(PDF) Hybrid Wind, Solar and Storage Power Plant in Electricity Market

DTU Wind Energy

Master of Wind Energy

HybridWind,SolarandStoragePowerPlant

inElectricityMarket

Emilio Barrachina Gascó

DTU Wind Energy Master-0010

July 2020

DTU Wind Energy Master - 0010

August 2020

2nd version

Hybrid Wind, Solar and Storage

Power Plant in Electricity Market

DK2 and SE2 Hybrid Power Plants

Supervisors:

Professor Poul Ejnar Sørensen

Researcher Kaushik Das

External supervisor:

Professor Henrik Stiesdal

Student:

Emilio Barrachina Gascó

Fanzara, Spain

August 6th 2020

b

Preface

This master thesis was prepared at DTU Wind Energy Department at the Technical

University of Denmark in fulﬁllment of the requirements for acquiring the DTU Wind

Energy Master diploma.

The supervisors for this master thesis have been professor Poul Ejnar Sørensen and

researcher Kaushik Das, both from DTU Wind Energy Department.

The external supervisor for this master thesis has been professor Henrik Stiesdal from

Stiesdal A/S company.

The complete period of work for this master thesis development has been ﬁve months,

starting on February 2020 till the end of June 2020.

Fanzara, Spain

July 1st, 2020

Emilio Barrachina Gascó

MEng and MSc

emilio.barrachinagasco@iicv.es

Abstract

During the last two decades we have seen how the renewable energy sources have been

expanded all around the world evolving hand in hand because of the increase of envi-

ronmental conscious market acceptance. These renewable energy sources have a great

challenge which is the inherent variability as they have stochastic renewable power

generation.

The unreliability of wind and solar photovoltaic to generate the energy demanded

is the main reason for including energy storage technologies so that we gain dispatchabil-

ity of WPPs and PVPPs. Energy storage systems have a really special advantage in this

sense, as they can shift energy generation to later periods with higher electricity prices

and energy demand.

We are living really interesting times if we talk about hybrid wind and solar with

energy storage systems and I’m pretty sure they’re going to be crucial in the near future,

and so this can happen, integration of several energy storaging within the hybrid wind

and solar pv plants is key to boost penetration of these technologies into the energy

market.

Index terms

CorRES, time series, hybrid wind and solar pv power plant, battery energy storage sys-

tem, wind power, solar photovoltaic power, correlation coeﬃcient, spot market prices

and RES negative correlation, power and energy surplus, bidding model methodol-

ogy, real time model methodology, forecast error minimization, reducing curtailment,

reducing penalties.

Future perspectives

Last summer when I was attending to the Wind Energy Master Graduation week event

at DTU Risø Campus many seminars were oﬀered within its fantastic and complete

program prepared by Merete Badger and Nina Julh Madsen. Among them a Henrik

Stiesdal’s seminar aimed to last year ﬁnalist master students, Master of Wind Energy:

What now?.

After listening to Henrik I decided that in my master thesis he had to be included

due to his huge impact within the wind energy ﬁeld of course, and for his impressive

enthusiasm communicating, vision of the future and extensive renewable knowledge in a

ﬁeld as exciting as wind energy, that will undoubtedly mark the future in a signiﬁcant

way for the next generations.

So we have had recently a phone call interview talking about many interesting and

relevant energy matters as follows. Needless to say that I am deeply grateful for the kind

attention that Henrik Stiesdal has always had with me in the course of this master thesis.

Henrik Stiesdal is going to place some thermal storage technology in the ﬁrst world’s

energy artiﬁcial island which is going to be located at the Danish North Sea. It’s A

10 [

GW

]oﬀshore wind and host electricity storage and power-to-X as well as housing, OM

facilities and HVDC converters for transmission and interconnectors. More information

can be consulted at Copenhagen Infrastructure Partners’s website.

This project is going to be developed in 3 stages in line with increasing Danish electricity

demand so Henrik Stiesdal’s participation will take place once the project has progressed

through the several phases planned but Henrik has already provided the project promoter

a pilot for phase one. Stiesdal Storage Technologies A/S has developed a thermal battery,

a grid-scale energy storage concept that can provide a backup to RES for a much longer

period than conventional batteries do.

iv Future perspectives

Talking to Henrik Stiesdal I asked him about HPP technologies and future perspectives

so he thinks that:

"HPP technology is going to be necessary in most places of the world. You can al-

ways introduce more load we could say that we solve things by avoiding having overﬂow

or wind and solar power production by having battery systems so then we can use the

excess power. But unless you accept very low eﬃciency of power rates then you’ll have

problems in regenerating when you don’t have enough wind and solar. So the half of the

problem is the overﬂow but we can’t avoid the other half of the problem that’s the under

supply of energy and for that we need storage. Storage of electrical energy promises to

make wind and solar power more viable by oﬀering a solution to the ﬂuctuations in the

energy supply they produce".

