Sizing of the French secondary sources

Master thesis: Sizing of the French secondary sources

Clément CHAUVEAU 2020

Examiner:

Lennart SÖDER Supervised by:

Abolfazl KHODADADI Julien GAUDIN

Sébastien FINET

2

Abstract

The study carried out in this master thesis comes from a larger project from the ENTSO-E (gathering of the European Transmission System Operators (TSO)) focusing on introduction of a European platform for the exchange of balancing energy from frequency restoration reserves. The frequency restoration reserves are the power reserves that allow the TSO to balance the consumption and the production. The management of these reserves is therefore strategic for the TSO. RTE (the French TSO) takes part to this European project and this has led RTE to challenge its secondary sources sizing method. The sizing method goal is to do a level of reserve prescription for the next day. The objective of the thesis is to evaluate a new secondary sources sizing method. To achieve the objective, two main tasks are identified. The first is the adaptation of a sizing method suggested by the ENTSO-E. This new sizing method should be applicable to the French power system and its specificities.

This requires some modifications from the original suggested method. We will decline the method into different scenarios with different parameters. The second task focuses on the evaluation of the different scenarios and the comparison with the current sizing method.

The evaluation will investigate three aspects for each scenario:

• The static aspect, which investigates the statistical characteristics of the scenario’s prescription.

• The dynamic aspect, which evaluates the impact of the scenario’s prescription on the power grid.

• The economic, which investigates the scenario’s cost.

At the end of the study, RTE will dispose of a set of scenarios that are applicable to the

French system with satisfying performances in order to select the best scenario considering

its cost/efficiency balance.

3

Sammanfattning

De studier som genomförts i detta examensarbete kommer från ett större projekt, från ENTSO-E (gathering of the European Transmission System Operators (TSO)) med fokus på en presentation av en europeisk plattform för utbyte av balanserad energi från frekvens återställnings reserven. Frekvens återställnings reserven är kraftreserver som gör det möjligt för en TSO (systemansvarig) att balansera förbrukningen och produktionen. Behandlingen av dessa reserver är därför strategisk för en TSO. RTE (den franska TSOn) deltar i detta europeiska projekt och detta har medfört att RTE ifrågasätter sin nuvarande sekundära mätmetod. Målsättningen för mätmetoden är att veta vilket krav på reserven som behövs för nästa dag. Syftet med detta examensarbete är att utvärdera en ny mätmetod. För att nå målet identifieras två deluppgifter. Den första är anpassningen av den mätmetod som föreslås av ENTSO-E. Denna nya mått metod bör tillämpas på det franska kraft systemet och dess förutsättningar. Detta kräver vissa ändringar från den ursprungligt föreslagna metoden.

Vi kommer att använda metoden i olika scenarier och med olika parametrar.

Den andra deluppgiften fokuserar på utvärderingen av de olika scenarierna och jämförelsen med den aktuella mätmetoden. Utvärderingen kommer att undersöka tre aspekter för varje scenario:

- Den statiska aspekten som undersöker de statiska egenskaperna för scenariots krav.

- Den dynamiska aspekten som utvärderar effekterna av scenariots krav på elnätet.

- Den ekonomiska som undersöker scenariots kostnad.

Slutsatsen är att RTE kommer att kunna använda ett par uppsättningar av de scenarier som

är tillämpliga på det franska systemet med tillfredsställande resultat till att kunna välja det

bästa scenariot med tanke på dess kostnad och effektivitet.

4

Acknowledgements

First, I would like to thank my supervisor Abolfazl Khodadadi for his support and confidence during all my master thesis. I also thank my examiner Lennart Söder for his interest in my master thesis.

I would also like to thank Julien Gaudin and Sébastien Finet for their support and the share of their knowledges during my entire thesis in RTE.

I thank all my colleagues at RTE from their welcome and the cooperation throughout my master thesis.

Finally, I would like to thank all the people who participated in the elaboration of my

master thesis as well as to those who helped me with the writing of this report.

5

Contents

Abstract ... 2

Sammanfattning ... 3

Acknowledgements ... 4

List of Figures ... 7

List of Tables ... 8

Abbreviations ... 9

1. Introduction ... 10

1.1 Background ... 10

1.1.1 The European organization and RTE ... 10

1.1.2 The Load Frequency Control ... 12

1.1.3 The FRCE target parameters ... 14

1.1.4 The ongoing change for the aFRR ... 15

1.1.5 The current aFRR dimensioning in France ... 15

1.1.6 Method suggested by the ENTSO-E ... 17

1.2 Objective of the study ... 19

1.3 Disposition ... 19

2. Description of the general procedure ... 21

2.1 Basic application ... 21

2.1.1 The measured data manipulation ... 21

2.1.2 The transformation of the data into the needed value ... 22

2.1.3 The extreme percentiles and the aFRR prescription ... 22

2.2 Adaptation of the method ... 23

2.2.1 The question of Δ15 ... 23

2.2.2 The prescription time step ... 24

2.2.3 The grouping of days ... 25

2.3 Evaluation of the method ... 26

2.3.1 The static evaluation ... 26

2.3.2 The dynamic evaluation ... 27

2.3.3 The economic evaluation ... 31

2.3.4 Selection of the scenario ... 33

3. Adaptation of the method ... 34

3.1 Choice of the Δ as base material ... 34

6

3.1.1 The characteristics of the French system ... 34

3.1.2 The introduction of Δ30 ... 34

3.1.3 Comparison between Δ30 and the historical aFRR ... 35

3.1.4 Choice of Δ30 ... 36

3.2 The prescription time step ... 37

3.2.1 The possible prescription time steps ... 37

3.2.2 Impact on the prescription ... 37

3.2.3 Conclusion on the prescription time step ... 39

3.3 The grouping of the days together ... 40

3.3.1 The weekdays ... 40

3.3.2 The months ... 41

3.3.3 Length of the data set ... 43

4. Evaluation of the different scenarios ... 45

4.1 Static evaluation ... 45

4.1.1 Statistical characteristics... 45

4.1.2 The aFRR saturation ... 47

4.2 Dynamic evaluation. ... 48

4.2.1 The model calibration. ... 48

4.2.2 Evaluation of the different scenarios ... 49

4.3 Economic evaluation. ... 51

4.4 Discussion on the results ... 52

5. Conclusion and future work. ... 53

A. ACE and RFCE target parameter ... 55

A.1 ACE = 0 if the area is not responsible for the frequency deviation: ... 55

A.2 FRCE target parameter level 1 and level 2 calculation [3] ... 56

B. The activation process... 59

C. The impact of a better repartition of the aFRR over the day and the year ... 61

Bibliography ... 64

7

List of Figures

F

IGURE

1.1: T

HE

5 R

EGIONAL GROUPS IN THE

ENTSO-E. ... 11

F

IGURE

1.2: P

ROCESS RESPONSIBILITY STRUCTURE

. ... 12

F

IGURE

1.3: E

XAMPLE OF A

FRR

PRESCRIPTION AND SCHEDULED A

FRR. ... 17

F

IGURE

1.4:

