Degree project in
Capacity Market
Arthur Géze
Stockholm, Sweden 2013
XR-EE-E3S 2013:002 Electric Power Systems
Second Level,
KTH Royal Insitute of Technology
Master Thesis
Modelling and simulating the French capacity market
Author:
Arthur G` eze
Supervisor:
Yelena Vardanyan
Examiner:
Dr Mohammad R Hesamzadeh
A master thesis submitted in fulfilment of the requirements for the master degree in power systems engineering
in the
Electricity Market Research Group (EMReG) KTH Royal Institute of Technology
February 2014
I, Arthur G` eze, declare that this thesis titled, ’Modelling and simulating the French capacity market’ and the work presented in it are my own. I confirm that:
This work was done wholly or mainly while in candidature for a degree at this University.
Where any part of this master thesis has previously been submitted for a degree or any other qualification at this University or any other institution, this has been clearly stated.
Where I have consulted the published work of others, this is always clearly at- tributed.
Where I have quoted from the work of others, the source is always given. With the exception of such quotations, this thesis is entirely my own work.
I have acknowledged all main sources of help.
Where the master thesis is based on work done by myself jointly with others, I have made clear exactly what was done by others and what I have contributed myself.
i
The author would like to emphasize the fact that all data used during this master thesis
are either public data or personal estimations and are not in any way representative of
RTE’s reality. The results shown below are thus based on arbitrary hypothesis which
do not reflect in any way the structure of the future French capacity market.
Abstract
Electricity Market Research Group (EMReG) School of Electric Engineering
Master Thesis
Modelling and simulating the French capacity market by Athur Geze
As a capacity mechanism is currently being designed in France, a new capacity market will be created. The French transmission system operator RTE needed to conduct some studies on this market behaviour and the parameter influences. This report analyses in detail the French capacity mechanism and presents a model of the capacity market in order to study it. It then introduces a simulator called CL´eMix that uses this model to run Monte Carlo and Agent based simulations. Several studies concerning the actors strategies and possible use of market power are then presented and their results analysed.
En kapacitetsmekanism ska appliceras i Frankrike och ska skapa en kapacitet marknad. Fran- ska systemasvarigen beh¨ovde att genomf¨ora studier i den h¨ar marknadens beteende och sina parametrars p˚averkan. Denna rapporten analyserar i detalj den franska kapacitetsmekanismen och presenterar en modellering av kapacitetsmarknaden f¨or att studera den. Sedan introducerar det en simulator som hetter CL´eMix och som anv¨ander den h¨ar modelen f¨or att k¨ora Monte Carlo och Agent Based simuleringar. Flera studier om akt¨orernas beteende och eventuell anv¨andning av marknadsmakt presenteras och sina resultat analyseras.
Acknowledgements
The author would like to thank the team at KTH consisting of Yelena Vardanyan and Mohammad Hesamzadeh, who both helped him write this report and follow the right directions for his work.
The author also wishes to thank Aur` ele Fontaine, for his essential help as a tutor for the master thesis at the RTE Market Department, as well as the rest of the (ME)
2division and Market Department, including Silvano Domergue, Arthur Hubert, and others for the provided help in making this master thesis a success and for the kind welcoming inside the company during the 5 months the master thesis lasted.
iv
Declaration of Authorship i
Abstract iii
Acknowledgements iv
Contents v
List of Figures vii
List of Tables viii
1 Introduction 1
1.1 The reasons behind the European capacity mechanisms . . . 1
1.2 Overview of the French electricity market . . . 2
1.3 The French capacity mechanism. . . 4
1.4 Problem definition . . . 5
1.5 Objectives . . . 6
1.6 Overview . . . 6
2 Modelling of the capacity market 8 2.1 Mechanism’s detailed description . . . 8
2.1.1 Certification . . . 9
2.1.2 Redeclaration . . . 9
2.1.3 Retailer actions . . . 11
2.1.4 Capacity market . . . 11
2.1.5 Peak period . . . 11
2.1.6 Post PP market . . . 12
2.1.7 Differences payment . . . 12
2.2 Modelling of the capacity market . . . 13
2.2.1 Modelling of the market’s elements . . . 13
2.2.2 Module Structure. . . 15
3 Simulator structure 18 3.1 General structure . . . 18
3.2 Database and other entrance parameter . . . 19
3.3 Actor’s Strategies. . . 20
3.3.1 Market Strategies. . . 20
3.3.2 Redeclaration strategies . . . 21
3.4 Agent Based Model. . . 23
3.4.1 The Roth Erev Algorithms . . . 23
v
Contents vi
3.4.2 Nash Equilibrium. . . 25
3.4.3 Implementation of the Model . . . 25
4 Simulations 26 4.1 Inputs . . . 26
4.1.1 Capacities . . . 26
4.1.2 Clients. . . 27
4.1.3 Peak consumption control projects . . . 27
4.1.4 Parameters . . . 27
4.1.5 Scenario . . . 28
4.2 Simulations’ Results . . . 28
4.2.1 Choice of the best redeclaration strategy. . . 28
4.2.2 Comparison of the market strategies . . . 29
4.2.3 Influence of renewable energies . . . 30
4.2.4 Certification strategies for a short actor . . . 31
4.2.5 Certification strategies for a long actor. . . 31
4.2.6 Use of market power by long actors. . . 33
5 Closure 35 5.1 Summary of the thesis work . . . 35
5.2 Conclusions concerning the model and the simulator . . . 35
5.3 Conclusions from the simulations . . . 36
5.4 Recommendations for future studies . . . 36
A Model’s Database 37
B Simulation’s parameters 39
Bibliography 41
1.1 French electicity mix . . . 2
1.2 European HHI . . . 3
1.3 French compared consumptions . . . 4
1.4 Capacity market principles . . . 5
2.1 Mechanism’s temporality . . . 8
2.2 Example of redeclarations . . . 10
2.3 The 3 types of capacities. . . 14
2.4 Simulator’s modules . . . 16
2.5 Bottleneck penalty model . . . 17
3.1 Simulator’s structure . . . 18
3.2 The usual 4 year scenario . . . 19
3.3 Basic bidding market strategy. . . 21
3.4 Temporality for Bj¨orn . . . 22
3.5 Temporality for Helmut . . . 22
3.6 Temporality for Enzo. . . 23
3.7 The Roth-Erev Algorithm . . . 24
3.8 Agent based module diagram . . . 25
4.1 Simulations’ scenario . . . 28
4.2 Redeclaration strategies’ results . . . 29
4.3 Market strategies’ results . . . 30
4.4 Simulation’s results on renewable energies . . . 31
4.5 Certification strategies (short actor) . . . 32
4.6 Certification strategies (long actor) . . . 32
4.7 Market’s reference state . . . 33
4.8 Capacity retention actors’ results . . . 33
4.9 Capacity retention compared results . . . 34
vii
List of Tables
2.1 List of objects and properties . . . 15
3.1 Simulator’s parameters. . . 20
3.2 Redeclaration strategies’ signals. . . 23
3.3 Roth-Erev symbols . . . 24
A.2 Model’s database . . . 38
B.2 Simulation’s parameters . . . 40
viii
Introduction
1.1 The reasons behind the European capacity mechanisms
The European capacity owners are facing a major profitability issue. Due to the discovery of large amount of oil shale gas sources in North America [1], the USA and Canada became net energy exporters and are able to provide the European energy market with a cheap income of coal [2]. The European consumption of American coal has thus increased almost 50% between 2010 and 2011 [3]. This allowed the old coal plants to compete with the newly constructed gas plants in the merit order of the electrical market bids in most European countries. This phenomenon was even amplified by the extremely low cost of the CO2 ton in the European market, making highly polluting coal plants relatively more profitable than new low emission combined cycle gas plants [2] [4].
