de caracter tecnico e instrumental necesarios
para realiyar la adecuada gestion tecnica del sistema ele'ctrico, vol. BOE n° 197, 20053,
1998. Available:
http://www.ree.es/operacion/pdf/po/PO_resol_30jull 998_b.pdf
Spanish industry and energy department (2000) Ministerio de industria y energia,

Resolution de 10 de marô de 2000, de la Secretaria de Estado de Industriay Energia, por la
que se aprueba elprocedimiento de operation del sistema (P.O. - 7.4) «Servitio complementario
de control de tension de la red de transporte», vol. BOE n° 67, 5204, 18 March 2000,
pp. 11330-11346. Available:
http:/ / www.ree.es/operacion/pdf/po/PO „resol_l 0mar2000_correc.pdf
Spanish industry and energy department (2006a) Ministerio de industria y energia,

Resolution de 24 de mayo de 2006, de la Secretaria General de Energia, por la que se aprueban
REFERENCES 273
diversos procedimientos de operation para su adaptation a la nueva normativa electrica, vol.

BOE n° 128, 9521, 2006. Available:
http://www.ree.es/operacion/pdf/po/PO_resolJ24may2006Y7-2.pdf
Spanish industry and energy department (2006b) Ministerio de industria y energia,

Resolution de 13 dejulio 2006, de la Secretaria General de Energia, por la que se aprueba el
procedimiento de operation 1.5 «Establetimiento de la reserva par la regulation frecuentia-
potentia», BOE n° 173, 13170, 2006, pp. 27473-4. Available:
h ttp ://www.ree.es/operacion/pdf/po/RES_P 0_L 5.pdf
Staropoli and Jullien (2006) C. Staropoli and C. Jullien, "Using Laboratory Experiments
to Design Efficient Market Institutions. The case of wholesale markets", presented
at the Journees de la section electrotechnique du club EEA 2006, Gif-sur-Yvette,
France, 15-16 March 2006.
Sterpu (2005) S. Sterpu, "Controle de performances de producteurs independants pour

une meilleure securite des reseaux electriques", PhD Thesis, Laboratoire
d’Electrotechnique de Grenoble (LEG), Institut National Polytechnique de
Grenoble (INPG), Grenoble, France, 2005. Available: http://tel.archives-
ouvertes.fr/docs/00/17/21/64/PDF/These.J5tefanSterpu_final.pdf
Stoft (2002) S. Stoft, Power system economics: designing markets for electricity. Piscataway, N.J.;
New York, N.Y.: IEEE Press: Wiley-Interscience, 2002.
Strbac and Kirschen (2000) G. Strbac and D. S. Kirschen, "Who Should Pay for
Reserve?" The Electricity Journal\ vol. 13 (8), pp. 32-37, October 2000.
SvIC (2005) (December). "Svenska Kraftnat keeps the electricity system in balance".
[Online]. Available: littp ://www.svk.se/web/Page.aspx?id=5216
TenneT (2003) "Samenvatting uit "Summary of the current operating principles of the
UCPTE" met betrekking tot de primaire en secundaire regeling", TenneT,
274 REFERENCES
Arnhem, The Netherlands, December 2003. Available:

http://www.tennet.nl/images/UCTE-regels_tcm41-12792.pdf
TenneT (2006) (January). "Primary reserve". [Online]. Available:

http://www.tennet.org/ english/system_services/primary/
Tesseron (2008) J*-M. Tesseron, "Mission : Surete - Lutter contre les blackouts",
presented at the IEEE France General Meeting, Paris, France, 14 March 2008.
Testud (1991) G. Testud, "Le reglage de la tension. Reglage primaire et domaine de

fonctionnement de l'alternateur", Fiche Echos Systeme n°08, Electricite de France
(EDF) R&D, Clamart, France, September 1991.
Texier-Pauton (2007) R Texier-Pauton, "Donnees et etudes necessaries a la preparation

de la negotiation du contrat services systeme 2008-2010", Internal report
No FC/FDO/COPM AM /2007/036, Electricite de France (EDF) Direction
Optimisation Amont/Aval & Trading (DOAAT), Saint-Denis, France, 11 May
2007.
Thomas (2003) S. Thomas, "The Seven Brothers", Energy Policy, vol. 31 (5), pp. 393-403,
April 2003.
Tirole (1988) J. Tirole, The theory of industrial organisation. Cambridge, Mass. ; London: MIT
Press, 1988.
Tyagi and Srivastava (2006) B. Tyagi and S. C. Srivastava, "A decentralised automatic
generation control scheme for competitive electricity markets", IE E E Transactions
on Power Systems, vol. 21 (1), pp. 312-320, February 2006.
UCPTE (1991) "Active power control in the UCPTE system - Inventory", Union for the
Co-ordination of Production and Transmission of Electricity (UCPTE), Brussels,
Belgium, 1991.
REFERENCES 275
UCTE (2004a) "Final report of the Investigation Committee on the 28 September 2003
blackout in Italy", Union for the Co-ordination of Transmission of Electricity
(UCTE), Brussels, Belgium, April 2004, Available:
h ttp ://www.ucte.org/_library/otherreports / 20040427_UCTE_IC_Final„report.p
df
UCTE (2004b) "UCTE Operation Handbook", Version 2.5E, Union for the Co
ordination of Transmission of Electricity (UCTE), Brussels, Belgium, 20 July 2004.
Available:
http://www.ucte.org/_library/ohb / Operation_Handbook_20.07.2004.pdf
UCTE (2005) "Geographical Distribution of Reserves", Version 1.0, Union for the Co
ordination of Transmission of Electricity (UCTE), Brussels, Belgium, July 2005.
Available:
http:/ / www.ucte.org/_library / otherreports / Geograp hical_Dis tribution_o f_Reser
ves.zip
UCTE (2006) UCTE, "Geographical Distribution of Reserves", Version 1.0, Union for
the Co-ordination of Transmission of Electricity (UCTE), Brussels, Belgium, July
2005 (available since June 2006). Available:
h ttp://www.ucte.org/_libraiy/otherreports/Geographical_Distribution_of_Reser
ves.zip
UCTE (2007) "Final Report - System Disturbance on 4 November 2006", Union for the
Co-ordination of Transmission of Electricity (UCTE), Brussels, Belgium, January
2007. Available: http://www.ucte.org/_libraiy/otherreports/Final-Report-
20070130.pdf
U.S.-Canada Power System Outage Task Force (2004) U.S.-Canada Power System
Outage Task Force, "Final Report on the August 14, 2003 Blackout in the United
States and Canada: Causes and Recommendations", April 2004. Available:
https://reports.energy.gov/BlackoutFinal-Web.pdf
276 REFERENCES
Varian (1999) R. Varian Hal, Intermediate microeconomics: a modem approach, 5th edition, New
York; London: W.W. Norton & Co, 1999.
Vu el al (1996) H. Vu, P. Pruvot, C. Launay, and Y. Harmand, "An improved voltage

control on large-scale power system", IE E E Transactions on Tower Systems, vol. 11
(3), pp. 1295-1303, August 1996.
Weisman (2005) D. L. Weisman, "Price regulation and quality", Information Economics and
Polity, vol. 17 (2), pp. 165-174, March 2005.
Williams and Lane (2004) H. E. Williams and D. Lane, Web Database Applications with PHP
and MySQL,, 2nd ed. Sebastopol, CA, USA; Cambridge, MA, USA: O'Reilly, 2004.
Wilson (2002) R. Wilson, "Architecture of power markets", Econometrica, vol. 70 (4), pp.
1299-1340, July 2002.
Wood and Wollenberg (1996) A. J. Wood and B. F. Wollenberg, Power Generation,

Operation and Control, 2nd ed. New York: Wiley Interscience, 1996.
Wu et al (2004) T. Wu, M. Rothleder, Z. Alaywan, and A. D. Papalexopoulos, "Pricing

energy and ancillary services in integrated market systems by an optimal power
flow", IE E E Transactions on Power Systems, vol. 19 (1), pp. 339-347, February 2004.
Yan and Stern (2002) J. H. Yan and G. A. Stern, "Simultaneous Optimal Auction and
Unit Commitment for Deregulated Electricity Markets", The Electricity Journal, vol.
15 (9), pp. 72-80, November 2002.
Young (1994) H. P. Young, "Cost Allocation", in Handbook of game theoiy with economic
applications, vol. 2. Amsterdam; Oxford: Elsevier, 1994, pp. 1193-1235.
Zhong (2003) J. Zhong, "On some aspects of design of electric power ancillary service
markets", PhD Thesis, Department of Electric Power Engineering, Chalmers
REFERENCES 277
University of Technolog)?, Goteborg, Sweden, 2003. Available:

ftp:/ / ftp .elteknik.chalmers.se/Publications/PhD / ZhongJ inPhD .pdf
Zhong et al. (2004) J. Zhong, E. Nobile, A. Bose, and K. Bhattacharya, "Localized

reactive power markets using the concept of voltage control areas", IE E E
Transactions on Power Systems, vol. 19 (3), pp. 1555-1561, August 2004.
Zhong et al. (2005) Z. Zhong, C. Xu, B. J. Billian, L. Zhang, S.-J. S. Tsai, R. W. Conners,
V. A. Centeno, A. G. Phadke, and Y. Liu, "Power system frequency monitoring
network (FNET) implementation", IE E E Transactions on Power Systems, vol. 20 (4),
pp. 1914-1921, November 2005.
Zhu et al (2000) J. Zhu, G. Jordan, and S. Ihara, "The market for spinning reserve and its
impacts on energy prices", presented at Power Engineering Society Winter
Meeting, Singapore, 2000.
Zobian and Ilic (1997a) A. Zobian and M. D. Ilic, "Unbundling of transmission and
ancillary sendees. I. Technical issues", IE E E Transactions on Power Systems, vol. 12
(2), pp. 539-548, May 1997.
Zobian and Ilic (1997b) A. Zobian and M. D. Ilic, "Unbundling of transmission and
ancillary sendees. II. Cost-based pricing framework", IE E E Transactions on Power
Systems, vol. 12 (2), pp. 549-558, May 1997. ■
278 REFERENCES
REFERENCES BY CATEGORIES 279
REFERENCES BY
CATEGORIES
G eneral
Oxford Dictionary (2001) IEA (2005)

Williams and Lane (2004) Hoinlds and Lindner (2007)
M a th em atics
Lasdon (1968) Hamilton (1994)

Fuller (1976) Bertsimas and Tsitsiklis (1997)
Cohen (1978) Peradotto (2001)
Open University (1988) Lemarechal (2004)
Berry and Lindgren (1990) Mathworks (2007)
Saporta (1990) Numerical Algorithms Group (2007)
P ow er s y s te m s
Gautier (1990) Wood and Wollenberg (1996)

British Electricity International (1991) Johannet (1997)
Grainger and Stevenson (1994) Bastard (1998)
Kundur (1994) Anderson (1999)
IEEE Std 1159-1995 (1995) Hingorani and Gyugyi (2000)
280 REFERENCES BY CATEGORIES
D G EM P/D ID EM E (2003) Robert and Prestat (2004)

U.S.-Canada Power System Outage Task
Force (2004)
Kirschen and Strbac (2004a)
R eliability o f p o w er s y s te m s
Billinton and Allan (1996) Keller and Franken (2006)