EXAMPLE OF A

Δ_15. ... 18

F

IGURE

2.1:

DESCRIPTION OF THE METHOD PROCESS

. ... 21

F

IGURE

2.2: R

OLE OF THE A

FRR

WITH

𝛥15. ... 23

F

IGURE

2.3: R

OLE OF THE A

FRR

WITH

𝛥30. ... 24

F

IGURE

2.4: C

ONTROL PROCESS

. ... 28

F

IGURE

2.5: T

HE A

FRR

CONTROLLER ON OPENMODELICA

. ... 29

F

IGURE

2.6: M

ODELED A

FRR

CONTROLLER

. ... 30

F

IGURE

3.1: D

ISTRIBUTION OF THE

Δ

30 AND THE CORRESPONDING PERCENTILE

. ... 35

F

IGURE

3.2: C

OMPARISON BETWEEN THE DIFFERENT PRESCRIPTION TIME STEPS

. ... 38

F

IGURE

3.3: C

OMPARISON OF THE A

FRR

PRESCRIPTION BETWEEN THE DIFFERENT WEEKDAYS

. ... 40

F

IGURE

3.4: C

OMPARISON OF THE MEAN VALUE OF THE SCHEDULED A

FRR

BETWEEN THE DIFFERENT WEEKDAYS

. ... 41

F

IGURE

3.5: C

OMPARISON OF THE A

FRR

PRESCRIPTION BETWEEN THE DIFFERENT MONTHS

. ... 42

F

IGURE

3.6:

COMPARISON BETWEEN TWO DATA SET LENGTH

. ... 44

F

IGURE

4.1: C

OMPARISON BETWEEN THE SIMULATED AND THE HISTORICAL A

FRR

ACTIVATION

... 49

F

IGURE

B.1: T

HE FULL ACTIVATION PROCESS

... 59

F

IGURE

C.1: L

EVEL

2

REPARTITION WITHIN THE DAY FOR THE SCHEDULED A

FRR. ... 61

F

IGURE

C.2: C

OMPARISON OF THE

L

EVEL

2

REPARTITION WITHIN THE DAY BETWEEN TWO SCENARIOS

. ... 62

F

IGURE

C.3: C

OMPARISON OF THE

L

EVEL

2

REPARTITION WITHIN THE DAY BETWEEN TWO SCENARIOS GROUPED BY HOUR

. ... 62

8

List of Tables

T

ABLE

3.1:

COMPARISON BETWEEN THE CURRENT STATE AND THE APPLICATION WITH

Δ_30. ... 36

T

ABLE

3.2:

TIME STEP TRANSITION

. ... 38

T

ABLE

3.3:

COMPARISON OF THE MEAN A

FRR

PRESCRIPTION BETWEEN THE DIFFERENT PRESCRIPTION TIME STEPS

. .... 39

T

ABLE

3.4: D

ESCRIPTION OF THE MONTH GROUPS

. ... 42

T

ABLE

4.1:

STATISTIC EVALUATION OF THE SCENARIOS

. ... 46

T

ABLE

4.2:

SATURATION EVALUATION OF THE SCENARIOS

. ... 47

T

ABLE

4.3: FRCE

TARGET PARAMETERS EVALUATION FOR THE DIFFERENT SCENARIOS

. ... 50

T

ABLE

4.4: C

OST COMPARISON FOR THE DIFFERENT SCENARIOS

. ... 51

T

ABLE

B.1: T

HE ACTIVATION PROCESS VARIABLES

... 59

9

Abbreviations

ACE ... Area Control Error aFRR ... automatic Frequency Restoration Reserve CE ... Continental Europe CRE ... Commission de Régulation de l’Energie FAT ... Full Activation Time FCR ... Frequency Containment Reserve FRCE ... Frequency Restoration Control Error IGCC ... International Grid Control Cooperation LFC Area ... Load-Frequency Control Area mFRR ... the manual Frequency Restoration Reserve MR ... Must Run PICASSO ... Platform for the International Coordination of Automated Frequency Restoration

and Stable System Operation

RGCE ... Regional Group Continental Europe

RR ... Replacement Reserve

RTE ... Réseau de Transport d’Electricité

SO GL ... System Operator Guide Lines

TSO ... Transmission System Operator

10

1. Introduction

1.1 Background

This part will provide some general information that are needed in order to understand the following project.

1.1.1 The European organization and RTE

In Europe, there is an association called the European Network of Transmission System Operators for Electricity (ENTSO-E) which represents 42 electricity transmission system operators (TSO). The ENTSO-E has been founded in 2009 and is legitimated by the EU’s Third Legislative Package for the Internal Energy Market. In this association there is 35 European countries represented. The purpose of the ENTSO-E is to facilitate the cooperation between European TSOs in order to organize the energy market and support the European energy and climate agenda. The ENTSO-E is producing network codes and policies and publishes reports on the different TSO performances. The ENTSO-E also has a role in the R&D field. Indeed, the ENTSO-E has a R&D roadmap which identifies the R&D actions needed in order to meet the EU requirements for the electricity sector. The ENTSO-E has a strong transparency policy and provides free data on the European energy sector on his Central Information Transparency Platform. ENTSO-E’s actions lead to standardize some aspect of the TSO’s organization in order to optimize the integration of the different policies or markets.

The 42 TSOs from the ENTSO-E are not in the same synchronous area. In fact, there is 5 different synchronous areas:

• Continental Europe (CE) : (Albania, Austria, Belgium, Bosnia-Herzegovina, Bulgaria, Czech Republic, Croatia, Denmark (West), France, FYROM, Germany, Greece, Hungary, Italy, Luxemburg, Montenegro, Nederland, Poland, Portugal, Romania, Serbia, Slovakia, Slovenia, Spain, Switzerland and Turkey).

• Nordic (Denmark (East), Finland, Norway and Sweden).

• Baltic (Estonia, Latvia, Lithuania).

• Great Britain.

• Ireland-Northern Ireland.

11

Figure 1.1: The 5 Regional groups in the ENTSO-E.

In France there is only one TSO called “Réseau de Transport d’Electricité” (RTE). RTE is

monitoring one of the largest grids in Europe with more than 100 000 km of power lines and

2 700 substations. RTE is interconnected with six of the neighboring countries. More than

100 TWh are exchanged through these interconnections every year. RTE handle the lines

with a 5 different voltage levels: 400 kV, 225kV, 150kV, 90kV and 63kV. The main role of RTE

(as every TSO) is to maintain the balance between the supply and the consumption and

deliver reliable and economical electricity to the customers. The balance between the supply

and the consumption cannot be perfect and there are then frequency deviations in the

system. These frequency deviations are handled by the Load Frequency Control process

(LFC).

12 1.1.2 The Load Frequency Control

The Load Frequency Control has usually 3 steps:

• The Frequency Containment Process (primary control).

• The Frequency Restoration Process (secondary control).

• Reserve Replacement Process (tertiary control).