This change in the merit order made most of the gas turbines and combined cycled gas plants, that were mostly built in the late 2000’s [5], non profitable in an ”energy-only” electric market.
A lot of owners are starting to consider closing them in order to save their operational costs, and to wait for better conditions [6]. A massive shutdown of these plants would be highly damaging for the whole European grid and for local markets in many countries; especially during winter peak periods when gas turbines and combined cycle plants are highly needed in order to maintain the balance of supply and demand.
In order to deal with this problem, most European countries are planning to take actions. The implementation of a capacity mechanism is one of them. For the moment Several EU countries have started implementing a capacity market. These mechanisms, even if they share a common denomination, are completely different in their constructions and goals. These mechanism in- cludes the Spanish capacity mechanism that mainly consists of subsidies to the plant owners by daily capacity payment to flexible combined cycle plants [7], or the British mechanism that relies on a centralised auction for capacity four years in advance [8].
1
Chapter 1. Introduction 2
1.2 Overview of the French electricity market
The French capacity mechanism was built in order to fit perfectly the settings of the French electricity market. This market is unique in Europe because of several elements:
First, French the energy mix is highly dominated by nuclear production [9]. Hydropower and non- renewable energies represent the most of the remaining production. Other renewable energies, despite of their fast growth, are still far behind [9].
Figure 1.1: The French electricity generation mix
Secondly, the French electricity market is highly concentrated, economically speaking. It consists of very few big actors, which own most of the market shares giving them higher possibilities of using market power, compared to a perfect market consisting of a huge number of small actors.
The concentration of a market can be estimated by using the Herfindahl–Hirschman Index, or HHI. It consists of the sum of the squared market shares of all the actors as described in the following formula:
H =
N
X
i=1
s2i (1.1)
Here H is the value of the of the HHI, with N being the total number of actors on the market, and sibeing the market share (in %) for the actor i. The HHI is therefore between 0 (theoretical
value for a market with an infinity of small actor) and 10.000 (for a monopoly with one actor owning 100% of the market).
With an electric generation market’s HHI of 8800 (the maximum HHI index being 10.000 for a single actor market), France ranks amongst the five most highly concentrated electricity markets in the EU [9] as shown in figure 1.2. Some actors of the French electricity market thus posses the possibility for important uses of market power.
Figure 1.2: European HHI indexes [10]
This situation is mostly due to the fact that the historical public electricity provider and producer, EDF (Electricit´e de France), still owns all the nuclear production, most of the hydro-plants, and almost all the gas turbines [11]. In order to reduce this actor’s market power, the first part of the NOME law (Nouvelle Organisation du March´e de l’Electricit´e: New Organisation of the Electricity Market), consisted mainly in introducing the ARENH (Acc´es R´egul´e `a l’Electricit´e Nucl´eaire Historique: Regulated Access to the Historical Nuclear Electricity). This disposition allows alternative electricity retailers to pay EDF in order to buy its nuclear energy at its production costs. This disposition extends to half of EDF’s nuclear production, which is sold at its production costs to small retailers [12].
Finally, the French electric consumption is highly thermo-sensitive. This is due to the extended use of electric heating. France thus accounts for almost half of the total European thermo sensitivity; and even if it is a power-exporter during most of the year, it becomes a power-importer during the winter peak-period [2]. Such a difference of almost 50% in the consumption is easily noticeable in the figure1.3where the curves represent the total French instant consumptions for January 1st 2013 (red curve) and June 30th 2013 (green curve):
Chapter 1. Introduction 4
Figure 1.3: French raw consumption curves for summer and winter (RTE website data)
1.3 The French capacity mechanism
The French capacity mechanism is set in place by the second part of the NOME law that set in place a capacity obligation for all electricity retailers [12]. Compared to all the other European capacity mechanism, it was designed to be non-centralized and adaptable. The formal rules were set in place by a decree issued in December 14th 2012 [13] (a decree is defined as: ”an official order that has the force of law ” [14]). The goals of this mechanism are mainly to ensure the national security of the electric supply during the consumption peak period, but also to encourage consumption control and power shedding during this period, thus lowering the thermo-sensitivity of the system.
The mechanism is based on a simple idea: in the one hand, the energy retailers are given an obligation to possess a certain amount of capacity certificates depending on their customers’
consumption during the peak period, and on the other hand capacity owners are given these certificates in exchange for their commitment to have their capacities available during the peak period. The amount of certificates given to a capacity owner depends on its available capacity during the peak period.
It thus creates a demand for certificates from retailers, and a supply coming from the capacity owners, leading to an over the counter (OTC) market for capacity certificates. This market is supposed to help the capacity owners fix their non-profitability problems, and to encourage installation of the right amount of capacity to ensure the system safety during the peak period.
Moreover, the price of the certificates is supposed to reflect the system’s security costs.
The figure1.4explains graphically the basic principle of the mechanism: the retailers’ obligations create a demand for certificates, while certification of capacities creates a supply (the exact amount of supply and demand being defined following a system security criterion). A capacity market thus occurs, that is supposed to reflect the system’s security cost.
Figure 1.4: Basic principles of the French capacity market
The main principles and the global organisation of this market are described in the decree [13], but the final set of rules, describing the parameters’ values and the exact organisation of the market is to be issued on November 2013.
1.4 Problem definition
The created capacity market however was not really studied in detail. A lot of questions are raised by the future market actors, but also mainly by the French transmission system operator (TSO), RTE. They are concerned about the future behaviour of the capacity market and the influence of the mechanism parameters (price of penalties, number of possible re-declaration of availability for the capacity owners...) as the final rules are currently being discussed and negotiated between RTE and the actors (EDF, GDF-Suez, E.On,...).