Kariuki and Allan (1996) Abiri-Jahromi et al. (2007)
Abbott (2001) Kirschen and Jayaweera (2007)
Kirschen (2002) Ortega-Vazquez and Kirschen (2007)
Joskow and Tirole (2004) Tesseron (2008)
T ech n ica l fe a tu r e s o f an cillary s e r v ic e s
Paul et ah (1990) Da Silva et al. (2001)

Jaggy and Longley (1991) Nobile et al. (2001)
Testud (1991) Proctor (2001)
Jaleeli et al (1992) Gross et al (2002)
Kundur (1994) Alvarado et al (2003)
Corsi et al. (1995) RTE (2004)
Fourment et al. (1995) Ingleson and Ellis (2005)
Vu et al. (1996) Vu et al. (1996) Sterpu (2005)
Wood and Wollenberg (1996) Zhong et al. (2005)
Bakken and Faanes (1997) Elizondo et al (2006)
Eurelectric (1997) Margotin et al (2006)
Zobian and Ilic (1997a) Raineri et al (2006)
Hirst and Kirby (1998) Saluden (2006)
Eurelectric (2000) Tyagi and Srivastava (2006)
Cali et al (2001)
E co n o m ics
Tirole (1988) Klemperer (2002)

Young (1994) Pekec and Rothkopf (2004)
Mas-Colell et al (1995) Weisman (2005)
Bergstrom (1999) IMF (2006)
Klemperer (1999) Rothkopf (2007)
Varian (1999)
P ow er s y s te m e c o n o m ic s
Bushnell and Oren (1994) Cramton and Stoft (2005)

Baughman et al. (1997a) Frontier Economics and Consentec
Baughman et al. (1997b) (2005)
Bernard et al. (1998) Garcia (2005)
Shahidehpour et al. (2002) Perrot and Molinier (2005)
Stoft (2002) Boucher et al. (2006)
Wilson (2002) ETSO tariffs task force (2006)
Yan and Stern (2002) I-Iaas et al. (2006)
CER (2003) Joskow (2006)
Ma et al. (2003) Kluge (2006)
Newbeiy et al (2003) Liu et al. (2006)
Thomas (2003) Luh et al. (2006)
Pignon (2003) O’Neill et al. (2006)
Kirschen and Strbac (2004a) Quintana and Bautista (2006)
Kirschen and Strbac (2004b) Sioshansi and Pfaffenberger (2006)
Newbeiy et al. (2004) Staropoli and Jullien (2006)
Baldick et al. (2005) Rious (2007)
M arkets for freq u en cy control ancillary s e r v ic e s
Hirst and Kirby (1996) Kirby and Hirst (1996)

Hirst and Kirby (1997) Eurelectric (2004)

Zobian and Ilic (1997b) Goel et al (2004)
Hirst and Kirby (1998) Papadogiannis and Hatziargyriou (2004)
Brien (1999) Wu etal (2004)
Singh (1999) Arroyo and Galiana (2005)
Kirby and Hirst (2000) Doherty et al (2005)
Strbac and Kirschen (2000) E.ON (2005)
Zhu et al (2000) Frontier Economics and Consentec
Cali etal (2001) (2005)
Cigre Task Force (2001) Galiana et al, (2005)
Nobile et al (2001) DePillis (2006)
Oren (2001) Hogan (2006)
Proctor (2001) Mickey (2006)
Chao and Wilson (2002) Raineri et al (2006)
Arnott et al (2003) Tyagi and Srivastava (2006)
Kirby and Hirst (2003) UCTE (2006)
Zhong (2003) Abiri-Jahromi et al (2007)
Chicco and Gross (2004) Ortega-Vazquez and Kirschen (2007)
M arkets for v o lta g e con trol ancillary s e r v ic e s
Hirst and Kirby (1996) Da Silva et al (2001)

Kirby and Hirst (1996) Gross et al (2002)
Hao and Papalexopoulos (1997) Zhong (2003)
Hirst and Kirby (1997) Electricity Commission of New Zealand
Zobian and Ilic (1997b) (2004)
Hirst and Kirby (1998) Eurelectric (2004)
Bhattacharya and Zhong (2001) Zhong et al (2004)
Cali et al (2001) Lin et al (2005)
Cigre Task Force (2001) Raineri et al (2006)
A ustralia
NEMMCO (2001) NEMMCO (2005)

NEMMCO (2004)
B elgium
Belgium economic department (2002)
California
Brien (1999) CAISO (2005b)

CAISO (2005a)
France
Paul et al (1990) Margotin et al (2006)

Fourment et al. (1995) EDF (2006)
D G EM P/D ID EM E (2003) RTE (2006)
French economic and industiy Em u (2007)
department (2003) RTE (2007a)
Robert and Prestat (2004) RTE (2007b)
EDF (2004) RTE (2007c)
Lemarechal (2004) Texier-Pauton (2007)
RTE (2004) CRE (2008)
French economic and industiy Tesseron (2008)
department (2005)
G erm any
E.ON (2005) BDEW (2007a)

BDEW (2007b) BDEW (2007c)
G reat Britain
National Grid (2007) National Grid (2008b)

National Grid (2008a) National Grid (2008c)
T he N eth erlan d s
DTe (2005) TenneT (2006)

TenneT (2003)
N ew Z ealand
Electricity Commission of New Zealand N ZIER (2005)

(2004)
Electricity Commission of New Zealand
(2005)
PJM
PJM (2005a) PJM (2007)

PJM (2005b) PJM (2008)
PJM (2006)
Spain
Spanish industry and energy department Spanish industry and energy department
(1998) (2006a)
Spanish industiy and energy department Spanish industry and energy department
(2000) (2006b)
OMEL (2005)
S w ed en
Nordel (2004) SvK (2005)
Europe
UCPTE (1991) Boucher et al (2006)

UCTE (2004a) ETSO tariffs task force (2006)
UCTE (2004b) Haas et al. (2006)
Frontier Economics and Consentec UCTE (2006)
(2005) UCTE (2007)
UCTE (2005) Meeus and Belmans (2008)
North A m erica
Hirst and Kirby (1996) Ingles on and Ellis (2005)

Kirby and Hirst (1996) NERC (2005)
Ma etal (2003) DePillis (2006)
U.S.-Canada Power System Outage Task NERC (2006)
Force (2004) O ’Neill et al (2006)
NERC (2004) NYISO (2008)
■
PUBLICATIONS
Jou rn als
Y. Rebours, D. Kirschen, M. Trotignon, and S. Rossignol, "A Survey of Frequency and
Voltage Control Ancillary Sendees—Part I: Technical Features”, IE E E Transactions
on Power Systems, vol. 22 (1), pp. 350-357, February 2007.
Y. Rebours, D. Kirschen, M. Trotignon, and S. Rossignol, "A Survey of Frequency and

Voltage Control Ancillary Sendees—Part II: Economic Features", IE E E
Transactions on Power Systems, vol. 22 (1), pp. 358-366, February 2007.
Y. Rebours, D. Kirschen, and M. Trotignon, "Fundamental design issues in markets for

ancillary services", The Electricity Journal, vol. 20 (6), pp. 26-34, July 2007.
C o n feren ces
Y. Rebours, "Some salient features of the management of frequency and voltage control
ancillary sendees", presented at the IEEE Power Engineering Society General
Meeting, Tampa, FL, USA, 2007.
Working p ap ers
Y. Rebours and D. Kirschen, "What is spinning reserve?" Release 1, The University of

Manchester, 16 September 2005. Available:
287
288 PUBLICATIONS
http:// www.eee.manchester.ac.uk/ research/groups/eeps/publications/reportsthe

ses/aoe/rebours%20et%20al_tech%20rep_2005Apdf
Y. Rebours and D. Kirschen, "A Survey of Definitions and Specifications of Reserve

Services", Release 2, The University of Manchester, 12 October 2005. Available:
http://www.eee.manchester.ac.uk/research/groups/eeps/pubfications/reportsthe
ses/aoe/rebours%20et%20al_tech%20rep_2005B.pdf ■
THE AUTHOR
Yann Rebours received his Electrical Engineer’s degree from the Ecole Superieure
d’Electricitc, France and his Master’s degree in electrical engineering from the University of
Wollongong, Australia, in 2004. During his Master, he mainly worked on the voltage
compensation of a 2-MW induction machine. He then started in March 2005 his PhD at the
University of Manchester, U.K., in partnership with Electricite de France R&D at Clamart,
France. ■
289
290 THE AUTHOR
APPENDICES
A.1 B a s ic s o f S ta tis tic s
Since the definition of some statistic indicators may vary [Berry and Lindgren (1990)], the
statistics and data analysis method used in this thesis, as well as their abbreviations and
symbols are presented in this section. Note that we consider that all the existence
conditions are met to calculate these various statistics (e.g., no division by zero, vectors
have an appropriate size and sums do not tend toward the infinity).
A.1.1 Univariate data

S Univariate data refer to the observations made for only one random variable X . Let us
consider a sample data defined by N elements, x[l] to x[N], sorted in an ascending
order;
The sample mean (3? or M), simply called mean or average in this thesis, is defined as
follows:
(A.1.1)
'ri The sample mean tends to the expected value of the random variable X when the number
of observation increases [Berry and Lindgren (1990)];
'ri The sample variance ( a 2) represents the weighted scattering of the data137:
137 N o te that the sample variance is defined slighdy differently in com parison to die variance o f the
random variable X (also called population variance). T he variance o f die random variable generally cannot be
calculated because m ost o f the time the complete population is unknow n [Berry and Lindgren (1990)].
291
292 APPENDICES
(A.1.2)
^ The sample standard deviation (p) gives a friendlier picture of the dispersion than the
variance because the standard deviation is homogeneous to X:
(A.1.3)
^ The sample median (5?), also called second quartile, is the value that splits a sample in two
parts, with 50 % o f the sample on each side. It is calculated as follows [Berry and
Lindgren (1990)]:
N +1
if N is odd => x
(A.1.4)
~N~ [N 1\
if N is even => — x + X — +1
2 _2 _ 2 J
S The first quartile (FQ) is the median of the samples below the median (the median
excluded)138. It thus represents the boundary that holds the 25 % first sample values;
'd The third quartile (TQ) is the median of the samples above the median (the median
excluded)538. It thus represents the boundary that gathers the 75 % first sample values;
^ More generally, the nthpercentile (p,) is the boundary for the n % first values of the
sample. It is calculated as follows in our study [Mathworks (2007)]:
We define:
j W = M xL _ I (A.1.5)
We find i such as:
j [ i \ < n < j [ i + 1] (A.1.6)
We can then calculate the nth-percentile:
N
A = x Ul+ — :* G M - jM ) (*[*'+!] - *[*']) (A. 1.7)
138 ]\f0 t;e that M icrosoft Excel does n o t use exactly the same algorithm to define th e first and third
quar tiles.
A.l BASICS OF STATISTICS 293
A The interquartile range (IQR) represents how the majority of the data dose to the centre
are spread. It thus a good indicator of the dispersion of the “reasonable” data. The
interquartile range may be used to find outliers, i.e. observations that are standing out of
the crowd. The IQR is defined as follows:
%2&=TJ2-F& (A-1-8)
A The interquartile mean (IQM) is the average of the 50 % samples that are around the
median and is defined as follows, where N* represents the set of the natural numbers
except zero:
3N
N 2 -± -
if e n * => — Yx[k]
4 N r=—
n +i
3N +1
N -1 2 -4 -
if— - e N £ * •[£ ]
N + 1 /i—
N -l
----H
4 i
IQM ~ 3 N -2
(A.1.9)
N -2 *
if eN
AT —2 X:—
a:+2
..... +1 4
3 N -1
N - 3 2 *
if —— - e N
IV —1 ^_n+i+1
4
A A range of values is called a class interval The number of samples in a given class interval
gives the frequency of this class. When the frequencies are divided by the size of the sample,
the relative frequencies are obtained, that are called probability in this report by misuse
of language;
A The relative frequencies are plotted in many figures in this thesis. Such a figure gives a
so-called frequency distribution and an example is given in Figure A.1.1. Note that several
frequency distributions could be defined since different class intervals may be chosen.
In addition, contrary to histograms, the areas in a frequency distribution do not mean
anything. However, class intervals in this report have all the same size for a given
frequency distribution to facilitate the reading. Lastly, the sum of the relative
frequencies is obviously equal to one.
294 APPENDICES
Probability
Value
Class interval
Figure A.1.1: E xam ple o f a frequency distribution
A.1.2 Bivariate data