The ENTSO-E has defined in the SO GL the process responsibility structure in the article 141 as the following [1].

Figure 1.2: Process responsibility structure.

A Scheduling Area is responsible for the scheduling process in that area. A Monitoring Area

has in addition to the scheduling the obligation to calculate and measure the active power

interchange in real-time in that area. A LFC Area has the additional obligation to fulfil the

Frequency Restoration Control Error Target Parameters by using the Frequency Restoration

Process. LFC Block is additionally responsible for the dimensioning of FRR and RR. The

Synchronous Area has the obligation to fulfil the Frequency Restoration Control Error Target

Parameters by using the Frequency Containment Process. RTE is responsible for both the LFC

Block and the LFC-Area.

13

The objective of the Frequency Containment Process is to stop the frequency deviation and maintain the balance between the generation and the demand within the synchronous area.

Therefor all the TSOs of a synchronous area respond to the system frequency deviation. A frequency deviation results from an imbalance between generation and consumption. These imbalances occur after an incident (power plant outage for example) or during normal system operation (mismatch between the consumption and the consumption forecast). The activation of the Frequency Containment Process is triggered for a frequency deviation larger than ± 20 mHz and the maximal activation for a frequency deviation greater than

± 200 mHz (for CE). The activation of the primary reserve is decentralized, automatic and proportional to the frequency deviation. That means that every power plant taking part to the Frequency Containment Process will modulate their output proportionally to the frequency deviation. The frequency deviation is the same on every point in the synchronous grid and then the primary reserve activation is done by all the TSOs in the synchronous area.

At the end of the Frequency Containment Process (less than 1 minute), the balance between the generation and the demand within the synchronous area is restored but the system frequency is not at its nominal level anymore. And the wider the synchronous area, the smaller the resulting frequency deviation at the end of the Frequency Containment Process.

The objective of the Frequency Restoration Process (the process is described in annex B) is to restore the frequency to its nominal level. If the frequency is back to its nominal level, then the primary reserve is refunded. Indeed, the system frequency deviation equals to zero and then the power plant production also back to the nominal value (the output proportional to the frequency deviation equals zero). The Frequency Restoration Process activation can be divided in two parts using two types of reserve, an automatic reserve (aFRR) and a manual reserve (mFRR). The activation of the Frequency Restoration Process only takes place in the LFC-Area responsible for the frequency deviation (see annex A for a small example). The automatic activation of the Frequency Restoration Process is based on the Area Control Error (ACE, equivalent to the FRCE Frequency Restoration Control Error) which is the sum of two different variables.

𝐴𝐶𝐸 = 𝐾 ∗ (𝑓 − 𝑓

₀

) + 𝛥𝑃

_𝑖

(in MW) (1.1)

With:

𝑓: The grid frequency in Hz.

𝑓

₀

: The nominal frequency in Hz.

𝐾: The K-Factor is an estimation for the change of Active Power output of a LFC Area resulting from a Frequency Deviation in MW/Hz.

Δ𝑃

_𝑖

: The difference between the actual power flow through the interconnection and the

planned power flow in MW.

14

At the end of the Frequency Restoration Process (15 minutes), the system frequency is back to its nominal value and the primary reserves are fully available. But aFRR is not fully available anymore.

The objective of the Reserve Replacement Process is to refund the secondary reserve or act as a supplement to the secondary reserve after large incidents. This consists in changes in generation or load on a contractual, market or regulatory basis. At the end of the Reserve Replacement Process, the secondary reserve is back to a sufficient level and the system frequency is still at its nominal value.

In this study we will focus on the reserve for the automatic Frequency Restoration Process (the aFRR). To be more precise we will investigate the aFRR prescription i.e. the level of power in the aFRR required by RTE.

aFRR prescription: The level of power in the aFRR required by RTE 1.1.3 The FRCE target parameters

The regulation of the European grid is in constant evolution and the regulator needs to evaluate the grid state/quality. To do so there are different indicators or target parameters.

In our case, we will focus on the FRCE target parameters. The European commission has introduced a new regulation concerning the FRCE target parameters the 2

^nd

of August 2017[1]:

All TSOs of the CE and Nordic synchronous areas shall endeavor to comply with the following FRCE target parameters for each LFC block of the synchronous area:

(a) the number of time intervals per year outside the Level 1 FRCE range within a time interval equal to the time to restore frequency shall be less than 30 % of the time intervals of the year;

and

(b) the number of time intervals per year outside the Level 2 FRCE range within a time interval equal to the time to restore frequency shall be less than 5 % of the time intervals of the year.

The time to restore frequency is 15 minutes and the FRCE for a LFC block is equivalent to his ACE. This mean that the absolute value of the mean value of the ACE in range 15 minute must be 30 % of the time smaller than the level 1 and 5 % of the time smaller than the level 2.

|𝑚𝑒𝑎𝑛

_{15𝑚𝑖𝑛𝑢𝑡𝑒𝑠}

(𝐴𝐶𝐸)| ≤ 𝑙𝑒𝑣𝑒𝑙 1 30 % 𝑜𝑓 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒. (1.2)

|𝑚𝑒𝑎𝑛

_{15𝑚𝑖𝑛𝑢𝑡𝑒𝑠}

(𝐴𝐶𝐸)| ≤ 𝑙𝑒𝑣𝑒𝑙 2 5 % 𝑜𝑓 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒. (1.3)

15

The level 1 and 2 have different value depending on the LFC-Block and change every year. For RTE, their values in 2019 are [2]:

Level 1 = 225,851 MW Level 2 = 427,120 MW

The calculations of these levels are explained in Annex A.

1.1.4 The ongoing change for the aFRR

The Electricity Balancing Guideline [6] of 23 November 2017 objective is to create multiple European electricity markets where the TSO can share balancing resources. This should lead to minimizing the risk for the electricity supply and maximizing the social welfare.

According to these guidelines RTE will introduce different markets to trade primary, secondary and tertiary resources. The introduction of these markets will have multiple impacts on the French power system’s doctrine. For example, if we can focus on the aFRR market (PICASSO, Platform for the International Coordination of Automated Frequency Restoration and Stable System Operation). The implementation of this market implies a merit order activation for the aFRR as it is market based and the actors biding their

capacities. Today, the activation is proportional to the ACE. This is done by computing a level called N which can vary between -1 and 1 and represent the share of the reserve that must be activated. N = 0.5 means that we want an activation of 50% of the positive aFRR. This level N is the same for every power plant taking part to the aFRR activation without consideration of the cost.

Moreover, with the European platform some aFRR can be activated in France to solve a frequency deviation in another country. The French balancing doctrine will evolve with these platforms and induce change in the need in aFRR and the available aFRR for RTE. These evolutions will probably have an impact on the French electricity quality. Namely, the level 2 of the FRCE target parameters might be affected and as the margin is not comfortable this may turn into a problem in the future.

As a result, RTE decided to challenge his aFRR sizing method and is looking into a method that will comply with the coming evolutions in the balancing market.