RTE thus needed to develop a simple and efficient model for the capacity market, and a simulator allowing them to conduct some studies on the influence of parameters and the actors’ strategies.
With these studies RTE should be able to have concrete and precise arguments in the discussions explaining all the actors which are the best values for the system.
At first, the capacity market modelling was created for a serious game that was to be used for educational purposes. Another RTE department with more experience on the development of serious games later took charge of this project. This future game is called CL´eM and is to be
Chapter 1. Introduction 6
finished around September 2013 and to be used for internal workshops, or for the actors to gain a better knowledge of the market workability.
1.5 Objectives
This master thesis was conducted at the RTE office in La D´efense (Paris’ business district), at the Market Department in the (ME)2 group (Mod´elisation de March´e et Etudes Economiques:
Market Modelling and Economic Studies) between January 2013 and June 2013 under the su- pervision of Mr. Aur`ele FONTAINE, member of the (ME)2division. Its goals were to:
• Develop a model for the capacity market
• Use this model to create a simulator in Visual Basic for Applications (VBA)
• The simulator must be easy to use, well documented, and easily adaptable to the future market rules
• The behaviour of the actors must be changeable, following several strategies
• Several basic studies must be conducted, trying to figure out the actors’ best strategies and possible use of market power.
This Master thesis was realised in conjunction with Arthur HUBERT, student at the ´Ecole nationale sup´erieure des mines de Paris (Mines ParisTech). The market model was initially developed with him, and then while the author developed the simulator and the actor strategies, Mr. HUBERT wrote several theoretical studies that were useful to the author for the simulator [15]. The final version of the studies and of the simulator were thus conducted by both Arthur GEZE and Arthur HUBERT, in order to validate Mr. HUBERT’s theoretical work, and Mr.
GEZE’s simulator structure.
As explained, the master thesis only focuses on the capacity market part of the whole mechanism.
The formulas calculating the exact amount of certificates given to an actor or the amount of obligations will not be taken into account by this work. Other internal studies from RTE focus on this particular problem. This work will thus consider a capacity only by the number of certificates it gives to its owner, and won’t consider its formal availability, capacity, or the length of the peak period. The trading or certification terms will also not be taken into account, as with the exchange formal terms of the OTC market.
1.6 Overview
The next chapter presents the capacity mechanism in detail and shows the different actions the actors can initiate as well as the part of the mechanism, whose rules are not fixed yet. This is followed by the model, which is derived from the mechanism and its structure choice and hypothesis.
Chapter 3 discusses the simulator (called CL´eMix), and how the model was coded in VBA, mainly by explaining the module structure and parameters. Then, the different actors’ strategies (bots) are exposed.
Chapter 4 presents various simulations and studies obtained using CL´eMix, and reflections on the actors’ strategies and the existence of market power. It thus tries to prove the model’s usefulness and efficiency.
Finally, the last chapter draws conclusions about the master thesis work, and displays some thoughts and leads for the future development and improvements of the model.
The appendix section contains data used in the simulator in order to perform the chapter 4.
Appendix A contains the database used to simulate the market conditions. Appendix B contains the simulations mechanism parameters.
Chapter 2
Modelling of the capacity market
2.1 Mechanism’s detailed description
The principles of the mechanism are described in a decree published in the French Official Journal [13]. Even though its exact rules are still being discussed, RTE needed a simulator capable not only of simulating quite precisely the future market, but also of adapting to all the possible rules.
This is the reason why most of the hypothesis’ chosen for the modelling are based only on the decree rules, and allow any possible variation around these decree rules.
The temporality of the mechanism is really important to understand, which will allow seeing all the possible actions for the actors. Figure2.1displays the temporality of the mechanism as fixed by the decree rules.
Figure 2.1: French capacity mechanism’s temporality as described by the decree [13]
8
2.1.1 Certification
First, all existing capacities must be certified by their owners. The exact date of this first certification will depend on the type of capacity, but will take place between four to three years before the delivery year for most of the capacities. The certification is based upon a contract between RTE and the capacity owner, and gives him capacity certificates in exchange for a certain available amount of power during the peak period. Since the amount of certificates largely depends on data given to the TSO by the capacity owner and follows a future formula known by both parties, the capacity owners still possess some control over the number of certificates they will receive, and thus could be able to base a strategy on the use of under or over-certification.
Then, during the rest of the year, newly installed capacities can also be certified following the same protocol as for already existing ones. However, the certification dates can be much closer to the delivery year (and might even be just before for power shedding certification). This ability to adapt the amount of certified capacities during the 4 years before the delivery period is unique to the French market and was designed in order to help the market supply of certificates adapt to the demand: if the demand is high because of under-capacity, the price for certificates will be high and capacity owner will be encourage to invest in new projects for capacity or power shedding, or by making their Combined Cycle Generation Plants (CCGP) available in order to gain the missing-money from the capacity market. On the other hand, if the supply is too high, the price will be too low, and actors will be encouraged to shut some plants for this year, resolving the over-capacity issue.
Once certified, a capacity (existing or newly installed) is included in the Certified capacities register (Registre des capacit´es certifi´ees). This public register lists all the certified capacities and their exact certification values, in order to help market actors and the TSO to estimate the total supply of capacities in the market. Another register exists: the Capacity Guarantees Register (Registre des garanties de capacit´e). This register is used by the TSO to keep track of all the exchanged certificates and transaction on the capacity market, in a confidential way.
2.1.2 Redeclaration
The redeclaration mechanism (r´e´equilibrage in French) was instituted in order to help the actors to adapt to variations of their availability during the four year period after the initial certification and to be aware of the real certificates supply as capacities availability may vary during the years before the Peak Period (PP) of the delivery year. The redeclaration application must come from a Certification Perimeter Manager or CPM (Responsable de P´erim`etre de Certification). This actor is financially responsible for any difference between the certified value and the actual availability observed during the peak period. A CPM can thus be a capacity owner, but can also consists of an association of several small capacity owners willing to reduce their financial risks by summing their small and fluctuant capacity levels into a less fluctuating big capacity level.
The redeclaration is basically a new contract replacing the old certification contract, and setting the certified value for the perimeter of plants to a new value.
Chapter 2. Modelling of the capacity market 10
In order to encourage the actors to certify their capacities directly at the right value, a small premium fee will be applied at each redeclaration. This premium must be proportional to the absolute value of the difference between the new and the old certification levels as the figure 2.2is explaining. For example if an actor decides to redeclare its capacity 50 MW higher than his first forecast, he will pay a fee proportional to 50 MW. And If he later decides to redeclare to 120 MW, he will have to pay a fee proportional to 30 MW (the difference between 150MW and 120MW). This would make his total fee proportional to 80 MW even though the difference between his last and first forecast is only 20 MW.