Y bivariate data refer to the observations made for two random variables X and Y139. Let
consider now a bivariate sample data defined by 2 x N elements, noted from x[l] to
x[N] and from y[l] toj[N ], classified in any order, but each couple of data x[Je\ andj/[A]
represents one sample;
Y The sample covariance1441 C represents the linear relationship between x and y and is
defined as [Berry and Lindgren (1990)]:
(A.1.10)
Y The sample correlation coefficient r y, (sometimes called cross-correlation) gives a friendlier

picture of the linear relationship between the two series of data than Cxr>is comprised
between -1 and 1, and is defined as:
C
(A.1.11)
^ It is important to keep in mind that Cxy and rx? can be equal to zero despite a clear non
linear relationship between the two variables;
Y Cxy and rxy represent the strength of the linear relationship between X and Y, but do not give
information about the significance of the strength of the linear relationship. The
significance can be estimated with for example a one-tailed test (t-test);
139 The general term to designate data with observations on several variables is multivariate data.
140 The sample covanance is som etimes called cross-covanance to make the distinction with the
covanance m atnx o f a time series [Fuller (1976)]
■A A significant strong linear relationship between X and Y does not mean that X causes Y
or vice-versa;
Y A scatter plot shows the relation between two variables by plotting each observation in a
two-dimensional graph, where the horizontal axis represents one random variable and
the vertical axis the second one. This kind of graph may be very helpful to detect
specific patterns.
A.1.3 Time-dependent data
A time series is a sequence of observations made at regular interval of time [Open University
(1988)]. When the observations of the random variables are made over time, the basic
statistics presented in the previous sections can be misleading in some cases, although they
can give a good insight of some particular data arrangements (e.g., see section 3.4.3).
Y Let us consider a univariate time series defined by Al elements, noted from x[/] to x[N\,
classified in chronological order;
Y The sample autocorrelation coefficient rlag>comprised between —1 and 1, is somewhat similar

to r as rh&gives the correlation of the process with the previous observations that were
made during the lag. The autocorrelation coefficient is defined as follows for a given
lag [Hamilton (1994)]:
N-tug
L - ~){x[k + x)
% = — ------«------------------------- (A-1.12)
£=1
Y From the previous formula, the autocorrelationfunction (ACF) can be drawn with the lags
on the horizontal axis and the corresponding autocorrelation coefficients on the vertical
axis;
Y The ACF shows the influence of all the observations from x[k—la£^ to x"[/^—1] on x[C].
To estimate the specific influence of x[k—laf\ on v[X^], the partial autocorrelation function
(PACF) should be used instead. Since its formulation is not straightforward [Hamilton
(1994)], its mathematical formulation is not given here. To calculate the PACF in this
thesis, the Microsoft Excel statistical add-in from Numerical Algorithms Group (2007)
was used;
296 APPENDICES
Y An important concept for time-dependant data is stationarity, since a stationary time

series can be studied as a classical random variable (see section A.1.1). A time series is
said to be stricdy stationary if the probability laws describing the shifted time series
x[A+/J are die same as those describing the original time series x\fe\ [Open University
(1988)]. Stricdy stationary time series are rare. Practically, a time series is weakly stationary
(or covariance-stationary) when [Hamilton (1994)]:
■ The expected values and variances are constant over time;
■ Each autocorrelation coefficient is constant over time.
Y Generally speaking, a time series may be split into three components [Open University
(1988)]:
■ A trend component^ which represents a non-periodic long-term variation;
■ A seasonal component, which shows a periodic variation;
■ A stationary irregular component, which basically represents the short-term noise around
the trend and the seasonal components.
A.1.4 Descriptive analysis

This thesis tries to understand the causes underlying the de-optimisation cost, but does not
specifically attempt to predict the de-optimisation cost (see Chapter 3). The process of
understanding the data is called descriptive analysis. In general, a descriptive analysis
summarizes the characteristics (trend and dispersion) of the variable, estimates its density
and detects atypical values (the outliers). Then, it may be necessary to transform data with for
example a logarithmic transformation. Such transformations are useful to separate the
different components of a time series, as introduced in section A.1.3. Lastly, descriptive
analysis helps to understand the links between variables [Peradotto (2001)].
In this thesis, the following steps are performed to analyse the computed de
optimisation cost:
Y Data selection', in this process, data are extracted from the OTESS database and some
basic calculations are performed (e.g., subtraction between two costs). Most o f the data
are defined for each time step of the day whereas the de-optimisation cost has a
meaning only over a day. Therefore, basic statistics for each parameter are taken, such
as the minimum, the maximum, the mean, the standard deviation and the difference
between the minimum and the maximum over the 48 first steps (the 48 following steps
ate ignored, as explained in section 3.3.6). In total, around 100 variables have been
defined;
C Handling missing data: the missing data are completed as explained in section 3.3.4.3;
T Calculation of basic statistics', the frequency distributions are plotted for the valid data and
basic statistics are given to evaluate the quality of the series;
T Seasonality', the seasonality of the time series is analysed with the ACF and the PACF (see
section 3.4.2);
Data smoothing (or filtering): in order to remove the noise around the main variation of
the series, a smoothing is performed with a window-moving average according to
(A. 1.13). In other words, this technique tries to remove the stationary component from
the original series. For example, Figure 4.4 represents the smoothing process with a 3-
element window. Note that the new set of data is smaller than the window since the
beginning and the end of the series cannot be smoothed. In this report, the window is
chosen according to the seasonality period;
nwdtm'—1
y
/ i *xwigm!w
I .. ,
;/'/WflW'“ l
k=/—
’ smoothed |7] —-------- y—;------ , where window is odd (A. 1.13)
window
Original series
2-element window
0
1 1
Smoothed series
>|C
Ti < >c
X
F igure A.1.2: Schem atic representation o f the sm o o th in g p rocess
T Getting the stationary component this is done by removing the trend and seasonal
components of each series over time and any correlation correction. Seasonal patterns
can be detected by calculating the auto-correlation coefficients of the series or by
performing a Fourier Transform or a caterpillar decomposition [Garcia (2005)]. In this
298 APPENDICES
study, the trend and seasonal components are removed by simply subtracting the
smoothed series from the original one;
S Normalisation', this makes possible comparing data with different units or orders of
magnitude. The normalization used in this report is shown in (A.I.14), where x and a
are respectively the mean and the standard deviation of the series to be
normalized [Garcia (2005)].
r n x\t} —x
'd Multidimensional analysis', this report relies on a basic multidimensional analysis, for which
correlation coefficients are calculated and scatter plots are drawn. More complex
techniques such as clustering (e.g. with the help of Kohonen networks) could be
performed for further studies [Garcia (2005)]. ■
A.2 TECHNICAL DESCRIPTION OF OTESS 299
A.2 T ech n ica l D escrip tion of OTESS
This section gives an overview of OTESS from a technical point of Hew. It is assumed here
that the reader is familiar with the optimisation process described in section 3.3. OTESS
stands for OuTilpour /’Etude des Sendees Systime (Tool for the study of ancillary sendees). This
software has been initially developed to manipulate the operating data of EDF Producer in
order to estimate its de-optimisation cost because of frequency control reserves (see
Chapter 3). However, because it is object-oriented architecture, this softaware could be
quickly adapted to any information system. Furthermore, the studies possible with OTESS
are not limited to ancillary sendees. In fact, any kind of research related to the generation
optimisation algorithm can be performed, such as assessing the impact of a new portfolio or
different market prices.
A.2.1 Basic functionalities

As presented in section 3.3.5.2, the main features of OTESS are the following:
V Handle and modify inputs and outputs of EDF’s dispatch algorithm;

V Extract the selected data from the optimisation algorithm’s output tiles and store them
into a relational database;
V Store any kind of additional data in its database, such as such as the target frequency,
the balancing mechanism prices or the final dispatch planning (theprogramme d’appef);
H Easily modify data of the database with user-defined scripts;
H Allow complex querying on cross-related tables;
H Display data with various graph types directly from queries(e.g., scatterplots or
frequency distributions, as shown in Chapter 3);
V Perform basic data mining techniques;
V Export data under various formats (e.g., CSV or XML141);
V Share data between users since it is a server-based software witha web-basedhuman-
machine interface;
V Run on both Linux and Windows environment;
V Based on free software;
141 CSV stands for Com m a-Separated Values and XM L for Extensible M ark-up Language
300 APPENDICES
S Strongly modular because object-oriented.
A.2.2 Hardware architecture

The architecture of OTESS is based on the famous LAMP architecture (Linux, Apache,
MySQL and PHP) and is shown in Figure A.2.3 where the basic operations are as follows:
Operation 1: the operational data are stored as zipped text files on a specific server. They
are thus transferred via FTP to the main machine, where they are unzipped;
Operation 2: the datasets are modified by OTESS according to a given script. For
example, the demand for reserve can be set to zero (see section A.2.6);
S Operation 3: the modified datasets are sent via FTP to several simulation machines in
order to reduce the computation time (see section 3.3.5.1 for more details on these
machines);
S Operation 4: a KSH script launches automatically ED F’s dispatch alogorithm for the
modified datasets. Following the optimisation process, new datasets are created as
outputs;
^ Operation 5: the results are sent back via FTP to the main machine;
S Operation 6: only a part of the dispatch results are used to feed in the relational database
of OTESS (see section A.2.3 for more details on the database structure);
S Operation 7: the user makes use of the functionalities of OTESS in order to extract the
data that are useful via tailored-made scripts and a generic web-based interface (see
section A.2.5 for a few screenshots).
EDF’s
historical
Dataset Datasets
datasets
generator
Storage server
Database
feeder >■
Datasets EDF’s
] Batch [ dispatch
algorithm
S i Database
\ Simulation machines
Database
analyser
Main machine
Figure A.2.3: H ardware architecture o f OT E SS
A.2.3 Software architecture