1.1.5 The current aFRR dimensioning in France

The current aFRR prescription method comes from an empiric noise management approach

defined and approved by the French energy regulator (CRE: Commission de Régulation de

l’Energie) in the annex 1 of the “Accord de bloc RFP”[7]. It has been used by RTE without a

modification for the last decade. The empiric noise management approach was

16

recommended by the ENTSO-E and the EU in the beginning of the 21

^st

century. The method is based on the demand forecast. There is an aFRR prescription for every 30 minutes of the day (48 levels of prescription for 1 day) with a minimal prescription level of 500 MW.

The demand forecast is computed one day ahead and is the sum of the French consumption and the exchanges with the connected neighbors.

𝐷

_{𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡}

= 𝐶

_{𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡}

+ 𝐸

_{𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡}

(1.4)

𝐷

_{𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡}

: The demand forecast in MW.

𝐶

_{𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡}

: The consumption forecast in MW.

𝐸

_{𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡}

: The exchanges forecast in MW (positive when exporting).

Then we have to distinguish two cases: the high variation of the demand between two prescription time steps and the rest with a lower variation. The prescription time step is the time step between two aFRR prescriptions. In France the prescription time step is currently 30 minutes. It means that one day is divided into 48 time frames with one aFRR for each of these time frames.

Let’s define Δ

_{𝑝𝑟𝑒𝑠𝑐}

the variation between two prescriptions in MW/30min.

If Δ

_{𝑝𝑟𝑒𝑠𝑐}

≥ 6 000 𝑀𝑊/30𝑚𝑖𝑛:

𝑃𝑟𝑒𝑠𝑐 = 𝛥

_{𝑝𝑟𝑒𝑠𝑐}

6

(1.5)

Else:

𝑃𝑟𝑒𝑠𝑐 = max (500, (√10 ∗ 𝐷 + 22500 − 150)) (1.6) Presc is the aFRR prescription for the corresponding 30 minutes and with all the 48 Presc we have the aFRR prescription for the next day.

This method is sufficient to respect the FRCE target parameters. In 2018 RTE achieved 11.69% of the time above the level 1 (maximum is 30 %) and 4.05% for the level 2 (maximum is 5%). But we observe that there is only a small margin for level 2.

This difference comes from diverse factors, but the aFRR dimensioning is one of them.

Moreover, this aFRR prescription method leads to a rather flat prescription with some peaks (when there is a high variation in the demand forecast between two prescription time steps). The consequences of these peaks are that the aFRR prescription is hardly satisfied in real-time due to technical constraints. But this is not the only reason for mismatch between the aFRR prescription and the aFRR in real-time. In France, there might be a modification of the aFRR prescription during the intra-day operation when the forecast conditions change.

This is done by an operator a few hours before real time in order to minimize the risk. This is

usually to smooth the peaks and add more volume on specific hours (between 00h and 02h

for example). To this you can add that any variation of the production planning mainly due

17

to balancing operation which can affect the aFRR level, with adding some generators to compensate a forecast mismatch for example. The result of both the modification due to technical constraints and operator modification will be called the scheduled aFRR in this report. The scheduled aFRR is the amount of aFRR that were in fact available that day. The difference is shown in Figure 1.3.

Figure 1.3: Example of aFRR prescription and scheduled aFRR.

The current method gives a consistent aFRR prescription. But there are some modifications that must be applied after the method output and the presence of peaks is troublesome for the producer, which sometimes struggles to follow the aFRR prescription. There is also the ongoing project of introducing a European market for the aFRR and this will probably have negative impact on the FRCE target parameters of the French system. With all this in mind RTE the French TSO, have decided to look into a new method suggested by the ENTSO-E to compute the aFRR prescription.

1.1.6 Method suggested by the ENTSO-E

The power grid is facing new problems with the introduction of renewable energy and new market and exchange mechanism. These modifications in the grid led to new types of frequency deviations. The TSOs from the Continental Europe (Regional Group Continental Europe RGCE) have done a recommendation for the aFRR minimum amount [2] in order to respond to these new challenges. This method is using a statistical approach based on historical data. The base material for this method is the ACE and the recommended minimum amount of aFRR has to ensure that:

- The positive aFRR is larger than the 1st percentile of the difference of the 1- minute average 𝐴𝐶𝐸

_𝑜𝑙

and the 15 minute average 𝐴𝐶𝐸

_𝑜𝑙

of the LFC Block of the corresponding quarter of hour.

And:

18

- The negative aFRR is larger than the 99th percentile of the difference of the 1- minute average 𝐴𝐶𝐸

_𝑜𝑙

and the 15 minute average 𝐴𝐶𝐸

_𝑜𝑙

of the LFC Block of the corresponding quarter of hour.

The 𝐴𝐶𝐸

_𝑜𝑙

means the ACE open loop. This can be interpreted as the resulting ACE in absence of the automatic secondary control:

𝐴𝐶𝐸

_𝑜𝑙

= 𝐴𝐶𝐸 − 𝑎𝐹𝑅𝑅

_{𝑎𝑐𝑡𝑖𝑣𝑎𝑡𝑒𝑑}

in MW (1.7)

With 𝑎𝐹𝑅𝑅

_{𝑎𝑐𝑡𝑖𝑣𝑎𝑡𝑒𝑑}

the amount of activated aFRR in MW.

The 𝐴𝐶𝐸

_𝑜𝑙

comes from historical data and it has at least a 5 s time step. Let’s define Δ

₁₅

as:

𝛥

₁₅

= 𝑚𝑒𝑎𝑛

_{1𝑚𝑖𝑛𝑢𝑡𝑒}

(𝐴𝐶𝐸

_𝑜𝑙

) − 𝑚𝑒𝑎𝑛

_{15𝑚𝑖𝑛𝑢𝑡𝑒𝑠}

(𝐴𝐶𝐸

_𝑜𝑙

) (1.8)

Δ

₁₅

: The difference of the 1-minute average 𝐴𝐶𝐸

_𝑜𝑙

and the 15 minutes average 𝐴𝐶𝐸

_𝑜𝑙

of the LFC Block in MW.

Figure 1.4: example of a Δ_15.

We have one Δ

₁₅

every minute. To follow the recommendation, we must regroup them and then we take the extreme percentile (1 % and 99 %) in order to find the aFRR minimum amount. This 1% corresponds to the requirement stated in the SO GL [1] that:

All TSOs of a LFC block shall ensure that the positive reserve capacity on FRR or a combination of reserve capacity on FRR and RR is sufficient to cover the positive LFC block imbalances for at least 99 % of the time.

The assumption in this method is that the 15 minutes ACE open loop mean value must be

handled by the mFRR and the variations around that mean value must be handled by the

aFRR.

19

The 1 % percentile corresponds to the minimum for the positive aFRR and the 99 % percentile correspond to the minimum for the negative aFRR. We have to take the opposite of the percentile because the 𝐴𝐶𝐸

_𝑜𝑙

represent the need in aFRR.

The time period considered for those measurements shall be representative and include at least one full year period ending not earlier than 6 months before the calculation date.