Figure 2.2: An example of the redeclaration system
Moreover, if the actor is redeclaring lower than his previous certification value, he must ”hand back” the difference in certificates to the TSO in order to keep the good number of certificates on the market. Similarly, if the new value is higher than the old one, the actor is given newly created certificates from the TSO. Once redeclared, the new certification value is of course reported in the Certified Capacities Register for everyone to be informed of the change in the total supply.
It is not decided yet if it will be possible to redeclare only during certain occurrences, or con- tinuously during the whole 4 years before PP. Furthermore, it has not been decided for the moment whether a maximum number of redeclaration for an actor will be instituted. The value of the premium, and the possibility of a premium which value will increase as PP is approaching (making early redeclarations less expensive than last minutes ones) are still under the discussion too.
Finally, the only decision concerning the redeclaration mechanism was that it will not be possible to redeclare after, or during PP, because at the end of PP all the actors are able to calculate their own actual level of availability. They could thus all redeclare their certification values at this value, causing all differences between the certified capacity values and the actual capacity values to be null. This would no longer push actors to certify their capacities at their best-forecast value, which is one of the mechanism’s first goals.
2.1.3 Retailer actions
On the retailer side, actions are more limited. As seen in Figure2.1, the retailers can only have an estimation of their final capacity obligation (the formula used to calculate it will be public).
However, they can decide to invest in order to lower their consumer consumption during the peak period, in order to reduce their obligation level.
2.1.4 Capacity market
During this 4-year period before PP, the actors also can trade certificates in addition to all the previously seen actions. This creates the capacity market, which is the main issue of this thesis.
The capacity market will work as an Over The Counter market (OTC), working by bilateral contract between two actors agreeing to trade capacity certificates in exchange of payment.
All the certificates transaction must be reported to RTE, who will note them in the confidential Capacity Guarantees Register. This will allow RTE to be able to calculate every actor’s certificate balance after PP.
The decree still allows the mechanism to use a fixing-based exchange mechanism with market sessions if the French minister in charge of energy decides so [12].
During the whole period when the capacity market takes place, RTE will regularly publish and update the Certified Capacities Register, giving thus the actors the exact amount of certificates issued on the market and their initial owners. Furthermore, RTE has to not only issue a prevision of the global obligation for the delivery year, every year during the four years preceding the delivery year, but also issue the actual amount of certificates exchanged this year, and the average price of the exchanges. This data will allow actors to have a rather precise view of the global demand for certificates and of the market values.
2.1.5 Peak period
During the delivery year, PP exact location is not the same for calculating the retailers’ obligation or the capacity owners availability. Actually two different definitions will be used: PP1 is used for calculating the retailers’ obligation, based on their clients’ consumption during this period.
PP2 is used in order to calculate the capacities availability, based on the actual availability during this period.
The reason why these 2 different definitions are used mostly come from the uses of power shed- ding. A retailer that has invested in a consumption control project can use it in two ways.
First, he can use it in order to reduce his clients’ consumption during the peak period, and thus his final obligation. The second possible use of a consumption control is to declare it as power shedding.
Power shedding can be certified exactly like capacities. A 10MW power shedding can thus be considered as a 10MW capacity in term of certificates, as adding 10MW to the power production
Chapter 2. Modelling of the capacity market 12
or withdrawing 10MW from the power consumption can be considered the same in term of system’s generation-consumption equilibrium. An actor can thus be given certificates in exchange for making his power shedding available to be activated during the peak period, exactly like a capacity owner receiving certificates in exchange for making his capacities available.
Therefore, an actor in possession of a consumption control project can choose between using it in order to reduce his obligation (by activating it during the peak period), or obtaining certificates (by declaring it as power shedding and leaving it available during the peak period). The point is that with the first method the project must be activated during all the peak period whereas, with the second method the project is just required to be available in case of electric stress on the system, and might indeed not even have to be activated.
Therefore, if PP1 and PP2 were the same period, actors would always choose the second method.
This is why, by separating PP1 and PP2 (and maybe making PP2 ”longer”), the actor is given a real choice between having to activate their peak control during a short period (PP1), or making their power shedding available during a longer time.
2.1.6 Post PP market
Finally, after the end of the peak period, all actors are given their exact level of availability and obligation by RTE. It starts the period known as the Post PP market. During this last market that lasts 15 days (between the moment the actors are given their exact levels, end the moment all imbalance fees in obligation or availability are due) the actors can exchange certificates for the last time. Furthermore, the NOME law states that all long actors must make a public offer for their certificates [12]. Studies have shown (and this idea was later confirmed by the simulator) that during this market, since the amount of certificates on the market is fixed (no more certification or redeclaration) and that the exact obligation is known, the prices will tend toward two extreme prices.
If the market is long (meaning that certificate supply is bigger than demand), the price of certificates will tend to zero. Therefore, all actors at the end will be long (ie with more certificates than their obligation) or at the equilibrium.
On the contrary if the market is short (the supply of certificates is not as important as the demand), the price will tend toward the price of the penalty, which is known exactly at this moment. All actors will thus be either short (they have more obligation than certificates) or at the equilibrium because all formerly long actors had to sell their certificates.
2.1.7 Differences payment
The last part of the mechanism consists in penalties for actors whose final certificate account is non-null. All short actors must pay a fine proportional to their amount of missing certificates, while long actors receive money. However, the financial balance of the account used for collecting negative differences and paying positive ones cannot be negative. Therefore, positive differences
can be paid at a lower price than the normal penalty price for positive obligation differences.
Actually, in the case of a long system, since all actors are long or null, there will not be any payment for positive differences.
The differences between certification and the real availability are paid using the same type of mechanism with a positive only account. Note that the penalty prices might not be the same, and their formulas also are completely different (even though RTE internal studies shows that it would not be a good market design because of the presence of integrated actors with both clients and capacities).
2.2 Modelling of the capacity market
As explained before, even though the global design of the mechanism is defined, some parts are still not agreed on. The modelling must thus be sufficiently open to allow all the problems to be simulated even with rather important changes on the mechanism parameters or on the simulation sequence. This is the reason why a modelling based on sequences of modules representing the different mechanism phases, and allowing each actor to only perform one type of actions per module was chosen.
The model also had to be particularly focused on the capacity market. The cross interactions between this market and the electricity market were not simulated and most of the parameters depending on the electricity market were considered totally external, even though the capacity market is closely linked to the electricity market.