OTESS relies on a relational database. The entity relation modelling of the main tables of
OTESS is given in Figure A.2.4142. To give an idea of data stored, Table A.2.1 to
Table A.2.4 give die structures of four data tables. In fact, there are 27 tables, amongst
which 15 are dedicated to OTESS management (e.g., relations between tables, user
management, session, script locations, graphs...). Nevertheless, this relational database can
be easily modified through a dedicated interface in OTESS.
142 F o r m ore inform ation on entity relation modelling, see for example Williams and Lane (2004).
302 APPENDICES
day script units
provides
global_dispatch 1 X fo r th e ^ s.
_statement \a iv e n
M M
1
dataset
for t h e ^ ' x .
demand
for the
timestep_dispatch for the
timestep given
statement given
supply
Figure A ,2,4: E n tity relation m od ellin g o f the m ain tables o f O T E SS
T ab le A.2.1: Structure o f the table “u n it”
N am e Data type Unit D escription

U N IT JD INT(3) >0 Unit ID (unique primary key)
NAME VAR(IO) Generating unit name, identical to the
one used by ED F (e.g., ‘CHIN2T 1’)
MAIN_TYPE (‘thermal’,’hydro’) Group main type: thermal or hydro
TYPE CHAR(l) For therm al: ‘N ’ : nuclear, ‘H ’ :
bituminous coal, ‘C’ : coal, ‘F’ : fuel, ‘G’ :
gas turbine, ‘B’ : bagasse, ‘E ’ : exchange ,
‘P’ : Phenix)
For hydro : 'L' : lake downstream , 'E' :
lock downstream , 'F' : fatal downstream
, 'P' : pumping , 'D' : floodgate , 'R' :
restitution
CITY„NAME VAR(255) - Name of the place
CITY_CODE INT(5) >0 - Zip code of the place
T ab le A.2.2:: Structure o f the table “d em an d ”
N am e Data type Unit Description

DATASETJiD INT(8) >0 - See table “dataset” (co-primary key)
TIMESTEP_ID INT(8) >0 - See table “timestep” (co-primary key)
NRJ SCI >0 or MW Forecasted power demand
<0
SEC SCI >0 MW Secondary reserve demand
PRI SCI >0 MW Primary reserve demand
T able A.2.3: Structure o f the table “supply”
N am e Data type U nit Description
DATASET_ID INT(8) >0 - See table "‘dataset” (co-primary key)

U N IT JD INT(3) >0 - See table "unit” (co-primary key)
T IM ESTE PJD INT(8) >0 - See table "timestep” (co-primary key)
NRJ SCI >0 or MW Dispatched power
<0
SEC SCI >0 MW Dispatched secondary reserve
PRI SCI >0 MW Dispatched primary reserve
T able A.2.4: Structure o f the table “glob al_d isp atch _statem en t”
N am e Data type Unit Description
D A TA SETJD INT(8) >0 “ See table “dataset” (unique primary

key)
NRJ_AVG_GAP SCI >0 or MW Average gap between scheduled power
<0 supply and estimated power demand
over the first 48 steps
SEC_AV G_GAP SCI >0 or MW Average gap between scheduled
<0 secondary reserve and secondary
reserve demand
PRI_AVG_GAP SCI >0 or MW Average gap between scheduled
<0 primary reserve and primary reserve
demand
THM_COST SCI >0 or k€ Cost of the thermal power supply
<0 dispatch for the first 48 steps
HYD_COST SCI >0 or k€ Cost of the de-stocked water for the
<0 first 48 steps
NRJ_PENALTY SCI >0 or k€ Cost of the power supply mismatch for
<0 the first 48 steps
SEC_PENALTY SCI >0 or k€ Cost of the secondary reserve
<0 mismatch for the first 48 steps
PRI_PENALTY SCI >0 or k€ Cost of the primary reserve mismatch
<0 for the first 48 steps
TOTAL_COST SCI >0 or k€ Total cost of dispatch, including
<0 penalties, for the first 48 steps
304 APPENDICES
A.2.4 Param eters for data analysis

The following parameters have been computed and stored in two specific tables of OTESS
in order to perform multidimensional data analysis (operation 7 in section A.2.2 ). Only a
few of these parameters have exhibited a clear interest (see section 3.4.3). Note that the
relations between the marginal costs of reserves and other parameters have not been
included in this thesis for obvious confidentiality reasons.
S For the de-optimisation cost (table “descriptive_analysis_normalized_data_dates”):

■ VALID: indicates whether the data is original (1) or extrapolated (0). See
section 3.3.4.3 for more information on extrapolated data;
■ DATE: date of the day;
■ DAY„NUMBER: number of the day (from 1 to 365);
■ DAY„NAME: name of the day;
■ SEASON: season of the day;
* BANKHOLIDAY: indicates whether the day is a bank holiday (1) or not (0);
■ CHAN GIN G_TIME: indicates whether the Summer/Winter time is triggered
during this day (1) or not (0);
* TARGET_FREQUENCY: UCTE target frequency for this day (49.99, 50.00 or
50.01);
■ WEEKDAY: indicates whether the day is a weekday (1) or a weekend (0);
■ INITIAL_COST (_TOT, _THM, „HYD, _PEN): cost of dispatch with reserves
(total, thermal, hydro, A POGEE penalty) (in k€). See section 3.3.6 for more
information on the cost calculation;
■ COSTJDIFFERENCE (_TOT, _TOT_PERCENT): difference between the cost
of dispatch without reserves and the cost of dispatch with reserve (absolute —in k€,
relative);
* NRJ_DEMAND (_MEAN, _MIN, JMAX, _DELTAMINMAX, _STDDEV):
initial demand for energy for the day (mean, minimum, maximum, maximum minus
minimum, standard deviation) (in MW/h). See Section A.1.1 for more details on the
statistics used;
■ SEC_DEMAND (_MEAN, ...): initial demand for secondary reserve (in MW/h);
■ PRI_DEMAND (_MEAN, ...): initial demand for primary reserve (in MW/h);
■ SPIN_DEMAND (_MEAN, ...): initial demand for primary + secondary reserves
(in MW/h);
■ THM_SHARE_NRJ_DISPATCH (_MEAN, ...): the thermal generation divided

by the total generation;
* TFIM_SHARE_SEC_DISPATCH (_MEAN. ...): the secondary reserves provided
by thermal units divided by the total provided secondaiy reserves;
■ THM_SHARE_PRI_DISPATCH (_MEAN, ...): same as above but with primary
reserves;
■ THM_SHARE_SPIN_DISPATCH (_MEAN, ...); same as above but with primary
+ secondaiy reserves;
■ NRJ_DISPATCH (_MEAN, ...): dispatched generation (in MW/h);
■ SEC_DISPATCH („MEAN, ...): dispatched secondary reseives (in MW/h);
■ PRI_DISPATCH („MEAN, ...): dispatched primaiy reseives (in MW/h);
■ SPIN_DISPATCH (JVLEAN, ...): dispatched primaiy + secondaiy reseives (in
MW/h);
■ NRj_NEGATIVE_MARGIN („MEAN, ...): the dispatched generation minus the
minimum available generation (in MW/h);
■ NRJ_POSITIVE_MARGIN (_MEAN, ...): the maximum available generation
minus the dispatched generation (in MW/h);
■ SEC„MARGIN (_MEAN, ...): the maximum available secondaiy reseives minus
the dispatched secondary reseives (in MW/h);
* PRI_MARGIN (_MEAN, ...): same as above with primaiy reseives (in MW/h);
■ SPIN_MARGIN (_MEAN, ...): same as above with primary + secondaiy reseives
(in MW/h);
■ NRJ„MARGINAL„COST (_MEAN, ...): marginal cost of energy (in €/M W /h);
■ SEC_„MARGINAL_COST (_MEAN, ...): marginal cost of secondaiy reseives (in
€/M W /h);
■ PRl_MARGINAL_COST (_MEAN, ...): marginal cost of primaiy reseives (in
€/M W /h);
■ NRJ„GAP (_„MEAN, ...): dispatched generation minus the initial demand for
energy (in MW/h);
■ SEC_GAP (_MEAN, ...): dispatched secondaiy reseives minus the initial demand
for secondaiy reseives (in MW/h);
■ PRI_GAP (_MEAN, ...): dispatched primaiy reseives minus the initial demand for
primaiy reseives (in MW/h);
306 APPENDICES
■ SPIN„WEIGHTED_SUM_MARGINAL_COSTS (JMEAN, ...):
X'R = —- ------- ----------- 1-------------- (in €/M W /h). See section 3.4.3.1 for more
l )ri AiijHiuh ^ M dhpakb
explanation on this indicator;

« ALL_WEIGHTED_SUM_MARGINAL„COSTS (JMEAN, ...): same as above,
but considering energy + secondary + primary instead of secondary + primary only
(in €/M W /h).
V For the marginal costs of reserves (table “descriptive_analysis_normalized_
data_times teps”):
VALID: indicates whether the data is original (1) or extrapolated (0). See
section 3.3.4.3 for more information on extrapolated data;
DATE: date of the day; ,
TIMESTEP: number of the time step (from 1 to 48);
D AY_NUMBER: number of the day (from 1 to 365);
DAY_NAME: name of the day;
SEASON: season of the day;
BANKHOLIDAY: indicates whether the day is a bank holiday (1) or not (0);
CHANGING_TIME: indicates whether the Summer/Winter time is triggered
during this day (1) or not (0);
TARGET„FREQUENCY: UCTE target frequency for this day (49.99, 50.00 or
50.01);
WEEKDAY: indicates whether the day Is a weekday (1) or a weekend (0);
NRJ_DEMAND: initial demand for energy for the time step (in MW/h);
SECJ3EM AND: initial demand for secondary reserve (in MW/h);
PRIJDEMAND: initial demand for primary reserve (in MW/h);
SPINJDEMAND: initial demand for primary + secondary reserves (in MW/h);
THM_SHARE_NRJ_DISPATCH: the thermal generation divided by the total
generation during the time step;
THM_SHARE_SEC_DISPATCH: the secondary reserves provided by thermal
units divided by the total provided secondary reserves;
THM_SHARE_PRI__DISPATCH: same as above but with primary reserves;
THM_SHARE„SPIN„DISPATCH: same as above but with primary + secondary
reseives;
NRJ_DISPATCH: dispatched generation during the time step (in MW/h);

SECJDISPATCH: dispatched secondaiy reserves (in MW/h);
PRI_DISPATCH: dispatched primaiy reserves (in MW/h);
SPIN_DISPATCPI: dispatched primaiy + secondaiy reserves (in MW/h);
NRJ_NEGATIVE_MARGIN: the dispatched generation minus the minimum
available generation (in MW/h);
NRJ_POSITR'rE_MARGIN: the maximum available generation minus the
dispatched generation (in MW/h);
SEC_MARGIN: the maximum available secondaiy reserves minus the dispatched
secondaiy reserves (in MW/h);
PRI_MARGIN: same as above with primaiy reserves (in MW/h);
SPIN_MARGIN: same as above with primaiy + secondaiy reserves (in MW/h);
NRJ_MARGINAL__COST: marginal cost of energy for the time step (in €/M W /h);
SEC_MARGINAL__COST: marginal cost of secondaiy reserves (in €/M W /h);
PRI_MARGINAL_COST: marginal cost of primaiy reserves (in €/M W /h);
NRJ_GAP: dispatched generation minus the initial demand for energy (in MW/h);
SEC_GAP: dispatched secondaiy reserves minus the initial demand for secondaiy
reseives (in MW/h);
PRI_GAP: dispatched primaiy reseives minus the initial demand for primaiy
reseives (in MW/h);
SPIN_WEIGHTED_SUM_MARGINAL_COSTS:
Xi x R . + Xi x R ( ,
X'R = — - -— ^ ------------------ (in €/M W /h). See section 3.4.3.1 for more
RPn' tthpnhh + Rm' dispatch,
explanation on this indicator.