1.2 Objective of the study

As explained in the previous part, the Load Frequency Control performance for a TSO depends on the performances of the primary, secondary and tertiary controls. This performance is evaluated for each LFC Area with the FRCE target parameters. And RTE has only a little margin with the level 2 (more than 4 % for a 5 % limit). The FRCE target parameters are using the ACE as base material. If the ACE is poorly corrected by the Frequency Restoration Process, then its mean value will rise and lead to bad FRCE target parameters for the corresponding LFC Area. The reason for an inefficient ACE correction can be an insufficient amount of aFRR. If the aFRR is lacking power, then the sizing of the reserve is not adapted to the current situation. We can then challenge the method currently used to do the aFRR prescription. After a quick study of the current aFRR prescription method and its outcome, we can consider a new method to replace the current one. We have chosen to investigate the method suggested by the ENTSO-E which is a statistical approach based on the historical ACE.

The objective of this study is to evaluate the compatibility and applicability of the new method to the French system. This method should be compatible with the introduction of the European market for the aFRR and the future evolution of the power system regulation.

We will also investigate what kind of modification to the method must be done in order for the method to be applicable to the French system. This method will be evaluated with different aspects:

• The static aspect.

• The dynamic aspect.

• The economic aspect.

This study has taken place in RTE, the French TSO in France.

1.3 Disposition

The report is structured in the following way:

The Chapter 1 is the introduction to the background of the study and its objective.

The Chapter 2 describes the methodology used in this document and present the model

used.

20

The Chapter 3 describes the different ways to modulate the method in order to adapt it to the French system.

The Chapter 4 evaluates the different scenarios with three different aspects: the static, dynamic and economic aspect.

The Chapter 5 is a conclusion and a discussion on the project.

21

2. Description of the general procedure

In this chapter we will explain the procedure of this master thesis. The objective of this chapter is to present the methodology that was followed in this thesis. This procedure will stay general so that it could be applied to other power systems.

2.1 Basic application

The first step of the procedure is to directly apply the recommended method. To do this we need an historical data set with all relevant data.

Figure 2.1: description of the method process.

2.1.1 The measured data manipulation

As previously explained, the method is based on the real time data. The requirement for this data is that it must have been measured on the grid with a maximum refresh rate of 5 seconds. The data set will contain at least one full year of data and multiple columns. This leads to a huge sized data set with more than 6 millions of lines only for a one-year data set.

The resulting .csv file is bigger than 1 GB and cannot be open with Excel for example.

Moreover, to apply the method we need to manipulate the data set and use some statistical

tools such as the percentile or the mean value.

22

To solve this problem, we have used the R programming language [5]. This is a free software environment for statistical computing. This language allows manipulating large amount of data. This programming language is adapted to the problem and easily available for RTE.

The first step of the application is to gather the different data that we need in our data set and that are available in the measurement at our disposal. We also have to gather the data we need to compute the values that are not measured. This can represent multiple files in different file formats or different refresh rates with valuable data mixed with other data irrelevant for our application.

Then we will merge the different files together, convert the values in the same unit, purge the aberrant values and handle the time change system if needed and select the values we want. This step is essential because the ENTSO-E’s method is based on the historical data and we need to have the best input in order to have the most accurate output.

The outcome of this step is one or multiple files with the different data we need to conduct the application. As we are manipulating tables, we have used a .csv file format.

2.1.2 The transformation of the data into the needed value

We have now all the raw data, but not all the values we need to conduct the method. We have to compute some values that are not present yet in the data set. We will then transform the available values to get the missing values. This step might depend on the values that are measured from the TSO and the way they operate but the result remains the same, at the end we have the ACE open loop of the LFC area with a maximum time step of 5 s.

The next step is the calculation of the values used in the ENTSO-E’s method:

• The 1-minute mean value of the ACE open loop

• The 15-minute mean value of the ACE open loop

• Δ

₁₅

which is based on the equation (1.8)

The result of this step is a file with the Δ

₁₅

values with a maximum time step of 5 s.

2.1.3 The extreme percentiles and the aFRR prescription

Now that we have all the Δ

₁₅

, we can list them in increasing order in order to take the extreme percentiles. These extreme percentiles will be called p

1%

and p

99%

in the following document.

The method says that:

• The positive aFRR prescription is - p

1%

• The negative aFRR prescription is - p

99%

The aFRR prescription is constant over the day and will keep the same value until the data is

updated with most recent values.

23 2.2 Adaptation of the method

In this part, we will adapt the method from the ENTSO-E to the characteristics of the power system. Indeed, the method is purposely kept quite general and based on general assumptions. Such as : the aFRR is effective for the 15 first minutes and then the tertiary reserve (Replacement Reserve, RR) and the manual Frequency Restoration Reserve (mFRR) are used to refund the aFRR. But each TSO has some specific characteristics that must be considered such as the production planning time step or the minimal activation time for the RR and mFRR.

2.2.1 The question of Δ

₁₅

The 𝛥

₁₅

is the base material of the recommended method. So, this is the first parameter to investigate. We must analyze the composition of 𝛥

₁₅

. The 𝛥

₁₅

is computed with the formula (1.8). The 1-minute mean value of the ACE open loop is used to reduce the volatility of the ACE open loop which is measured on a 5 s step. The 15-minutes mean value of the ACE open loop comes from the fact that the time range of the aFRR action is 15 minutes and past these 15 minutes it is for the mFRR or the RR to take action. It could be interpreted as a 15- minutes mean deviation which must be handled by the RR or the mFRR and not by the aFRR which only compensates the variations around this mean value.

Figure 2.2: Role of the aFRR with 𝛥15.

The role of the aFRR and then its reserve prescription depends on the characteristic of the

RR and mFRR. These characteristics are not the same in every country and if we want to

properly adapt the method for a TSO, we have to consider these characteristics and if

needed to change the Δ

₁₅

to another value such as Δ

₃₀

or Δ

₂₀

for example. Δ

₃₀

and Δ

₂₀

are

respectively the difference of the 1-minute average ACE open loop and the 30-minutes

24

average ACE open loop of the LFC Block in MW and the difference of the 1-minute average ACE open loop and the 20-minutes average ACE open loop of the LFC area in MW

To choose the right base material for the method, we will at least consider the specifications of:

• The Full Activation Time (FAT) of the RR and mFRR.

• The minimal activation time for the RR and mFRR.

• The production planning time step.

• The doctrine of the TSO (some TSO use lot more mFRR and RR than other).

• The FAT of the aFRR.

• The imbalance settlement period.

This choice will have a big impact on the aFRR prescription because the use of the aFRR over the day is different as we can see with this example done with the same ACE data but Δ

₃₀

instead of Δ

₁₅

.

Figure 2.3: Role of the aFRR with 𝛥30.

The choice of the based material has to be motivated and justified with a comparison with the recommended base value which is Δ

₁₅

. In the following description of the general procedure we will consider Δ

₁₅

as base material but the procedure is the same for any other Δ.