"At the present time there is no a direct energy market for storage but it wasn’t a

wind market when we started there it was all about the subsidies in the early years, there

were subsidies in California, subsidies in Denmark and then the market came, and the

same will happen with the storage, it has to be some kind of support mechanism".

"But of course you can also have good part of your revenue from capacity, in other

words you’re always ready to deliver power even though you might not need it now you

actually get paid per

kW month

basis rather than per

kW h

basis. And the capacity market

is quite huge in many parts of the world, at least in the US, so in a well deﬁned capacity

market over there where you get a certain payment per

kW

and then we can do. So I do

think that can help in the distance case, so it’s not purely upping fast buying cheaper and

selling expensive".

Finally talking to Henrik about seasonal storaging and the variable demand that power

production has to face at every single moment he says: "normal ly say that we need three

type of storages: the short terms that batteries can do, that’s like half an hour or an hour

or so and then we need the mid-term that’s what we do with thermal storage as that can

do for a week, and then we need the seasonal storage which have to rely on chemistry so

it’s all about hybrid them and that can last of course. There is of course a limit on the

amount we can store but then we have ammonia instead, which is easy to keep so that

way for sure we’ll be able to solve it with these three types of storages"

Henrik Stiesdal

Wind power pioneer

CEO Stiesdal A/S

GWEC Ambassador

Taksigelser

Jeg vil gerne udtrykke min dybe og respektfulde tak til mine vejledere på hovedopgaven:

Professor Poul Ejnar Sørensen og Researcher Kaushik Das fra DTU Vindenergi, for deres

fantastiske vejledning, assistance, hjælp og støtte gennem hovedopgaven.Især i disse

måneder med COVID-19 med alt dette har forårsaget og konditioneret arbejdet med

universitetsundervisning og -forskning. Jeg værdsætter højt deres markante bidrag i form

af tid og ideer, hvilket har gjort, at dette interessante arbejde har været en stimulerende

og berigende oplevelse gennem 5 måneder.

Jeg vil også gerne udtrykke min dybtfølte tak til min eksterne vejleder: Henrik Stiesdal,

vindpioner fra Stiesdal A/S, for hans værdifulde samarbejde, venlig støtte og rådgivning

under dette arbejde.

Jeg vil også gerne udtrykke min tak til Researcher Matti Juhani Koivisto og Post-

doc Juan Pablo Murcia León fra DTU Vindenergi for deres venlige bidrag og støtte i

form af prognoser og målt energiproduktion til brug for denne hovedopgave.

Jeg vil gerne udtrykke min tak til Senior Scientist, Head of Master Studies, Merete Badger

hende altid fremragende eﬀektiv opmærksomhed. Jeg vil gerne udtrykke min tak til

Sekretær og Master Programme Coordinator Nina Juhl Madsen hende altid fremragende

eﬀektiv opmærksomhed.

Jeg vil gerne udtrykke min tak til alle undervisere involveret i DTU Master i Vin-

denergi for deres entusiasme, dedikation og undervisning igennem de 9 kurser.

Jeg vil gerne udtrykke min tak til mine medstuderende: Didde Okholm Kvorning,

Peter Fausbøll, Stephan Johannes Østergaard, José Alberto Navarro Martínez og Laín

Nieto Gómez. Det var virkelig en fornøjelse at være i masterprogrammet sammen med dig.

Jeg vil gerne udtrykke min tak til min ven, universitetskollega og forsker: Jaime Zabalza

Ostos fra Centre for Signal and Image Processing - University of Strathclyde(Glasgow,UK),

for hans gode råd i forbindelse med udarbejdelsen af denne hovedopgave.

Jeg vil gerne udtrykke min dybeste taknemmelighed til min moder, Juana Gascó Vivas,

til min fader, Emilio Barrachina Martí og til min tante Encarna Gascó Vivas, for al deres

indsats i form af at give mig min uddannelse.

Jeg vil gerne udtrykke min dybeste taknemmelighed til min onkel, Juan Carlos Gascó

Vivas, for hans super støtte og gode eksempel på kamp, som for os alle er med hans

daglige indsats i de vanskelige øjeblikke, som livet har sat ham igennem.

vi Taksigelser

Jeg vil gerne udtrykke min dybeste tak til min elskede kæreste, Sonia Marmaneu Vidal,

for al hendes forståelse og støtte under sådant et tidskrævende arbejde.