2.2.1 Modelling of the market’s elements
In order to simulate the market, 3 types of objects were considered necessary. Each actor thus plays the game with a portfolio of various occurrences from these 3 types in his portfolio:
Clients
This object is defined by the value of obligation it gives to its owner. It also has a certain value of deviation, in order to simulate the different type of clients (industrial, households...) which has larger or smaller values for the deviation depending on their sensitivity to external factors (temperature, economic situation...). The possibility for a retailer to ”get rid” of a client was not taken into account in the simulator since it focuses only on the capacity market, and tries to separate it from the energy market.
Peak consumption control projects
These projects were simulated as options. The actor owning one of this project can chose to pay a small fixed amount of money in order to obtain the option during the investment phases preceding the PP phase. Then during the PP phase, all actors that paid for the option can chose to activate the consumption reduction for a certain price per MW (generally quite high).
Chapter 2. Modelling of the capacity market 14
If they do, their obligation will be reduced by a certain amount of MW defined by the project’s proprieties.
These objects are thus defined by: the total amount of reduction they cause to their owner, a deviation on this value, an option price (generally low), and an activation price (generally high). In most cases, the option price is relatively lower than the activation price. The first one only represents some investments in power reduction capacity, while the second represents the compensations offered to customers in exchange of their consumption being shedder and is thus much higher than the option price.
Capacities
These objects can be divided into 3 categories described in the figure2.3:
Figure 2.3: The 3 types of capacities
The first category accounts for the already existing and profitable capacities such as nuclear plants, hydropower plants, and most wind and solar plants (which as seen later were not studied in this thesis). These plants will always be available in the electricity market and thus will always be certified during the initial certification.
The second category represents all the existing plants that are no longer profitable on an energy only market, such as CCGP or Gas turbines. These capacities have non-null missing money on the energy market, and one of the goals of the capacity mechanism is actually to help resolve this missing money problem. These capacities can appear on the capacity market if the price for certificates is high enough, but will not be opened for exploitation if the price is too low.
The last category includes all the capacities still in project (including power shedding). Here again, if the price is high enough actors will invest in these projects and new capacities will appear. On the other hand, if the price remains low (often a sign of over-capacity), the actor will not invest, and these capacities will not appear.
From the capacity market’s point of view for a given year, already existing non-profitable capac- ities can be considered as capacities in project, just by seeing their missing money as investment costs. Therefore, the last two categories were merged into one single category of capacities with non-null missing money.
The two types of capacities are thus:
Existing capacities:These capacities are already profitable and thus have a null missing money.
They are defined by their availability, their deviation, and a probability of daily non-availability during PP.
New capacities:These capacities represent the new capacities (projects), but also the existing capacities with non-null missing money (such as CCGP or Gas Turbines). They are defined by their availability, their deviation, and a probability of daily non availability during PP, but also by their missing money (the value it would cost their owner to make them available), and their limit of investment (as seen afterward, thermal plants can be invested during Y-4 and Y-3, while power shedding can be invested during Y-2 and Y-1).
The table2.1sums up all the objects of the model and all their given proprieties:
Object Properties
Client Obligation level
Deviation Peak consumption control projects Reduction level
Deviation Option cost Activation cost
Existing capacities Availability& Deviation Yearly availability
New capacities Availability
Deviation
Yearly availability Missing money Limit of investment
Table 2.1: List of objects and properties
2.2.2 Module Structure
The model is based upon the following modules that are divided into 2 groups: the phases during which the actors can act by selling/buying/investing/certifying etc., and the phases where the actors can’t act.
Certification phase During this phase, actors can decide to certify their existing capacities and the new capacities they invested during an investment phase. They can also change the amount of certification on any of their capacity that wasn’t invested in during the last investment phase (meaning any capacity that is not certified for the first time). This simulates the redeclaration mechanism, and a premium fee is paid by the actor for changing the certification level of any of its capacities. This redeclaration possibility can be switched off just by changing one of the entrance parameter of the phase, in case redeclaration could not be used anytime.
Investment phase Here the actors can invest in projects of capacities or Consumption control they might have in their portfolio. If they do so, they must pay the missing money or the option cost depending on the type of object they want to invest in. The investment limit must also be good as most new capacities can be invested in only during certain years. Once invested in, new capacities can be certified during the next certification phase, and Peak consumption control projects can be activated during the PP phase.
fixing phase This phase represents the biggest hypothesis of the model. Whereas the capacity market is supposed to work as an OTC market (at least during the first years), the present model
Chapter 2. Modelling of the capacity market 16
Figure 2.4: Presentation of the 6 simulator’s modules
simulates the exchanges through a fixing module. This is an important approximation, but it was believed to be unavoidable, as simulating an OTC market would have proved an enormous (if even possible) task. Here all the OTC exchanges are modelled by a fixing which can be defended by the fact that OTC markets will tend to work as Fixing exchange markets for big markets sizes. This fixing basically works using a classical fixing algorithm, and allows actors to buy and sell defined amounts of certificates at a certain price. Actors are allowed to make as many offers as they want.
PP phase During this phase, actors that paid the option price for Peak consumption control projects are able to activate them (and then pay the activation cost) in order to reduce their obligation.
Random variation phase This phase is supposed to introduce random changes in a simulation in order to test the actors’ strategies resistance to random variations. This phase can take 2 forms: The first one is the classic one: a random variation of availability or obligation is applied to each capacity or client. This variation follows a normal rule whose expectancy is the former value, and whose standard deviation is one of the object parameter. The second form of variation is called PP variation and is used to simulate in particular the uncertainties of the peak period.
Each client is applied a random variation similar to the classical one (normal law). However Peak consumption control projects also are applied a random variation following a normal law (as for the clients, the expectancy is the former value, and the standard deviation is a propriety of the object). Moreover, most capacities are applied a binomial law variation. The probability is given as propriety, and the number of random drawings is given as a parameter (number of days PP2 last). This simulates the fact that every day during the peak period, the capacity has a certain chance of not being available.
Final penalty phase This phase conclude a simulation. During this phase, all the actors whose
certificate account is different from zero must pay a penalty if negative (or might receive money if positive). Then another penalty is applied for actor whose availability levels are different from their certification levels. The chosen model of penalties is the so-called ”bottleneck” method from figure 2.5. A more complete study was written by Arthur HUBERT on this penalty model [15] shown on the figure2.5:
Figure 2.5: The bottleneck penalty model
The red curve (P REn) represents the penalty price for negative differences, while the green one (P REp) is for the positive differences.
This model for setting penalty prices contains 3 modes, depending on the final system difference (the difference in GW between the total certified capacity and the total obligation).