2.2.2 The prescription time step

In the current state of the application, the aFRR prescription is constant over the day. But

the ACE open loop (which can be interpreted as the need of aFRR) varies over the day. We

can then suppose that having a varying aFRR prescription over the day could improve the

ACE regulation. The objective is to introduce a prescription time step for the method. The

prescription time step is the time step between two aFRR prescriptions. This prescription

25

time has to consider some characteristics of the power system such as the aFRR constitution process time step. This means that if a TSO constitutes its aFRR through an aFRR market with and the clearing of the market occurs every 15 minutes, it doesn’t make any sense to have an aFRR prescription time step smaller than 15 minutes or not a multiple of 15 minutes.

Because the aFRR prescription (which is constant during one prescription time step) cannot be followed with the aFRR available in the market.

The choice of the prescription time step can have a big impact on the level of prescription.

That comes from the fact that choosing a time step can be interpreted as choosing how to regroup the 𝛥

₁₅

before taking the extreme percentile. Indeed, if we consider a 30 minutes prescription time step, this mean that the prescription corresponding to the time range 00:00:00 to 29:59:59 comes from the extreme percentiles of the group of Δ

₁₅

which is composed with all the Δ

₁₅

measured between 00:00:00 and 29:59:59. We have dispatched the Δ

₁₅

in different groups, each group corresponding to a prescription time range.

The outcome of this step is a list of possible prescription time steps that we want to investigate for future applications.

2.2.3 The grouping of days

After dividing the day by prescription time steps, we can now group days together. At the current state of the method application, we have an aFRR prescription varying over the day.

But the aFRR prescription is the same from one day to the next. The different groups are created only according to the daytime corresponding to the measure. Then the aFRR prescription is the same every day. This will remain the same until a modification of the data base. Now we want to create additional groups which depend on the day as well as the time of the day corresponding to the measure.

The objective of this step is to create day types which have a similar need in aFRR over the day and apply one aFRR prescription for each day type. The day types with a high need in aFRR will have a higher aFRR prescription than day types with smaller need in aFRR. The day type can be based on multiple factors. For example, we can consider weekdays and investigate if we can group some days together thanks to a similar aFRR prescription (and so a similar need in aFRR). But it can be something else such as group the days by average temperature and observe the impact on the aFRR prescription. At the end of this part we should have one aFRR prescription corresponding to each day type. The application of this method consists in considering one day (that we want the aFRR prescription of), determining to which day type this day belongs and then apply the corresponding aFRR prescription.

Then, it must be easy to assign one day to a day type and these day types must be based on

reliable criteria. For example, if you consider the mean temperature as criteria to define

different day types, the assignation of the day to a day type depends on the weather

forecasts. Moreover, if the TSO is responsible for a large LFC area even if the mean

temperatures of two different days are the same, their repartition over the area might be

very different.

26

The outcome of this step is several day types with distinct characteristics and distinct need in aFRR profile. The assignation of one day to a day type must be easy and reliable.

We have to make sure that the dispatching of the Δ

₁₅

into groups (the prescription time step and the day types) does not imply a too small number of Δ

₁₅

in the group. The method has a statistical approach and must have enough Δ

₁₅

values in order to have a representative percentile. A solution to a small number of values in a group is to increase the data set length. This means that we do not consider one year of historical data but two or more.

There are pros and cons to having a big data set length and it must be discussed and evaluated. In this study we have considered three different data set lengths: 1 year, 2 years and 3 years.

2.3 Evaluation of the method

The outcome of the previous part is multiple scenarios of different adaptations of the method (different time step prescription and/or different day type and/or different data set length). The objective of this part is to evaluate the different scenarios and compare them in order to choose a scenario that match the TSO requirement. The evaluation will consider three different aspects:

• The static evaluation: This evaluation looks into the statistical characteristics of the aFRR prescription

• The dynamic evaluation: This evaluation looks into the impact of the aFRR prescription on the aFRR activation and its impact on the ACE in the LFC area

• The economic evaluation: This evaluation looks into the cost of the aFRR prescription for the TSO.

2.3.1 The static evaluation

The objective of the static evaluation is to evaluate the statistical characteristics of the aFRR prescription for each scenario. We will compare the maximal, mean and minimal values of the aFRR prescription of different scenarios. This comparison might show some aFRR prescriptions that cannot be followed because of a too high maximal value. This will also probably show the influence the day types or the prescription time step.

In the static evaluation we will also evaluate is the aFRR saturation. First let’s define what we

call the aFRR saturation. As we explained in the beginning of this document in the chapter 1,

the aFRR purpose is to reduce the ACE to zero. We have at our disposal the ACE open loop

measurements. This ACE open loop represents the instantaneous need in aFRR. We consider

there is an aFRR saturation if the instantaneous need in aFRR (ACE open loop) is greater than

the aFRR. We can evaluate this saturation at the rate that the ACE measurement refresh

rate. But a short aFRR saturation will have only a small impact on the power system

frequency. That’s why we will not look into the number of aFRR saturation but the number

of long aFRR saturation (the aFRR is saturated more than 15 minutes). We will also look into

27

the number of high aFRR saturation. To do this evaluation we need to take ACE open loop values that were not used to compute this aFRR prescription. We will use the ACE open loop data from one year to evaluate scenario computed with the data coming from previous years. This will prevent any correlation bias between the method calculation and its evaluation.

This evaluation has for objective to show the impact of the different day types or prescription time steps on the aFRR prescription and its characteristics.

2.3.2 The dynamic evaluation

The objective of this evaluation is to simulate the impact of the modification of the aFRR prescription on the ACE of the LFC area. To do so we need a model of the Frequency Restoration Process. It can be modeled with different software such as openmodelica or matlab. The Frequency Restoration Process characteristics are different depending on the TSO. The model has to be adapted to the power system we want to evaluate. For the same reason as in the previous part with the static evaluation, we will use the ACE open loop data from one year to evaluate scenario computed with the data coming from previous years to prevent the correlation bias. The final objective is to compute the FRCE target parameters with the ACE coming from the simulation.

This evaluation has for objective to show the impact of the different day types or prescription time steps on the simulated FRCE target parameters.

The process activation structure comes from the process described in [2] and the

signification of different signs is given in annex B:

28

Figure 2.4: Control process.

In this figure, we have all the Load Frequency Control process, but we are only interested in the Frequency Restoration Process. As explained in chapter 1 the input of the aFRR controller has two components: Δ𝑃 and 𝐾(Δ𝑓 + 𝑓

_𝑜𝑓𝑓

). When you add up these two components you obtain the ACE. And the objective of the frequency control process is to reduce the ACE to zero. If we want to do a complete modeling of the power system in order to evaluate the impact of the level of aFRR prescription on the FRCE target parameters, we need to simulate the impact on both the frequency (Δ𝑓) and the power through interconnections (Δ𝑃). To simulate the frequency, we need the system inertia that depends on the number and the type of the producing power plants in the synchronous area. This inertia varies over the day. This way of simulating the power system is complex and requires a lot of data such as power plant planning, interconnection capacity or neighbor’s behaviour.