ALL_WEIGHTED_SUM„MARGINAL„COSTS: same as above, but considering
energy + secondaiy + primaiy instead of secondaiy + primaiy only (in €/M W /h).
A.2.5 Screenshots
In order to illustrate some of the basic functionalities of OTESS mentioned in
section A.2.1, Figure A.2.5 to Figure A.2.10 display a few sreenshots from OTESS (note
that OTESS is compatible with most of web browsers).
308 APPENDICES
E 9B S -I#Ix<
; ::.m
“3 in
3*—-«g e ■ 0 Paroton-*
Q 9C IO »»ith the script ‘2. Set PRI & SEC reserves to zero’
\C urrm utB & t |: i.*-'*. i -E : it '^ j m&ry Q+m O m r t mw r o w ]
\Soim *naW *lorr TW* >cn» « » s t » darnrwil m larrnt or sacanrtary aid pntnary»aquan<yeo»*olreserves a l t MWf<y a»»e97 areas |
O tttfn tn U n o h ftn Date—t i erew tf wmh th* kct$a and wmtgng for DaUMCt m aty to JnMrtiraoffw OM Asm stnthe
toktm APOGEE (itusfmufBtWH iiti cUttawarehQus*
Ht-OtOMKMB » N ccx a a sseso NCofaawses C Nb of datasets. 893
VKOftg V'«W to VlBHVtg V l««tg
1 1 Origin toAtm » tor APOGEE 2. After APOGEE » Oatwrerehoujc
'4 » c /n a n da ta se ts t& c re a * 0 I (s*>of new d& asflts to a r o a t 0 . - ••« | 1
I V im I S »«">«
Decesen In m e ongin foUtr * Too*
iloerl
» * h » » to * H n lw 1 * ita 'A P O G £ E -Ja?l]
*Tac«j
Figure A .2.5: M odification and im port o f datasets w ith O T E SS (operations 2 and 6 in sectio n A.2.2)
Adnr Uhn tfltrtwei Nmm out* 7 w m m m m m m — ■——

4* •4 difi *
im—mima^/trj&.itonr^teaa»— »] *** urn
Ca-^tjC* -|i»nip4' C• O«•>•«•- <«-
* B H ffig n OTESS
eDF mpumr _UnlWOTfWlU.titurn OKpUlS
. —. _
Datawarehouse: 2. Demand
zJ ItedBr a n Cant, ww war]
APOGEE input Pnmary (MW) f dem and P R J;

APOGEE input Secondary (MW) f d e m a n d 'S E C )
D ataset O ( dataset DATASETJD1
D ay Cda y DAY^
mu Mont> (d ay ’MONTH')
Power (MW) (-demand NRJ")
(day~YEAP~)
Starr (sm e step START)
OaUMtfO Kear Nontf) Day Starr P tm m lM W ) APOGEE Input P iim tty (MW) APOGEE input Secondary f»M )
r J . • *• • m •• * ' • ■* • •
• ts/xtf 20C« 06 09 11 3000000 54811 9 6710 620
• a y r Jon 2006 06 09 113000000 S2S449 671.0 620
• U / x « 2006 06 06 1 2 0 0 00000 55370* 674 675
• a / X 2061 2006 06 09 120000000 53514 15 674 675
• B -/* » 2061 2006 06 09 123000000 54978 1? 674 675
• a / x « 2006 06 06 123000000 56792* 674 675
----------ft—F F ------ A d M o m ilM *
Figure A.2.6: M an agem en t o f data in the database (operation 7 in sectio n A.2.2)

■ H B S
Rdner Edbon Afftchoge Fevens Outfc >
^ ^ e c e a e rtr . 4 * £ Wj -
AdfiM I 4 i http f) i:7 0.0 l/OTESS/outrU.KWMwe.pho ■»] ^CK Dsn*
C o .'j k C - K - O H m lm m - j5> I t W a M I TF <*!»»«»» . » b w o y > r.
I B ia jO T E S S
O utput scripts
T hese scripts select appropriate data from the datawarehouse. transform It and then store the result into the table output_dala' T hese scripts are thus
useful to show complex relationships between data
Script: 1. C ost o f tim e control (49.99) • Q> S x
Description:
Launch script: *»«**■ i
Script: 1. C ost o f tim e control (50.01) • a s x
Description:
Launch script: ***'<» I
Script: 1. C ost w ith reserves - C ost w ith o u t reserves <* a y* x
Description:
Launch script: t - * * I
Script: 1. O riginal cost - cost w ith w in d generation » a / x
F ig u re A.2.7: C re a tio n o f n ew d a ta in th e d a ta b a se (o p e ra tio n 7 in se c tio n A.2.2)
r Q H9btoqu«<<> ^ Orthngrephe >* * Envoy*'

—
SWMCHh-ihR (OTESS
C alcu latio n view : 0. A nalysis c alcu latio n s

t; 11 Anafvsit calculations Modify Create a new view
C alculation: 1 .M ean » s y x
Filter: VALID = 'V AND DATE >= '2007-01-01' AND DATE <= '2007-06-31'
Order:
Limit:
» Number o f records: 8352
Function '(1 variable) Mean: sam ple m ean' on:

X 1 = descriptive_analysis_raw_data_timesieps , SEC_DEMAND • Gs / X
equals 706.01727490425
C alculation: 2.ln terqu artile M EAN ® a s x
Filter: VALID - 1 ' AND DATE > '2006-07-31' AND DATE < '2007-08-01'
Order:
Limit:
» Number of records 352
Function'll variable) Mean: IQM (Interquartile mean)' on:
F ig u re A.2.8: B asic c a lc u la tio n o n th e d a ta (o p e ra tio n 7 in se c tio n A.2.2)

310 APPENDICES
U»*lr - C' O•*•••— 9 i**M*I * *e~wr».

~3 «*» -
^ ^ i i w aripft
Graph: Demand for primary reserves (from C O N SO file) • a ✓ x

Comnwrt
>■ A dda now plo)
Plot 1: Primary reserve distribution • 14 ✓ X

» * ■
» V ■ demand PRI
» » Filer « dataset SCRIPTJD “T AND timestep START < '24:00:00:000'
» Order *
» Limit -
» Visible (1 or 0 | ■ 1
0 18-
I
0 16j
0 14.
0 12 -
r
0.4
0 08*
0 06-
0 04.
0 02 *
450
—
500
^riTftn
550 600 6 5 0 700 750 800 850 900 950 1000
P rm a ry rese rv e d em and (MW)
(Vj Prim ary rese rv e distnbution
Figure A.2.9: D isplay o f a graph w ith O T E SS (operation 7 in sectio n A.2.2)
Data analysis from the table 'descriptive_analysis_normalized_data_dates’

Data fiber * "VALID'-T
CalreUboe of I k cross-corretatjoe coefficieets of CO STD IFFER EN C X TO T with:
M a re a s D ru o e bv to o m D These OTESS OTXSS W ehsite ieclede Mathematics SrriesO p php on fa- 92

1 » SPIN WEIGHTED_SlTM_MARGINAI_COSTS_MEAN = 0 8563862237343 scaBgplot
2 » SECMARGINAL COST_MEAN = 0 83333204042379 te a m plot
3 » PRI MARGINAL COST_MEAN = 0 81505316266117 | scattg plot
4 » COST_DIFFERENCE_TOT_PERCEXT = 0 7816201371398 scaflgpkx
5 » INITIAL COST PEN = 0 "2983189442686 scattg p fa
6 » ALL WE1GHTED_SUM_MARGINAL COSTS -MEAN = 0 62904182684413 t scaBg plot
7 » NRJ MARGINAL_COST_MEAN = 0 6220*4OS692S65 scaBgplot
8 » SPIN_WE1GHTED_SL‘M_MARGINAL COSTS MAX = 0 59406596411463 sc alto plot
9 » SEC_MARGINAL COST.MAX = 0 58602588'62824 scana plot
10 » PR!_MARGINAL_COST_MAX = 0 55998599345198 I tcattgpiot
11 » N R J GAP_STDDE\' = 0 5404’ 529E69S scaBgplot
12 » SPIN_WE3GHTED_SUM MARGINAL COSTS STDDEV =0.53117211806563 scattg plot
13 » SEC_MARGINAL_COST_SrDDEY - 0 52943655662551 scattg plot
14 » SPIN WEIGHTED _SL\M MARGINAL COSTS MIN = 0 52692S63424786 i scattg plot
15 » PRI MARGINAL.COST STDD£V = 0 5264S1”8475142 scaBgplot
16 » SEC MARGINALJ^STJ3ELTAMINMAX = 0 5O4'2604'413*9 seaagpktt
17 » SPIN WnGHTED_SL"M_MARGINAL COSTS_DELTAMINMAX = 049708267656472 scaBgplot
18 » NRJ_MARGINAL_COST_STDDEY = 0 4844022086616 scaBgplot
19 » SEC_MARGINAL_COST_MIN = 0 48412235323812! scaBgplot
2 0 » ALL WEIGHTED SVM_MARGINAL_COSTS_STDDEV = 048396663506088 scaBgplot
21 » PRl_MARGINAL_COSf_DELTAMINMAX = 0 48303358260195 scaBgplot
Figure A .2.10: A nalysis o f data w ith O T E SS (operation 7 in sectio n A.2.2)
A.2.6 Code example

To perform complex tasks such as creating new datasets or modifying data, OTESS allows
users to define their own scripts. For example, the following script has been used to create
new datasets with the demand fot reserves set to zero (note the object-oriented
programming that allows any change in the structure of the dataset text files):
<?php
/* *\
d a ta s e t/s c r ip t/r e s e r v e _ d e m a n d _ a t_ z e r o .php
W ritte n by: Y a n n REB OURS (y a n n . r e b o u r s @ i e e e . o r g )

C rea ted : 2 0 0 7 .0 9 .1 1
L a s t m o d ifie d : 2 0 0 7 .1 0 .1 2
D e sc rip tio n : Copy th e d a t a s e t s and s e t th e dem and f o r th e p r im a r y an d
seco n d ary re s e rv e s a t zero
U tilisa tio n : u s e s t h e POST m e t h o d f o r :
- S C R IP T _ID : t h e i d o f s c r i p t t h a t we u s e
- d a t a s e t s : e n c o d e d l i s t o f th e nam es o f th e d a ta s e ts to copy
(C ) The U n iv e rs ity o f M an ch ester 2 007