Sidst, men ikke mindst, tak til Audrey, min elskede hund, som har været sammen

med mig i 13 år, men som desværre gik bort i april måned.

Oversættelse foretaget af min ven og medlærer, Peter Fausbøll

Acknowledgements

I would like to express my most sincere and respectful gratitude to my master thesis supervisors:

Professor Poul Ejnar Sørensen and Researcher Kaushik Das from DTU Wind Energy department,

for their fantastic guidance, assistance, help and support throughout this master ﬁnal project.

Especially during these months of COVID-19 with all that this has caused and conditioned the

work of university teaching and research. I really appreciate their notable contributions of time

and ideas provided, which have made this interesting work a very stimulating and enriching

experience during these 5 months period.

I would like also to express my sincere thanks to my external supervisor: Henrik Stiesdal, wind

industry pioneer from : Stiesdal A/S company, for their valuable cooperation, kind support

and advise during this work.

I would like like to express my gratitude to researcher Matti Juhani Koivisto and postdoc

Juan Pablo Murcia León from DTU Wind department for their kind contribution and support

providing forecast and measured power production data for this master thesis.

I would like to express my more sincere thanks to senior scientist and Head of Master Studies

Merete Badger her always excellent eﬃcient attention.

I would like to express my more sincere thanks to secretary and Master Programme Co-

ordinator Nina Juhl Madsen her always excellent eﬃcient attention.

I would like to express my sincere thanks to all the professors involved in the DTU Mas-

ter in Wind Energy for their enthusiasm, dedication and teaching all along the 9 courses.

I would like to express my gratitude to my master colleagues: Didde Okholm Kvorning,

Peter Fausbøll, Stephan Johannes Østergaard, José Alberto Navarro Martínez y Laín Nieto

Gómez, it has been a real pleasure sharing with them all this master time.

I would like to express my gratitude to my friend, university colleague and great researcher:

Jaime Zabalza Ostos from Centre for Signal and Image Processing - University of Strathclyde

(Glasgow,UK), for all his wise advise all along this master thesis work.

I would like to express my deepest gratitude to my mother, Juana Gascó Vivas, to my

father, Emilio Barrachina Martí and to my auntie Encarna Gascó Vivas, for all their eﬀorts

having given me the education they gave me.

I want to express my deepest gratitude to my uncle, Juan Carlos Gascó Vivas, for his super

support and great example of struggle that for all of us is with his daily eﬀort in the diﬃcult

viii Acknowledgements

moments that life has put him through.

I would like to express my deepest gratitude to my beloved girlfriend, Sonia Marmaneu

Vidal, for all her understanding and support in such a time-demanding job.

Last but not least to Audrey, my beloved dog who has been together with me during these last

13 years that sadly passed away last April month.

Agradecimientos

Deseo expresar mi más sincera y respetuosa gratitud a mis supervisores de tesina: Catedrático

Poul Ejnar Sørensen y al Doctor Kaushik Das del departamento DTU Wind Energy por su

fantástica guía, asistencia, ayuda y apoyo durante esta tesina. Especialmente durante estos

meses de COVID-19 con todo lo que ello ha ocasionado y condicionado las labores de docencia

universitaria e investigación. Realmente aprecio sus notables contribuciones e ideas aportadas,

las cuales han hecho de este interesante trabajo una experiencia muy estimulante y enriquecedora

durante estos 5 meses.

Querría expresar mi más sincero agradecimiento a mi supervisor externo de tesina: Hen-

rik Stiesdal, pionero de la industria eólica de la compañía: Stiesdal A/S, por su valuosa

cooperación, apoyo y consejo durante este trabajo.

Me gustaría expresar mi gratitud al cientíﬁco investigador Matti Juhani Koivisto y al post-

doctorando Juan Pablo Murcia León del departamento DTU Wind Energy por su amable

contribución y apoyo facilitando los datos de la previsión y medida de producción de potencia

utilizados en esta tesina.

Deseo expresar mi más sincero agradecimiento a la investigadora cientíﬁca senior y Direc-

tora del Máster Merete Badger su siempre excelente atención.

Querría expresar mi más sincero agradecimiento a la secretaría y Coordinadora del Máster

Nina Juhl Madsen su siempre excelente atención.

Me gustaría trasladar mi más sincero agradecimiento a todos los profesores implicados en

el máster de energía eólica de la DTU, por su entusiasmo, dedicación y enseñanzas a lo largo

de los nueve cursos que comprende este máster.