To solve this problem of complexity and resources, we decided to simplify the model and

only consider the ACE and not Δ𝑓 and Δ𝑃 separately. The model might be less precise than

the complex one, but it can still give an idea of the evolution of the FRCE target parameters

with the different scenarios. The modeling has been made with openmodelica.

29

Figure 2.5: The aFRR controller on openmodelica.

The input data are the 𝐴𝐶𝐸

_𝑜𝑙

, the positive and negative aFRR prescription and the IGCC.

IGCC stand for International Grid Control Cooperation [4]. This is the 𝑃

_{𝑖𝑛𝑡,𝑣}

from the Figure 2.4, and it represents the imbalance netting between the TSOs through virtual tie lines. The output is the activation of the aFRR. The filter at the output is smoothing the activation. In this model we suppose that, for the scenarios, the aFRR prescription has been respected and that the scheduled aFRR equals the aFRR prescription.

The ACE we need to simulate the FRCE target parameters is the output of “sum_IGCC”:

𝐴𝐶𝐸

_𝑜𝑙

+ 𝑎𝐹𝑅𝑅

_{𝑎𝑐𝑡𝑖𝑣𝑎𝑡𝑒𝑑}

+ 𝐼𝐺𝐶𝐶 = 𝐴𝐶𝐸 (2.1)

Now the only block that we did not talk about yet is the block modeling the aFRR controller

and the French system for the aFRR activation. The model of this block is represented in

Figure 2.6:

30

Figure 2.6: Modeled aFRR controller.

To explain this controller let’s start with the input/output. The inputs are the ACE after the IGCC correction (u), and both negative (Min) and positive (Max) reserve limits. The output is the aFRR activation (y). First, we have a PI controller in green with a proportional and an integral part. To calibrate this model, we have to adapt:

• The gain of the proportional part.

• The gain and the characteristic time of the integral part.

The output of the PI controller is the set-point for the aFRR activation. For the next step we

need to adapt the process according to the sign of the set-point because the negative and

positive aFRR limits are different. That’s the role of the block at the PI output. The output of

this block is either the set point or zero. The top output is positive (the set point if the set

point is positive or zero) and the bottom output is negative. Then comes the slew rate limiter

(blue rectangles). The specification for the power plant taking part in the aFRR is that they

must have a required Full Activation Time. The blue rectangle is limiting the slew rate to a

maximum slew rate equal to the Full Activation Time. The set point is divided by the reserve

limits and the slop is limited to a maximum value which is going from 0 to 1 in the range of

the FAT. The set point is then multiplied by the reserve limit. The slew rate limitation has

31

been modeled this way because the reserve limits are changing during the simulation and are different for the negative and positive reserves. The next step is to limit the output to the level set by the aFRR prescription (purple block). Finally, there is an anti-windup control to prevent integration wind-up in the PI controller.

2.3.3 The economic evaluation

The objective of this evaluation is to measure the economic impact of changing the aFRR prescription method. This evaluation will be based on an algorithm that depends on the aFRR constitution process of the TSO. The algorithm will evaluate the cost of the aFRR constitution for the TSO, the social welfare and the mean marginal cost. For this evaluation there is no need to prevent the correlation bias as we are not working with the ACE anymore.

The algorithm has to be adapted according to the aFRR constitution process. Hereafter will be presented the model of the algorithm used in the case of the French power system. This algorithm cannot be directly applied to another power system. The base of this algorithm is the aFRR constitution process which is not market based (yet) in France.

The base materials used for this algorithm are:

• The spot prices corresponding to the evaluated period.

• The production planning corresponding to the evaluated period.

• The aFRR prescription corresponding to the evaluated period.

Most of the information are in the production planning files. In these files, we have for every 30 minutes in the year (production time step) for each power plant operating in France:

• The type of power plant (nuclear, hydro, …).

• The minimum and maximum output.

• The maximum aFRR participation.

• The scheduled power production.

• The scheduled aFRR participation.

Based on this information we can compute the opportunity loss. This is the amount of money that the producers are losing when they take part in the aFRR constitution instead of biding it in the electricity market. For the positive reserve, the power plant must reduce his production under his maximal output in order to modulate the output if needed for the aFRR activation. They can then not bid this margin on the electricity market. Then the opportunity loss is:

𝑜𝑝𝑝𝑜𝑟𝑡𝑢𝑛𝑖𝑡𝑦 𝑙𝑜𝑠𝑠 = max ((𝑃𝑟𝑖𝑐𝑒

_{𝑠𝑝𝑜𝑡}

− 𝑚𝑎𝑟𝑔𝑖𝑛𝑎𝑙 𝑐𝑜𝑠𝑡), 0) in €/MW.h (2.2)

32

The point of the algorithm is to constitute the aFRR with a minimal cost to society (maximizing the social welfare). We will then do the clearing between the aFRR prescription and the available aFRR by increasing opportunity loss. It can happen that the available aFRR is smaller than the aFRR prescription. Then we must start an additional power plant in order to respect the aFRR prescription. This is called a Must Run (MR). We need an approximation for the Must Run opportunity loss. As an illustration, in France these Must Run are mainly handed by the hydro power plants. The hydro power plants are using the water they have in their reservoir in order to fulfill the requirement. The hydro power plant can provide around 15 % of his output as aFRR. Then if we need 20 MW for the aFRR to be produce by a Must Run, we need to start around 140 MW of hydro. And these 140 MW comes from the reservoir and cannot be used later to maximize the hydro power plant surplus. We consider that the power plant could have bid his power at the most expensive spot price of the current day and then the power plant have lost the difference between the two spot prices multiplied by the total production (around 7 times the needed aFRR). For indication, we can then define an adapted formula to compute the opportunity loss for the Must Run:

𝑜𝑝𝑝𝑜𝑟𝑡𝑢𝑛𝑖𝑡𝑦 𝑙𝑜𝑠𝑠 𝑓𝑜𝑟 𝑎 𝑀𝑅 (𝑡) = 7 ∗ (max

day

(𝑠𝑝𝑜𝑡 𝑝𝑟𝑖𝑐𝑒) − 𝑠𝑝𝑜𝑡 𝑝𝑟𝑖𝑐𝑒(𝑡)) in €/MW.h (2.3) The opportunity loss of a Must Run depends on the spot price at the moment of the Must Run. Then a Must Run during high prices is not as expensive as a Must Run during low prices.

For each 30 minutes time step we have a marginal opportunity loss. The energy price for the aFRR constitution is currently fixed by RTE and is constant during the year. In the coming years, the constitution of the aFRR will be based on market mechanism. The assumption is that the power plant will bid their opportunity loss in this market. The total cost for RTE will come from the marginal opportunity loss multiplied by the total amount of aFRR in the corresponding 30 minutes time step.

The output of the algorithm is:

• The number and the volume of Must Run

• The total cost of the Must Run

• The total opportunity loss (considered as the cost to society)

• The total cost for the TSO

This algorithm is based on different assumptions that should be met in the future with the introduction of the European aFRR market. The most important assumption is the marginal cost for the different production technologies (that will not be disclosed due to confidentiality rules).