(C) E le c tric ity de F ra n c e 2007
\* */
/////////////
// MODULES //
/////////////
re q u ire_ o n ce 1p a r a m e t e r s . i n c ' ; / / P a ra m e te rs
re q u ire_ o n ce ' S y s t e m / F o l d e r . p h p 1; / / To m a n a g e f o ld e rs and f i l e s
re q u ire_ o n ce 'H T T P /I n p u t.p h p 1; / / To h a v e a c c e s s t o th e a rg u m en ts
re q u ire_ o n ce ’D a t a s e t / S c r i p t . p h p ' ; / / C la ss t h a t m anages s c r ip t s
re q u ire_ o n ce 'D a ta s e t/F ile _ C o n s o .p h p ' ; / / To h a v e a c c e s s t o t h e CONSO f i l e
////////////////
// PARAMETERS / /
////////////////
/ / W e g e t t h e ID o f t h e s c r i p t
$ i d _ s c r i p t = H T M L _ In p u t: : g e t l n p u t s ( "S C R IP T _ ID ", "P O S T ");
/ / We g e t t h e l i s t o f d a t a s e t s t o c o p y
$ a r r a y _ d a t a s e t s _ t o _ c o p y = H T M L _ In p u t:: d e c o d e U r l V a r ia b l e ( H T M L _ I n p u t: : g e t l n p u t s ( " d a t a s e t s ",
"P O S T "));
/ / We g e t t h e i n f o r m a t i o n o n t h e s c r i p t
$ s c r i p t = new S c r i p t ( $ i d _ s c r i p t ) ;
/ / We d e f i n e t h e i n i t i a l and th e d e s tin a tio n fo ld e rs

$ p a th _ in itia l_ fo ld e r = $ s c r ip t- > g e tI n itia lF o ld e r ();
$ p a th _ d e stin a tio n _ fo ld e r = $ s c rip t-> g e tD e s tin a tio n F o ld e r();
/ / We c r e a t e a f o l d e r o b j e c t r e l a t e d t o t h e d e s t i n a t i o n f o l d e r
$ d e s t i n a t i o n _ f o l d e r = new F o l d e r ( $ p a t h _ d e s t i n a t i o n _ f o l d e r ) ;
//////////////////////////////
// C R E A T I O N OF TH E D A T A S E T S //
//////////////////////////////
/ / We d e f i n e t h e e m p t y r e s e r v e a s a n a r r a y o f 9 7 e l e m e n t s
$ e m p t y _ r e s e r v e = a r r a y _ p a d ( a r r a y ( ) , 9 7 , " 0 . QOOOOOOOOE+OO") ;
/ / We g o t h r o u g h t h e l i s t o f d a t a s e t s t o c o p y
fo re a c h ( $ a rra y _ d a ta se ts_ to _ c o p y as $ c u rre n t_ d a ta se t ) {
/ / We c h e c k w h e t h e r t h e d a t a s e t t o c o p y i s v a l i d a n d w h e t h e r th e ta rg e t d ataset does
n o t alread y e x is t
if ( i s _ d i r ( $ p a t h _ i n i t i a l _ f o l d e r . $ c u r r e n t _ d a t a s e t ) && !
is _ d ir($ p a th _ d e s tin a tio n _ fo ld e r . $ cu rren t_ d ata se t) } {
/ / We c o p y t h e d a t a s e t
$ d e s tin a tio n _ f o ld e r - > c o p y F o ld e r ($ p a t h _ i n i t i a l _ f o l d e r . $ c u rre n t_ d a ta se t ,
$ p ath _ d e stin a tio n _ fo ld e r . $ c u rre n t_ d a ta se t);
/ / We s e t t h e d e m a n d a t 0 i n t h e CONSO f i l e
$ f i l e _ c o n s o = new F i l e _ C o n s o ( $ p a t h _ d e s t i n a t i o n _ f o l d e r . $ c u rre n t_ d a ta se t .
" /C O N S O " ," e d it" );
$ file _ c o n s o -> se tS e c o n d a ry D e m a n d ($ e m p ty _ re s e rv e );
$ file _ _ c o n s o -> se tP rim a ry D e m a n d ($ e m p ty _ re s e rv e );
/ / W e f re e th e m em ory
/ / N o te : t h i s step is p a rtic u la rly im p o rtan t when th e re are hundreds of d a ta se ts
to m anage
312 APPENDICES
unset ($ file _ c o n so );
///////////////////
// RETURN HEADER / /
1111111111111111111
/ / W e r e t u r n to th e pag e t h a t h e lp s to m anage d a t a s e ts
h e a d e r ( " L o c a t i o n : " . 0TESS_L0CAT10N_R00T . " d a t a s e t / m a n a g e . p h p " ) ;
exit ;
?>
To give an idea, the APOGEE text file that defines the demand for reserves and
power for a given dataset looks as the following:
•C O N SO ' ' 2 8 / 0 3 / 2 0 0 7 * '1 3 : 1 6 '
• 2 9 /0 3 /2 0 0 7 ’
• CONSOMMATION•
6 . 10609000E+04
5 .7 9 0 4 3 0 0 0 E + 0 4 5 .66327000E+04 5 . 68756000E+04 5 . 67770000E+04 5 . 6 1 2 0 1 000E+04 5 .4 8171000E + 04
5 .4 7 6 7 3 0 0 0 E + 0 4 5 .37623000E+04 5 . 55873000E+04 5 .57700000E+04 5 . 5 7 5 4 8 000E+04 5 . 6 8 7 3 3 OOOE+04
S . 8 6 1 9 6 0 0 0 E + 04 6 . 11816000E+04 6 .52323000E+04 6 .59297000E+04 6 .6 7 4 8 4 000E+04 . 72475000E+04
6 . 814S4000E+04 6 . 81704000E+04 6 . 89083000E+04 6 . 87611000E+04 6 . 9 3 7 5 6 0 0 0 E + 04 .9 4 6 2 6 0 0 0 E + 0 4
6 .78709000E+04 6 . 90679000E+04 6 .77728000E+04 6 .74430000E+04 6 .66728000E+04 . 54041000E+04
6 .4 6 8 1 3 0 0 0 E + 0 4 6 .3 9915000E+04 6 .36389000E+04 6 .32568000E+04 6 . 2 8 6 4 5 0 0 0 E + 04 .2 9 8 5 7 000E+04
6 .32740000E+04 6 .44215000E+04 6 . 51604000E+04 6 .50362000E+04 6 .6 2 5 1 3 000E +04 ■70456000E +04
6 .5 0 1 8 1 0 0 0 E + 0 4 6 .3 7 3 4 1000E+04 6 . 32413000E+04 6 .46282000E+04 6 .2 6 9 3 1 000E +04 . 17510000E+04
5 .8 9 2 6 2 0 0 0 E + 04 5 . 73462000E+04 5 . 79264000E+04 5 .77593000E+04 5 .6 5 2 8 5 0 0 0E+04 5 .5 1 4 3 6 0 0 0 E + 0 4
5 .4 7 1 4 5 0 0 0 E + 0 4 5 . 37294000E+04 5 . 45065000E+04 5 .46943000E+04 5 .6 0 0 8 3 000E+04 5 . 69993000E+04
5 . 9 9 4 1 5 0 0 0 E + 04 6 .26598000E+04 6 .5 4 907000E+04 6 .6 5 4 09000E+04 6 .6 7 3 8 1 0 0 0 E + 0 4 6 . 7 6 1 5 0 000E+04
6 .8 4 9 7 2 000E + 04 6 . 87932000E+04 6 . 95726000E+04 6 .93596000E+04 6 .9 8 3 9 1 0 0 0 E + 0 4 6 . 9 7 8 4 1 000E+04
6 .8 0 6 5 7 0 0 0 E + 04 6 . 90857000E+04 6 . 80906000E+04 6 . 7 5 9 5 3 OOOE+04 6 .6 2 6 3 8 000E+04 6 . 4 7 1 8 4 000E+04
6 .3 6 2 8 5 0 0 0 E + 04 6 .27277000E+04 6 .21265000E+04 6 .14530000E+04 6 . 08399000E+04 6 . 0 9 2 2 4 000E+04
6 . 1 3 7 0 1 0 0 0 E + 04 6 .2 6 4 96000E+04 6 . 31714000E+04 6 .28654000E+04 6 .5 5 0 3 0 0 0 0E+04 6 . 61970000E+04
6 . 4 4 2 7 3 0 0 0 E + 04 6 . 30994000E+04 6 .28869000E+04 6 .46170000E+04 6 .23045000E+04 6 .1 3 9 1 7 000E + 04
' P R IM A IR E '
5 .5 3 2 5 6 2 3 5 E + 0 2
5 .55831191E*02 5 . 65831191E+02 5 . 66911208E+02 5 . 66911208E+02 5 .6 6 9 1 1 2 0 8E+02 5 . 6 5 8 3 1 1 9 1 E + 02
5 .6 6 4 3 1 1 9 1 E + 0 2 5 .71131191E+02 5 .70051235E+02 5 . 69451235E+02 5 .7 0 5 3 1 1 9 1E+02 5 . 6 2 5 3 1 1 9 1 E + 02
5 .5 1 5 5 6 6 9 1 E + 0 2 5 . 50521708E+02 5 . S3721708E+02 5 .49621708E+02 5 .4 7 1 8 6 2 3 6E+02 5 .4 3 4 3 6 2 3 6 E + 0 2
5 .4 4 5 1 6 2 5 3 E + 0 2 5 .45116253E+02 5 .4 4 1 16253E+02 5 .44116253E+02 5 .4 4 616253E + 02 5 .4 4 6 1 6 2 5 3 E + 0 2
5 .4 9 1 8 1 2 5 4 E + 0 2 5 .49181254E+02 5 . 51001267E+02 5 .51471281E+02 5 .56166267E+02 5 . 5 7 9 4 6 2 5 4 E + 02
5 . 58996254E+02 5 .57916237E+02 5 . 54756691E+02 5 .55081691E+02 5 . 54 0 01735E + 02 5 . 53 3 0 1 7 3 5E +02
5 .4 8 3 0 1 7 3 5 E + 0 2 5 .4 7 7 0 6 6 9 1 E + 0 2 5 .46626734E+02 5 .46076735E+02 5 .4 1 0 5 6 691E +02 5 .4 1 0 5 6 6 9 1 E + 0 2
5 .4 7 8 7 6 7 3 5 E * 0 2 5 .4 7876735E+02 5 .48956691E+02 5 .4 9256691E+02 5 .5 1 1 3 6 7 0 8 E + 0 2 5 .5 2 7 8 6 7 0 8 E + 0 2
5 . 83 9 1 1 2 0 8 E + 02 5 . 93911208E+02 5 . 93911208E+02 5 . 93911208E+02 5 .9 7 9 1 1 2 0 8 E + 0 2 5 . 9 7 9 1 1 2 0 8 E + 02
5 . 95911208E+02 5 . 96611208E+02 5 . 92611208E+02 5 . 92611208E+02 5 . 90611208E+02 5 . 9 0 6 1 1 2 0 8 E + 02
5 . 8 0 6 2 1 7 0 8 E + 02 5 .78746708E+02 5 .69661708E+02 5 .69111708E+02 5 . 7343 6253E+02 5 . 7 3 4 3 6 2 5 3 E + 02
5 .6 8 8 6 6 2 5 3 E + 0 2 5 .68866253E+02 5 .67866253E+02 5 . 67866253E+02 5 .6 7 4 6 6 2 5 3 E + 0 2 5. 746625
6 3E+ 02
5 . 7 1 3 4 6 2 5 4 E +02 5 .71346254E+02 5 . 71346254E+02 5 . 76826254E+02 5 . 77376254E+02 5 . 81196254E+02
5 . 8 1 9 9 6 2 5 4 E +02 5 .81996254E+02 5 . 80536708E+02 5 . 80951708E+02 5 .8 1 3 3 6 7 0 8E+02 5 . 8 0 6 3 6 7 0 8 E + 02
5 .7 9 2 5 1 7 0 8 E + 0 2 5 .77761708E+02 5 . 77761708E+02 5 .77761708E+02 5 . 6 8451708E +02 5 . 6B 451708E +02
5 .7 4 4 2 1 7 0 8 E + 0 2 5 . 78136708E+02 5 . 78136708E+02 5 .78136708E+02 5 .7 8 1 3 6 7 0 8 E + 0 2 5 . 7 9 4 0 1 7 0 8 E + 02
' TELEREG LA G E'
8 . 38000000E+02
6 .2 2 0 0 0 0 0 0E»02 6 . 62000000E+02 6 . 62000000E+02 6 .62000000E+02 6 . 62000000E+02 6 . 62000000E+02
6 . 62000000E+02 6 . 62000000E+02 6 . 62000000E+02 6 . 62000000E+02 6 .6 2 0 0 0 0 0 0E+02 6 .6 2 0 0 0 0 0 0 E + 0 2
6 . 6 2 0 0 0 0 0 0E<-02 7 . 16000000E+02 7 . 70000000E+02 7 .68919983E+02 7 .6 8 9 1 9 9 8 3 E + 0 2 7 . 6 8 9 1 9 9 8 3 E + 02
7 . 68919983E+02 7 . 68919983E+02 7 . 68919983E+02 7 .68919983E+02 7 . 68 9 19983E + 02 7 . 6 8 9 1 9 9 8 3 E + 02
7 .6 8 9 1 9 9 8 3 E + 0 2 7 . 68919983E+02 7 . 14919983E+02 7 .14919983E+02 7 . 14 9 1 9 9 8 3 E + 0 2 7 . 1 6 0 0 0 0 0 0 E + 02
7 .1 6 0 0 0 0 0 0 E + 0 2 7 . 16000000E+02 7 . 16000000E+02 7 .16000000E+02 7 .1 6 0 0 0 0 0 0E+02 7 . 1 6 0 0 0 0 0 0 E + 02
7 .1 6 0 0 0 0 0 0 E + 0 2 7 . 16000000E+02 7 . 16000000E+02 8 . 78000000E+02 8 .78000000E+02 8 .7 8 0 0 0 0 0 0 E + 0 2
7 .1 6 0 0 0 0 0 0E+02 7 . 16000000E+02 7 . 16000000E+02 7 .16000000E+02 7 .1 6 0 0 0 0 0 0E+02 7 . 1 6 0 0 0 0 0 0 E + 02
6 .2 2 0 0 0 0 0 0E+02 6 . 62000000E+02 6 . 62000000E+02 6 . 62000000E+02 6 .6 2 0 0 0 0 0 0E+02 6 .6 2 0 0 0 0 0 0 E + 0 2
6 .6 2 0 0 0 0 0 0E + 02 6 . 62000000E+02 6 . 22000000E+02 6 . 22000000E+02 6 .6 2 0 0 0 0 0 0 E + 0 2 62000000E+02
7 . 7 0 0 0 0 0 0 0E + 02 7 . 70000000E+02 7 .70000000E+02 7 .70000000E+02 7 .7 0 0 0 0 0 0 0E+02 70000000E+02
7 .7 0 0 0 0 0 0 0E + 02 7 . 70000000E+02 7 . 70000000E+02 7 .70000000E+02 7 . 70000000E+02 70000000E+02
7 .7 0 0 0 0 0 0 0 E * 02 7 . 70000000E+02 7 .70000000E+02 7 .70000000E+02 7 .7 0 0 0 0 0 0 0E+02 70000000E+02
7 .70000000E+02 7 . 70000000E+02 7 . 70000000E+02 7 .70000000E+02 7 . 70000000E+02 70000000E+02
7 .7 0 0 0 0 0 0 0 E + 02 7 . 70000000E+02 7 . 70000000E+02 7 .70000000E+02 7 . 70000000E+02 70000000E+02
7 . 70000000E+02 7 .70000000E+02 7 . 70000000E+02 7 .70000000E+02 7 . 70000000E+02 7 0 0 0 0 0 0 0E+02
Lasdy, to show how data from the database can be easily change, the following code
was used to calculate the percentage of time when the marginal cost of energy was higher
than the marginal costs of reserves (see section 3.5.1):
<?php
u s e r_ s c rip ts /o u tp u t/c o n s tra in t_ e n e rg y _ a s _ m o s t_ b in d in g .php