The first mode is located on the right of the figure and represents situations when the system’s safety is not endangered. Usually it means that the difference is positive (system in over- capacity), but this parameter can be set to other values than 0 GW if needed. While in this mode, the price of the penalty is indexed on the price market Pm(the statistical method is also a parameter) times a certain factor.
This factor is 1 + K(E) for a negative difference (short actor) as it will cost the actor more to pay the penalty than to have bought it at the market price before. The factor is thus 1 − K(E) for a positive difference (it would have earned the actor more money to sell its remaining certificates at the market price, than to wait for the penalty to receive money).
The second mode (located on the left of the figure) is when the security of the system’s supply is really endangered (difference lower than ELimit). Here the penalty for negative difference is fixed at Pa; a fixed price calculated to be around the cost of a gas turbine installation (around 60 ke/MW). Finally the third mode is in between these two limits (on the figure it is between ELimit and 0). Here the penalty price is linear as seen in Figure2.5.
Chapter 3
Simulator structure
3.1 General structure
Once the model for the market finalized, the simulator was set in place. For simplicity reasons it was chosen to develop it in Visual Basic for Application (VBA), since displaying results and organising the tables was easier in a simple Excel file. The main principle of the simulator can be seen in figure3.1:
Figure 3.1: Diagram of the simulator during a Monte Carlo simulation
The simulator is organised in modules, each representing one of the model module (see figure 2.4). These modules receive order from the modules directing the actors’ strategies, and then update the database and the result table. The module is applied following a scenario, entered as a parameter of the simulation.
Then, once the scenario is over, the financial results of each actor and several other data are stocked, and another simulation is run with other random selection value. This allows the simulator to run Monte Carlo simulations: the simulation is run several times (the number of iteration is set as a parameter), and the final expected financial result of each actor is considered equal to its average result in all the simulations. This also allows estimating the risk level for a strategy by obtaining the results variance.
The usual scenario used during most studies is detailed in figure 3.2. It allows simulating both organized market sessions (one per year) and a continuous market. It also allows simulating continuous redeclaration possibilities as well as one redeclaration occurrence per year.
18
Figure 3.2: The Usual 4 year scenario
3.2 Database and other entrance parameter
The database used for simulating the market contains all the objects of the 4 previously described types that will be used by the actors. Each object proprieties are listed in the table, and several other columns are used to store other data useful for the simulation. The size of the database and the amount of objects is theoretically not limited; even though the bigger it is, the longer the simulation will take. Thus some simplifications are expected.
This database also allows listing all the market actors. The maximal number of allowed actors for the moment is 18. Here again, the number of actors only make the simulation slower, so some small actors must be merged in order to accelerate the calculus time. An example of a database can be found in Appendix A.
The mechanism’s parameters can be changed between each Monte Carlo. They are listed in table3.1:
Chapter 3. Simulator structure 20
Parameter Note
param(1, 1) Duration of PP2 (days) param(2, 1) Duration of PP1(hours)
param(3, 1) Market price calculus statistic ( capacity owners) param(4, 1) Upper bound of the tolerance band for owners(MW) param(5, 1) Lower bound of the tolerance band for owners(MW) param(6, 1) Fixed penalty for owners (ke/MW)
param(7, 1) Incitation coefficient for owners K(E) param(8, 1) Market price calculus statistic (retailers)
param(9, 1) Upper bound of the tolerance band for retailers(MW) param(10, 1) Lower bound of the tolerance band for retailers(MW) param(11, 1) Fixed penalty for retailers (ke/MW)
param(12, 1) Incitation coefficient for retailers K(E) param(13, 1) Redeclaration premium 1 (ke/MW) param(14, 1) Redeclaration premium 1 (ke/MW) param(15, 1) Redeclaration premium 1 (ke/MW) param(16, 1) Redeclaration premium 1 (ke/MW)
param(17, 1) Final system difference calculus method (owners) param(18, 1) Final system difference calculus method (retailers)
Table 3.1: Simulator’s parameters
3.3 Actor’s Strategies
Once the model for the game set, the actors’ behaviour had to be simulated. The strategies were chosen to be simple, mainly because the goal was to try the simulator with automated actors behaviours, adaptable to all actors and situations. These small modules were called ”bots”. 3 basic types of behaviour were coded for the actors’ market strategy, and another 3 for their redeclaration strategy. Certification, Investment and PP consumption control activation were automated following some simple rules:
An actor invests in a new capacity only if it’s missing money (in ke/MW) is strictly inferior to the market price of the last fixing
An actor always certifies a capacity at its present value (being also the most probable value for the final availability of the capacity)
An actor invests in a consumption control project only if the market is its activation price plus its option price are inferior to the market price of the last fixing. Moreover they are activated only if the market is short.
3.3.1 Market Strategies
All of the 3 market strategies are based upon the same type of behaviour: an actor will first estimate his obligation. Then he can react in two ways: if he has enough existing capacities and new capacities in project to exceed his obligation, he will issue buying offer for his cheaper projects at their missing money thus ensuring that he will not invest in projects if he could
have bought cheaper certificates on the market. He will then sell all the rest of his projects at their missing money. If the actor does not have enough projects in his portfolio to exceed his obligation, he will offer to buy the remaining certificates at the penalty price.
Please note that the prices of the offers are maximal buying price and minimal selling price.
The market being modelled as a fixing market, all the transactions are made at the market price following a pay as cleared model
Figure 3.3: Actors’ basic bidding strategy
The difference between the 3 market bots lay in the temporality of the mechanism: Even if all the projects will be sold during their allowed periods (an actor can’t wait till the end to see if he needs to invest in a CCG plant), the different bots will sell the exceeding certificates (at a minimal null price) or buy their need for certificates (at the penalty price) at different times.
The three bots were given nicknames during the development of the simulator and these nick- names are still used to refer them. The first bot is called Bj¨orn, and will try to act safely by buying or selling as soon as possible. The second bot, called Helmut, will partially buy his certifi- cates needed during the first markets, but will wait for power shedding projects to be available to finish buying what he needs, hoping that power shedding certificates will be cheaper than thermal plant ones. Finally, the last bot is called Enzo. He will wait for the last market session to sell all his extra certificates (hoping for a short market), or to buy his certificates (hoping for a long market). Figures3.4,3.5, and3.6show these different behaviours:
3.3.2 Redeclaration strategies
Three strategies are implemented to simulate the actor’s strategies concerning redeclaration.
These strategies were called R0, R1, and R2. The first two strategies represent two stereotypical and extreme behaviours (redeclaring whenever it is possible, or just at the last minute). The last strategy represents a more developed and subtle strategy.