We have considered that the hydro power plants do not have an opportunity loss. Because

they can store the unused water for the electricity production for a later time. We have also

assumed that the opportunity loss of the negative reserve was equal to zero except for the

Must Run negative reserve. Because we still have to start the additional power plant and use

33

the water in the reservoir to do the negative reserve in case of insufficient negative aFRR available. This assumption also implies that the power plants are able to have dissymmetric aFRR capacity. Otherwise we would have to take the same amount of positive and negative reserve for each power plant and to have a dissymmetric aFRR prescription will be useless as we would simply consider the maximum of the absolute values of positive and negative aFRR prescription. The last assumption is that for the target method scenarios, the scheduled aFRR equals the aFRR prescription.

2.3.4 Selection of the scenario

After these three evaluations, the scenarios can be ranked according to their results on the

different evaluations. The selected scenario will then depend on the TSO’s doctrine and

objective. For example, a TSO has to find the balance between the different aspects that

satisfy its objectives.

34

3. Adaptation of the method

3.1 Choice of the Δ as base material

3.1.1 The characteristics of the French system

The characteristics of the French system are the assumptions used for the following study and are the following [8][9]:

• The production planning has a 30-minute time step.

• The imbalance settlement period is 30 minutes.

• The participation to the aFRR does not concern every power plant (the wind and solar power plants do not take part to the aFRR constitution).

• This participation is function of the generation type as a percentage of the production (5 % for a nuclear power plant, 10 % for a thermal power plant, around 15

% for the hydro).

• The activation of the aFRR does not follow the merit order. This is a pro rata of the available reserve monitored by the level N.

• Each generator taking part in the aFRR has a 400 s FAT (from 0 MW to the maximum available reserve) in a normal situation.

• The FAT for the RR and mFRR is between 15 minutes and 30 minutes depending on the technology.

• The minimal activation time for the RR and mFRR is 30 minutes.

• The mean value of the scheduled aFRR for 2019 is 712 MW. (As the reserve is currently symmetric, the mean value is the same for both the positive and negative reserve).

These characteristics may evolve in the coming year and the results of the following study may evolve with the evolution of the following assumptions.

3.1.2 The introduction of Δ

₃₀

In consideration of the French system, we need to confront the French mFRR and RR specifications to the Δ

₁₅

definition to make sure both visions are consistent. As we said in the previous part paragraph, the Full Activation Time of some of the mFRR and RR is 15 minutes which is consistent with the Δ

₁₅

definition but, the minimal activation time for the RR and mFRR is 30 minutes. Also, the production planning has a prescription time step of 30 minutes. We can then think that for the French power system, this is not the 15 minutes mean deviation that is handled by the RR and the mFRR but the 30 minutes mean deviation.

We can suggest using Δ

₃₀

as the base material of the method. Δ

₃₀

is computed as:

𝛥

₃₀

= 𝑚𝑒𝑎𝑛

_{1𝑚𝑖𝑛𝑢𝑡𝑒}

(𝐴𝐶𝐸

_𝑜𝑙

) − 𝑚𝑒𝑎𝑛

_{30𝑚𝑖𝑛𝑢𝑡𝑒𝑠}

(𝐴𝐶𝐸

_𝑜𝑙

) (3.1)

35

With Δ

₃₀

as base material we still have one Δ

₃₀

for each minute, then the rest of the method does not change. And we can do the direct application of the ENTSO-E’s method with the Δ

₃₀

distribution and the ACE open loop data of the period 2016-2018 (3 years):

Figure 3.1: Distribution of the Δ30 and the corresponding percentile.

The resulting aFRR prescription would be a constant value of 757 MW for the positive

reserve and 795 MW for the negative reserve. These values are both higher than the current mean scheduled aFRR. We can suppose that the FRCE target parameters will be better than with the current aFRR prescription which to a mean scheduled aFRR level about 50 MW smaller than with the Δ

₃₀

application. This method seems interesting as the outcome of the method is comparable to the current scheduled aFRR. The comparison of the mean values is not sufficient to have an idea of the impact of the aFRR on the frequency quality.

3.1.3 Comparison between Δ

₃₀

and the historical aFRR

To do the first application we have used the ACE data measured between 2016 and 2018. To evaluate this application, we will base our comparison on the 2019 ACE data. There are mainly two reasons to do so. It is like considering that the ENTSO-E’s method have been used for the aFRR prescription in 2019. And as the prescription of aFRR with the recommended method is based on the historical ACE data, this precaution prevents any correlation bias between the method and its evaluation.

The comparison will be based on three different points:

• The mean value of aFRR during the year of 2019

• The proportion of saturation of the aFRR i.e. the proportion of time where the 𝐴𝐶𝐸

_𝑜𝑙

is greater than the aFRR (5 s time step)

• The mean difference between the 𝐴𝐶𝐸

_𝑜𝑙

and the aFRR when the 𝐴𝐶𝐸

_𝑜𝑙

is greater

than the aFRR

36

These points will be evaluated for both the positive and negative aFRR. In order to have a better comparison we will use two historical data for the current prescription method: the historical aFRR prescription and the historical scheduled aFRR.

Table 3.1: comparison between the current state and the application with Δ_30.

Evaluation criteria 𝚫

_𝟑𝟎

Scheduled

aFRR

Historical aFRR prescription

Mean value (in MW) Positive 757 712 650

Negative -795 -712 -650

Proportion of saturation Positive 11,30 % 13,20 % 15,39 %

Negative 11,15 % 13,97 % 15,59 %

Mean value of saturation (in MW)

Positive 295 297 316

Negative 344 355 363

This table shows that the direct application of the method with the Δ

₃₀

has better results than with the historical data from the current method. Also, we can notice that for the historical data (scheduled aFRR and aFRR prescription) produced by the current method, the results are better for the positive deviations than the negative deviations. On the contrary, for the target method, the negative aFRR has a smaller proportion of saturation

¹

than the positive aFRR. This shows one of the target method’s advantages: it is dissymmetric and the higher the historical negative deviation the higher the positive aFRR prescription.

With this table, we can assume that the target method applied with Δ

₃₀

is consistent with the current situation in France and will provide an acceptable aFRR prescription.

3.1.4 Choice of Δ

₃₀

Regarding the information given in Table 3.1, we have made the choice to keep Δ

₃₀

as base material of the method for the further investigations. The Δ

₃₀

is more consistent with the French power system’s characteristics than the Δ

₁₅

. The quick evaluation of the direct application of the ENTSO-E’s method provides good results for the Δ

₃₀

application with a prescription in the same range as the current prescription.

These results might come from the French system characteristics such as the Full Activation Time of the mFRR and RR or the production planning time step. If these characteristics change in the future the results of this comparison can be different. If a major evolution occurs in the French power system or its regulation, then this decision must be challenged.

1 Saturation of aFRR shall be considered as normal situation in the French system since RTE is using mainly mFRR product to balance the system.