W ritte n by: Y a n n REBOU RS (y a n n . r e b o u r s @ i e e e . o r g )

C re a te d : 2 0 0 8 .0 4 .0 7
L a s t m o d ifie d : 2 0 0 8 .0 4 .0 7
D e sc rip tio n : G e t t h e p e r c e n t a g e o f ti m e w hen NRJ i s th e m ost b in d in g
U tilisa tio n : no in p u t needed
(C ) The U n iv e r s it y o f M an ch ester 2008

( C) E le c tric ite de F ra n c e 2008
* */
m i ntm m
/ 1 MODULES 1 1
nm u nun
req u ire_ o n ce 'p a r a m e te r s .in c '; / / P a ra m e te rs
req u ire_ o n ce 1D a t a b a s e / D a t a b a s e _ O t e s s . p h p 1 ; / / To a c c e s s t o th e O tess d a ta b a se
req u ire_ o n ce /o u tp u t_ s c rip t/O u tp u tD a ta .p h p '; // To s a v e t h e o u t p u t
////////////////
// PARAMETERS //
////////////////
/ / We g e t co n n e cted to th e d a ta b a s e
$ d atab ase = n e w D a t a b a s e _ O t e s s ( ) ,-
/ / We d e f i n e t h e t a b l e s t o s e l e c t
$ a rra y _ ta b le s_ to _ se le c t =
A rray ("d a y ", " tim e s te p _ d is p a tc h _ s ta te m e n t" , "tim e ste p ", " d a ta s e t" ) ,-
/ / W e d e fin e th e f ie ld s to s e le c t
$ a rra y _ fie ld s_ to _ se le c t = a rra y (
A r r a y ( " T a b l e " => " d a y " , " F i e l d " => "D ATE") ,
A r r a y ( " T a b l e " => " t i m e s t e p 1i , " F i e l d " = > " S T A R T " ') ,
A r r a y ( " T a b l e " => " t i m e s t e p d i s p a t c h s t a t e m e n t " , " F i e l d " => " N R J MC" ) ,
A r r a y ( " T a b l e " => " t i m e s t e p d i s p a t c h s t a t e m e n t " , " F i e l d " => " SEC" MC" ) ,
A r r a y ( " T a b l e " => " t i m e s t e p _ _ d i s p a t c h _ s t a t e m e n t " , " F i e l d " => " P R I " "MC" )
);
/ / We s e l e c t t h e d a t a o n t h e m a r g i n a l c o s t s
$ arra y _ a rra y _ m a rg in a l_ c o sts = $d atab ase-
> s e l e c t E l e m e n t s ( $ a r r a y _ t a b l e s _ t o _ s e l e c t , $ a r r a y _ f i e l d s _ t o _ s e l e c t , " ' S C R I P T _ I D ' = ’ 1 1 AND
' t i m e s t e p ' . ' S T A R T ' < 12 4 : 0 0 : 0 0 : 0 0 0 ’ " ) ;
/ / I n i t i a l i s a t i o n of th e co u n ters
$ a r r a y _ t o t a l = a r r a y ()?
$ a rra y _ n rj_ h ig h e s t = a r r a y ();
/ / We g o t h r o u g h a l l t h e d a t a ( i . e . , 4 8 x 8 7 9 d a y s )
fo re a c h ( $ a rra y _ a rra y _ jn a rg in a l_ c o sts as $ array _ m a rg in a l_ c o sts ) {
/ / I n i t i a l i s a t i o n of th e a rra y s
if ( e m p ty ($ a rra y _ n rj_ _ h ig h e st[$ a rra y _ m a rg in a l_ c o sts["S T A R T "]]) ) {
$ a r r a y _ n r j _ h i g h e s t [ $ a r r a y _ m a r g i n a l _ c o s t s [ "STA R T"]] = 0 ;
if ( e m p ty ($ a rra y _ to ta l[$ a rra y _ m a rg in a l_ c o s ts [" S T A R T " ]]) ) {

$ a r r a y _ t o t a l [ $ a r r a y _ m a r g i n a l _ c o s t s [ " START"] ] = 0 ;
/ / A c tu a lis a tio n of th e d a ta
$ a r r a y _ t o t a l [ $ a r r a y _ m a r g i n a l _ c o s t s [ " S T A R T " ] ] ++;
if ( $ a r r a y _ m a r g i n a l _ c o s t s [ " N R J _ M C " ] > $ a r r a y _ m a r g i n a l _ c o s t s [ " P R I M C"] &&
$ array _ m arg in a l_ _ co sts["N R J_ M C "] > $ a rra y _ m a rg in a l_ c o sts[" S E C _ M C " ] ) J
$ a r r a y _ n r j _ h i g h e s t [ $ a r r a y _ m a r g i n a l _ c o s t s [ "START"] ] ++;
/ / We c r e a t e t h e a r r a y w i t h t h e r e s u l t
$ a r r a y _ d a t a _ t o _ i n s e r t = a r r a y (),■
$ t i m e s t e p = 1 ,-
fo re a c h ( a rray _ k e y s($ a rray _ n rj_ h ig h e st) as $ s ta r t ) {
a r r a y _ p u s h ( $ a r r a y _ d a t a _ t o _ i n s e r t , A r r a y { "X "= > $ tim e ste p ,
"Y "= > 1 0 0 * $ a rra y _ n rj_ h ig h e st[$ sta rt]/$ a rra y _ to ta l[$ sta rt] ) );
$ tim e s te p ++;
}
/ / We i m p o r t t h e d a t a i n t o t h e d a t a b a s e
$ o u tp u t _ d a ta = new O u tp u tD a t a {);
/ / N o te : we d e l e t e t h e p r e v i o u s r e s u l t s
314 APPENDICES
$ o u tp u t_ d a ta -> in se rtD a ta { $ a rra y _ _ d a ta _ to _ in se rt, t r u e ) ;
////////////
// RETURN 1 1
illin iu m
e c h o 1<B R / > < A H R E F = " . . / . . / o u t p u t _ s c r i p t / v i e w . p h p " > R e t u r n to th e li s t of o u tp u t

scrip ts< /A > < B R / > ' ;
INDEX
—A — Sealed-bid, 206, 208, 211, 239, 246
ACE, 107,108,110,138 Sealed-bid second price See Auction,

Vickrey
ACF, 168
Sequential, 210, 212, 214
Adaptive deterministic security
boundaries, 86 Simultaneous, 211, 212, 214
Adequacy, 84 Single-sided, 204, 205
ADSB See Adaptive deterministic security Single-unit, 205, 206

boundaries
VCG See Auction, Vickrey-Clarke-
Agent Groves
Pivotal, 129,132 Vickrey, 206, 209
Auction Vickrey-Clarke-Groves, 206, 207, 208,

239
Dutch, 206, 207
AVR See Voltage control, Primaiy
English, 205, 206
— B—
Japanese, 205
Bid
Multiple-unit, 205, 206, 208
First rejected, 214
Philatelist See Auction, Vickrey
Last accepted, 214
315
316 INDEX
— C— Fixed, 144, 145, 146, 149, 198, 219,