The R0 strategy consists of redeclaring to the real availability every time it is possible (thus once every year in the usual scenario). This strategy is the most ”virtuous one” as it gives everyone the right signal concerning the real availability (though the Certified Capacities Register). This strategy can prove really expensive, in the case of for example a down-variation of the availability followed by an up-variation of the same amount the following year. In this case, the actor will
Chapter 3. Simulator structure 22
Figure 3.4: Temporality for the Bj¨orn strategy
Figure 3.5: Temporality for the Helmut strategy
pay the redeclaration fee twice, even though his final availability value is the same as his initial value
The R1 strategy on the contrary, consists of redeclarating only at the last moment. It could prove useful as in the example used before, no premium would be paid. However this strategy is dangerous for the system, as the market can no longer reflect the real situation of the system.
The R2 strategy consists of redeclaring at the last minute, like R1, but in the meantime, the actor will buy/sell the difference between the real and the certified value on the market. By making this, he prevents any risk inherent to R1, like having to buy expensive certificates at the last minute if the market is short, or being left with extra certificates on a long market. This strategy furthermore gives the right signal to the market (as the actor is actually buying/selling
Figure 3.6: Temporality for the Enzo strategy
certificates as if it was certified to the right value), even though the Register signal is false.
Redeclaration strategy Register value Market signal
RO
(redeclaration as soon as possible) TRUE TRUE
R1
(redeclaration as soon at the last minute) FALSE FALSE R2
(last minute redeclaration, but market cover) FALSE TRUE Table 3.2: Redeclaration strategies’ signals
3.4 Agent Based Model
Once the model and various strategies were implemented in the simulator, it became necessary to check the effects and the efficiency of the various strategies for the actors in order to obtain a relatively realistic market. The problem here was that since every actor can choose between 9 different strategy sets, the possibilities were too many to be tested by hand. Thus an Agent Based model was implemented on the simulator.
3.4.1 The Roth Erev Algorithms
The used agent based model is a reinforcement-learning model based on Roth-Erev algorithms [16]. These algorithms were developed in order to easily obtain the optimal strategies’ set for actors in a game.
It is based on the reinforcement learning principle. This consists in making the actors play the same game again and again. Each actor will then ”learn” from its previous games, and finally converge toward its best choice of strategy.
Chapter 3. Simulator structure 24
The principles of these algorithms can be found in the figure3.7and in the following equations.
Explanation of all the used symbols can then be found in table3.3:
Figure 3.7: Overview of the Roth-Erev reinforcemnt-learning algorithms
qj(t + 1) = [1 − ϕ]qj(t) + Ej(ε, N, k, t) (3.1)
Ej(ε, N, k, t) =
rk(t)[1 − ε], if j = k qj(t)[N −1]ε , if j 6= k
(3.2)
P robj(t) = eqj(t)/T PN
n=1eqn(t)/T/T (3.3)
Symbol Meaning
t Number of the current Agent-Based loop j Index number for a strategy
k Index concerning the chosen strategy during the loop t aj Action of choosing the strategy number j
qj(t) Propensity of the actor for the strategy j during loop t
P robj(t) Probability for the actor to choose the strategy j during loop t Ej(ε, N, k, t) Learning part of the propensity update equation
N Total number of different strategies available rk(t) Result of the loop number t, using k strategy
E Curiosity parameter
ϕ Forgetting parameter
T ”Temperature” parameter
Table 3.3: Roth-Erev symbols
This model gives each actor a certain propensity qj(t) (contrary to a probability, propensity is not between 0 and 1), and a certain probability P robj(t) for each strategy aj .
Before each Monte Carlo simulation, each actor randomly chose its strategy for this simulation based on the probabilities P robj(t). The chosen strategy is named ak .
At the end of the simulation, each actor evaluates his result: rk(t). Here the result function is (with a parameter of the simulation called ”risk reluctance” and noted µ):
rk(t) = Expected result(t)k− µ ∗ V ariancek(t) (3.4)
Each propensity is then updated following the algorithm’s formula. The probabilities are then also updated by using the Boltzmann formula on propensities (T being a ”temperature” pa- rameter allowing to accelerate or reduce the speed of convergence of the actor toward a unique strategy).
Then, after a certain amount of iterations, the model will converge toward an equilibrium sit- uation. If this situation is correctly reached, each actor has adapted his strategy to the best response to the others’ strategies.
3.4.2 Nash Equilibrium
In game theory, a Nash equilibrium is ”a profile of strategies such that each player’s strategy is an optimal response to the other players’ strategies” [17]. This basically means that at the Nash equilibrium each actor willing to change its strategy will lower its final gain.
It was seen that at the end of the Roth-Erev model, each actor converge toward the best response to the others’ strategies. This point is thus a Nash equilibrium, because a change in an actor strategy is supposed to cause inferior results for this actor. [16]
3.4.3 Implementation of the Model
Once implemented in the simulator, it is possible to perform without setting the actors’ strategies as entrance parameter. For a given set of mechanism parameters and a certain database repre- senting the market situation, the simulator runs a series agent based loops; each of one consists in a Monte Carlo simulation with a certain set of strategies. Finally the system converges and returns the best set of strategies for the actors, as described in the figure3.8.
Figure 3.8: Diagram of the Agent Based module of the simulator
Chapter 4
Simulations
4.1 Inputs
The following simulations are obtained with the simulator in its latest version (v0.40). They all feature the same mechanism parameters, database and scenario. The only change is located in the actors’ strategies. The first two studies are using common bot modules (see 3.3). The last three studies on the contrary use specially designed modules non-reusable for other actors or situations.
At this point it should be reminded that the goal of these studies were to try to simulate the capacity market with roughly estimated values, in order to test the relevancy and the usefulness of the model and simulator rather than perfectly simulating the French market. This is why most of the values used in the database rely on rough and raw estimations rather than on exact (and surely non-public) values. All actors were also renamed. Finally, the simulation main goals were more about analyzing the system behaviour and situation than being able to obtain precisely the results of a single actor.
The simulation database can be found in Appendix A if needed. The simulation database is based on several sources:
4.1.1 Capacities
The capacities (new or existing) are based on the RTE website data concerning the reference park [11]. Their availability was chosen at their installed capacity. This approximation was made because of the non-availability of a formal certification method yet. It might lead to some distortion in the real values, as not all the capacity types will be certified the same way. Their deviation was set at 10% of their availability, and their PP non-availability probability was set to 2% per day. Their missing money was given by RTE and based on internal studies.
Some production types (Nuclear and hydro power) were merged into bigger objects of capacities from the same type belonging to the same actor. This approximation was used in order to accel- erate the calculations of the simulator. The deviation of these big capacities was reduced in order to take random proliferation effect into account. Furthermore, their PP Random variation is not binomial as for the other capacities but Gaussian, in order to simulate the random proliferation.