224, 238, 239, 246
Congestion, 215
Marginal, 50, 128, 159, 160, 162, 171,
Contract
172, 179, 180, 181, 185, 204, 209,
Bilateral, 50, 52, 92,188, 191, 193,194, 213,214,219, 226, 244, 245
195,196, 215, 220, 236, 238, 248

Opportunity, 72, 147, 148, 197, 199,
Duration, 203, 219 200, 201, 203, 209, 219, 227, 238
Forward, 50,194 Procurement, 88, 209, 210, 211, 223
Future, 194 Reservation, 146,149, 239, 246
Control Running, 232, 240, 247
Speed, 58, 59, 60, 92 Transaction, 90, 145,189,194,195
Controller, 58, 92, 93, 104, 105, 106, 107, Utilisation, 147,149,184, 200, 224, 225
108,109,110,111,116
Variable, 146, 149, 151, 154, 159, 184,
Correlation coefficient, 168,171,173 198, 217, 219, 238
Cost CRE, 100
Allocation, 130, 131, 207, 223, 224, Cross-elasticity, 192
225, 226, 227, 239, 246

Curve
Aumann-Shapley, 226, 227

Demand, 50, 192, 193, 206, 219, 220,
Marginal, 226 239, 246
De-optimisation, 146, 149, 154, 155, Offer, 201, 227
159, 165, 166, 167, 168, 169, 170,

Supply, 50,199, 219, 220
171, 172, 173, 174, 175, 176, 177,
178, 179, 180, 182, 183, 184, 185, Customer damage function, 85
243, 244, 245
— D—
Deployment, 148
Demand
INDEX 317
Curve, 50,192, 193, 206, 219, 220, 239, Ferranti effect, 75

246
Financial transmission tights, 53, 210
Response, 192,195, 237, 238
Free riding, 127, 222, 225
Deployment end, 103, 106, 110
Frequency control
Deployment start, 103, 104, 110
Overlap regulation sendee, 138
Differentiation
Primaiy, 93, 94, 97, 100, 101, 102, 103,
Horizontal, 89 104, 105, 106, 108, 109, 111, 119,
133, 134, 137, 140, 173, 182, 189,
Vertical, 89
196, 200, 201, 212, 215, 216, 217,
Distributed generation, 47, 114, 126, 136, 218, 222, 226, 234, 236
229
Secondaiy, 92, 94, 97, 101, 102, 107,
108, 109, 110, 120, 122, 123, 124,
125, 133, 134, 137, 138, 140, 144,
Economic efficiency, 230, 232, 233, 235, 153, 157, 177, 192, 193, 196, 198,
237, 240, 247 200, 201, 202, 212, 215, 216, 217,
218, 221, 222, 224, 230, 234, 236,
EENS See Expected Energy N ot Served
248
Elasticity, 127, 141, 185, 189, 192, 211,
Supplemental regulation service, 138
220, 237, 243, 247, 248
Tertiary, 94, 97, 101, 102, 120, 122,

ERAP, 189, 190, 191, 192, 193, 194, 205,
123,124,125
206, 209, 210, 211, 213, 218, 219, 223,
224, 238, 245 FTR See Financial transmission rights
Exchange See Market, Decentralized Full availability, 103,106,110
Expected Energy N ot Served, 85 — G—
Externality, 187,195, 210, 213
— F— Turbine, 78, 116, 159, 163
FACTS, 80,116
318 INDEX
Generating unit, 43, 45, 46, 47, 51, 52, 54, — K—
55, 56, 58, 59, 60, 63, 64, 68, 70, 71, 72,
Kolmogorov test, 125
77, 78, 79, 80, 81, 92, 93, 94, 95, 98, 99,
101, 103, 105, 111, 116, 120, 122, 138, — L—
139, 147, 148, 173, 187, 190, 198, 215,
218, 225, 230, 241, 248 Lagrange multiplier, 158
— H— Lagrangian function, 158
HHI See Index, Herfindahl-Hirschman Load-serving entity, 189, 223
LSE See Load-serving entity
Incentives, 46, 125, 126, 127, 128, 129, — M—
132, 140, 141, 188, 189, 190, 191, 192,

Market
207, 213, 214, 215, 222, 223, 225, 227,
228, 230, 231, 232, 236, 237, 238, 239, Centralised, 50, 52, 203, 211, 239, 246
240, 242, 243, 246, 247
Clearing, 46, 188, 192, 202, 211, 214,
Penalties, 158, 159, 162, 199, 222, 223, 216, 217, 218, 221, 238, 245
228, 229, 230, 231, 240, 247
Concentration, 193, 213, 233, 235, 236
Rewards, 222, 228, 229, 231, 240, 247
Coordination, 212
Index
Decentralised, 49, 50, 52, 54, 64, 142,
Herfindalil-Hirschman, 235 203, 211, 239, 243, 246, 249
Lerner, 233 Liquidity, 50,199, 233, 236
Residual Supply, 235 Monitoring, 48, 137, 222, 228, 229,

230, 231, 232, 240, 246, 247
Indicator
Settlement, 203, 205, 213, 214, 215,
Cost, 234, 235
216, 239, 246,248
Reserve, 133,134,135
Spot, 49, 52, 147, 150, 188, 193, 194,
Insecurity See Security 195, 196, 200, 208, 219, 230, 233,
238
INDEX 319
Timing, 200, 216, 218, 219 Common, 214
Transparency, 193, 194, 195, 227, 230, Decomposition, 158, 209

246
Ex-ante, 53, 88, 213, 228
— N —
Ex-post, 53, 213, 228, 230
Nash equilibrium, 51
Nodal, 53, 203, 213, 215, 239, 246
Net Surplus See Profit
Separate, 214
Normality, 123,124,141
Sign, 197, 202
— O—
Symmetry, 202
Organisation
Zonal, 203, 213, 214, 215, 239, 246
Centralised, 107
Price cap, 46, 188, 192, 220, 221, 222,
Hierarchical, 107 239, 246
Pluralistic, 107 Offer, 220, 221, 222
OTC See Over-the-Counter Purchase, 220, 221, 222
Over-the-counter, 50, 51, 52, 54,194 Prime mover, 55, 58, 69,114
— P— Problem
PACF, 161,168 Dual, 157, 158, 209
Pareto efficiency, 51, 88, 128, 129, 205, Primal, 157, 209
206, 207, 208, 239, 246
Product substitution, 117, 128, 210, 211
Payment
Profit, 147,199, 223
Structure of, 197
Property rights, 187, 213, 215, 239
Pool See Market, Centralised
Public good, 43, 84, 87, 127, 128, 129,
Price 130,141, 222, 223, 243
320 INDEX
— Q— —T—
Quality Tax
O f supply, 131 Groves-Clarke, 129,130, 207
Power, 84, 86, 87, 118, 129, 140, 242, Tendering process, 105, 191, 194, 195,
248 196, 206, 208, 228
Quasi-steady state, 57, 58, 59, 61, 77, 103, Transformer, 47, 71, 72, 79, 86, 145, 146,
114 148
— R— Transmission tariff, 54, 152, 223, 224,

239, 246
Rational buyer, 209, 211
— V—
Reliability, 84, 85, 86, 87, 99, 118, 119,
140, 211,224,228 Value
Risk, 50, 52, 86, 92, 127, 146, 149, 191, Marginal, 214, 215
193,194,195,219, 225, 238
O f lost load, 85, 221, 222
RSI See Index, Residual supply
VOLL See Value, O f lost load
— s —
Voltage control
Security, 242
Basic, 196, 201, 215, 216, 217, 218,
Cost of, 86,128,132 222, 234
Level of, 85, 86,119 Enhanced, 196, 200, 201, 215, 216,
217, 218, 222
Smart grid, 92
Primary, 71, 98, 111
Stability, 91, 92, 93, 97, 111, 114, 249
Secondary, 98
Stationarity, 141
Tertiary, 98
Substitution
Voting, 127
Rate of, 128
INDEX 321
-W— Global, 44, 88, 118, 142, 195, 209, 210,

232, 242, 248, 249
Welfare

A Comprehensive Assessment of Markets For Frequency and Voltage Control Ancillary Services

Uploaded by

Copyright:

Available Formats

A Comprehensive Assessment of Markets For Frequency and Voltage Control Ancillary Services

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Comprehensive Assessment of Markets For Frequency and Voltage Control Ancillary Services

Uploaded by

Copyright:

Available Formats

A Comprehensive Assessm ent of

Markets for Frequency and

A diesis submitted to The University o f M anchester for die degree of

in die Faculty o f Engineering and Physical Sciences

School o f Electrical and Electronic Engineering

All rights reserved

INFORMATION TO ALL USERS

In the unlikely e v e n t that the a u thor did not send a c o m p le te m anuscript

Published by ProQuest LLC(2019). C opyright of the Dissertation is held by the Author.

All rights reserved.

A cknow ledgem ent 23

1.1 Increasing Welfare o f Power System Users 43

1.3.3 Transmission markets 53

1.4 Fundamentals o f Frequency Control 55

1.5 Fundamentals o f Voltage Control 65

1.6 Technologies Used to provide Ancillary Services 77

C hapter 2 D elivery of ancillary services 83

2.2 Needs o f Users for System Services 84

2.3 Specification o f the Quality o f Ancillary Sendees 88

2.4 Quantity o f Ancillary Sendees 118

2.5 Location o f Ancillary Sendees 135

2.6 Conclusion 140

C hapter 3 Cost of ancillary services 143

3.1 Introduction 143

3.2 Main Cost Components o f Ancillary Services 144

3.3 M ethodology and Hypotheses to Estimate Cost 149

3.4 Day-Ahead De-Optimisation Cost for a Producer 165

3.4.1 De-optimisation cost over two and a half years 165

3.5 Marginal Costs o f Frequency Control for a Producer 179

3.6 Cost o f Time Control in France 182

3.7 Conclusion 183

C hapter 4 P rocurem ent of S ystem Services 187

4.1 Introduction 187

4.2 Nom inating the Entity Responsible o f Procurem ent 189

4.3 Matching Supply and D em and 190

4.4 Choosing the Relevant Procurem ent Methods 193

4.5 Defining the Structures o f Offers and Payments 197

4.6 Organizing the Market Clearing Procedure 202

4.7 Avoiding Price Caps 220

4.8 Providing Appropriate Incentives 222

4.9 Assessing the Procurem ent M ethod 232

4.10 Summary 238

C hapter 5 Conclusions an d F u tu re R esearch 241

5.1 Rationale for the Thesis 241

5.2 Contributions to Knowledge 242

5.3 Short-Term Evolutions Desirable in France 247

5.4 Suggestions for Future W ork 248

References by categories 279

A .I.3 Time-dependent data 295

A .I.4 Descriptive analysis 296

.2 T ech n ical D e sc rip tio n o f O T E S S 299

A.2.1 Basic functionalities 299

A.2.2 Hardware architecture 300

A.2.3 Software architecture 301

A.2.4 Parameters for data analysis 304

A.2.5 Screenshots 307

A.2.6 Code example 310

Approximate number o f words in the main text (Chapters 1 to 5): 58 000

Table 2.1: General versus precise specifications of AS qualities 90