26
The data used do not include solar or wind power. The approximation of not taking them into account in the present simulations was judged valid by the author and the RTE team in charge of the project. The first reason of this hypothesis is that they only represent a small fraction of the French electric production (around 5% of the total production [9]). Secondly, their availability is less certain at a given moment (the peak period), due to their power source nature (the wind and the sun cannot be contracted to be available). Therefore, they will receive less certificates in comparison to a same sized more reliable capacity, making the amount of certificates coming from solar and wind plants even smaller. This hypothesis is backed by some simulation in 4.2.3.
The power shedding was also added, in quantities and at an approximated price based on as- sumptions made by the author and members of the RTE team. The decision to model power shedding even if its installed capacity seems to be smaller than the renewable energies’ one is based on the fact that one of the capacity mechanism’s aim is to support this source of ”capac- ity”. Observing the effect of the mechanism on power shedding’s actors is thus important for the model.
4.1.2 Clients
The clients values came from an internal RTE study and represent real possible obligation values (although it was obtained with a formula that might not be the adopted one at the end). The deviation for small clients was also set to 10% of the obligation value, while for the 3 biggest clients the proliferation was taken into account.
As described in section 1.2, the ARENH mechanism is a mechanism allowing small retailers to buy half of the nuclear production of the ”historical actor” (EDF) on the energy market, at its production costs. The NOME law also states that the nuclear capacity certificates has to be given with the nuclear ARENH power bought from the historical actor. This was intended in order to reduce the power of the historical actor on the capacity market by forcing it to transfer half of its nuclear certificates to small retailers.
In order to simulate the ARENH mechanism, all small retailers have a 50% lower obligation, while the historical actor (Seine Elec) increase its obligation corresponding to all the obligation reduced from smaller actors. This 50% value is based on rough estimations of RTE data.
4.1.3 Peak consumption control projects
Modelling the ability for actors to use consumption control is really important to the model. It is one of the few actions that pure retailers can entertain within the mechanism. Moreover the mechanism aims at supporting these action, as well as power shedding actions (see 2.1.5 for more information on power shedding and consumption control actions). These objects properties were thus based on the same estimations made for the power shedding objects.
4.1.4 Parameters
The model’s parameters are not available since they are still being discussed by the market actors. The used parameters are thus chosen by the author based on RTE’s suggestions and can be found in appendix B. The author wants to remind the reader that these particular parameters’
Chapter 4. Simulations 28
values does not represent RTE’s point of view. The number of games played during the Monte Carlo simulation is 1000.
4.1.5 Scenario
The used scenario is the same as pictured in the figure 3.2. There is nevertheless one difference with the formal scenario: since the PP random variation is not affected by the choice of the strategies, it was decided that the PP Phase of the scenario will not feature random variations as it would only bring some noise to the results of the simulations.
The figure4.1shows this formal scenario year by year with all the phases happening chronologi- cally, from the top (beginning of the market, 4 years before the peak period) to the bottom (end of the differences payment, after the peak period).
Figure 4.1: Scenario used for the simulations
4.2 Simulations’ Results
4.2.1 Choice of the best redeclaration strategy
The next three simulations are conducted in order to compare the results of the three possible redeclarations strategies on the market. For each iteration of a game, the market strategy of each
of the 17 actors is randomly chosen from the three available market strategies (Bj¨orn, Helmut, or Enzo). Therefore, each of the 1000 games played during a simulation has its own strategy set. The difference between the 3 simulations comes from the redeclaration strategy used by the actors. The first simulation set all the actors on R0, while the second set them on R1, and the third one on R2.
Figure 4.2: Simulations’ results for the different redeclaration strategies
The figure4.2features the system final difference between total availability and total obligation, the total financial result for the actors (featuring all the money they paid in missing money, that explain why it is negative), the spare money remaining on the penalty accounts, and finally the total financial result for the producer/capacity owners (as they are the only ones able to use the redeclaration mechanism).
It becomes then clear that R2 is the best strategy for the actors, and the system. The total system cost is the lowest (9% lower than with R0) and the owners result is increased by 18%
compared with the R0 strategy. It also appears that R1, as predicted is the most expensive for the system (even though not for the owners only).
This simulation makes it clear that the R2 strategy is the best for the capacity owner, and also for the system. The fact that the register information is false is compensated by the fact that the market signals are true and represent the real availability level.
4.2.2 Comparison of the market strategies
These 3 simulations are made by setting all actors to the R2 redeclaration strategy, and then by changing their market strategies from Bj¨orn, to Helmut, and then to Enzo.
Figure4.3shows the similar system values than the last one, except for the risk-increased cost, calculated by multiplying the system difference with the fixed penalty (60ke/MW) in case of negative final difference. It aims at representing the TSO’s investment in maintaining the system running and/or the social costs of power outages. The system costs are calculated by adding:
System total costs = T otal actor costs + Retailer spare money + P roducers spare money + Risk increase cost (4.1)
Chapter 4. Simulations 30
Figure 4.3: Simulations’ results for the different market strategies
These simulations show the impacting effect of a global Enzo strategy for the system. The system becomes under-capacitated by more than 2GW, and the system costs grow strongly by 19%. Moreover this strategy cost more for the actors.
Bj¨orn and Helmut strategies provide quite similar results at the end. Actually, although the Helmut strategy provides slightly better financial results for the actors (1.2% lower actor costs), it also is more risky for the system (80MW lower expected final difference).
4.2.3 Influence of renewable energies
The next two simulations aim at supporting the author’s hypothesis concerning the renewable energies. These two simulations take place with actors using the R2 redeclaration strategy, which was proved the most efficient in 2.2.1. The choice of the market strategy for each actor and for each game is made randomly between the three possible market strategies as it was the case in 2.2.1. Please note that in this master thesis, “renewable energies” include solar-power, wind- power, and other renewable sources, with the exception of hydro-power. The first simulation uses the standard database. In the second simulation, 3.21 GW of existing capacity is added to the market in order to represent the renewable energies certificates. This figure comes from the installed capacity of 10.7 GW of renewable energies [18], modified by a factor of 30%, representing the fact that since they have a lower availability than other production sources. This factor is chosen by the author based on discussions with RTE members.
The results can be found in the table 4.3:
Figure4.4thus shows that adding 3.2 GW of renewable energy to the system does not represent an important change. The system’s costs are decreased by 6% and the actor’s gain is increased by 4%. This is relevant with the fact that 3.21 GW are added to the market supply (thus a 3.6%
increase of the system initial supply).
These results can be used to support the hypothesis that renewable energies can be neglected from the database, as their current effects are small enough to be blended into other existing capacities (like hydro-power for example).