Quiero expresar mi gratitud a mis compañeros de clase: Didde Okholm Kvorning, Peter

Fausbøll, Stephan Johannes Østergaard, José Alberto Navarro Martínez y Laín Nieto Gómez,

ha sido un verdadero placer compartir con ellos todo este tiempo en el máster.

Querría expresar mi gratitud a mi gran amigo, compañero de carrera and gran cientíﬁco

investigador: Jaime Zabalza Ostos del Centre for Signal and Image Processing - University

of Strathclyde (Glasgow,Reino Unido), por sus sabios consejos durante todo el trabajo de la

tesina.

Querría expresar mi más profunda gratitud y reconocimiento a mi madre, Juana Gascó

Vivas, a mi padre, Emilio Barrachina Martí y a mi tía Encarna Gascó Vivas, por todos sus

esfuerzos para ofrecerme toda la educacion que ellos me han dado.

xAgradecimientos

Deseo expresar mi más profunda gratitud a mi tío Juan Carlos Gascó Vivas, por su apoyo y

gran ejemplo de lucha que para todos nosotros és en su esfuerzo diario en los momentos difíciles

que la vida le ha hecho pasar.

Querría expresar mi más profunda gratitud a mi querida novia, Sonia Marmaneu Vidal,

por toda su comprensión y apoyo en un trabajo tan demandante de tiempo.

Por último pero no por ello menos importante a Audrey, mi querida perra quien ha estado

conmigo durante estos últimos trece años y que tristemente falleció el pasado mes de abril.

List of Figures

1.1 Hybrid power plant + battery energy storage system . . . . . . . . . . . . . . . . . 2

1.2 TopologyofconnectedHPP............................... 2

1.3 Types of HPPs: integration and operation of diﬀerent generating modules . . . . . 3

1.4 World HPP and RES capacity share in 2024 . . . . . . . . . . . . . . . . . . . . . 3

2.1 Danish energy system - June 29th 2020 18:51 hrs . . . . . . . . . . . . . . . . . . . 5

3.1 Nordic area wind conditions - July 29th 2020 6.51 pm . . . . . . . . . . . . . . . . 8

3.2 DK2HPPlocation.................................... 9

3.3 DK2HPP......................................... 9

3.4 SE2HPPlocation .................................... 10

3.5 SE2HPP......................................... 10

3.6 CorRESblockdiagram ................................. 12

3.7 Scatter plot: Power generated per hour vs El. Spot market prices . . . . . . . . . 13

5.1 Bidding model algorithm ﬂowchart . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

5.2 Bidding model algorithm ﬂowchart - Charging detail . . . . . . . . . . . . . . . . . 21

5.3 Bidding model algorithm ﬂowchart - Discharging detail . . . . . . . . . . . . . . . 22

5.4 Bidding model algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

5.5 DK2Bidding-Scenario9................................ 25

5.6 DK2Bidding-Scenario9................................ 25

5.7 DK2forecast-Scenario9................................ 26

5.8 DK2 forecast - Scenario 9 - P_LIMbatt = 50MW .................. 26

5.9 DK2forecast-Scenario9................................ 27

5.10 DK2 forecast - Scenario 9 - P_LIMbatt = 50MW .................. 27

5.11 DK2 forecast - Scenario 9 - P_LIMbatt = 100MW .................. 28

5.12 Real time model algorithm ﬂowchart . . . . . . . . . . . . . . . . . . . . . . . . . . 30

5.13 DK2 measured - Scenario 9 - P_LIMbatt = 5MW .................. 32

5.14 DK2 measured - Scenario 9 - P_LIMbatt = 100MW ................. 32

5.15 DK2 measured - Scenario 9 - P_LIMbatt = 5MW and P_LIMbatt = 10MW . . . 33

5.16 DK2 measured - Scenario 9 - P_LIMbatt = 50MW ................. 33

5.17 Days with low Psurplus(t)−Pbatt = 5 and 10 [M W ].................. 34

5.18 Days with low Psurplus(t)−Pbatt = 50 and 100 [M W ]................. 34

5.19 Full model algorithm ﬂowchart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

6.1 DK2 Forecast duration curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

6.2 DK2 Measured duration curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

6.3 DK2 P_H P P _F(t)-P_LIMbatt = 5[M W ]Ebatt = 100[M W h].......... 40

6.4 DK2 P_H P P _F(t)-P_LIMbatt = 25[M W ]Ebatt = 100[M W h]......... 41

6.5 DK2 P_H P P _F(t)-P_LIMbatt = 50[M W ]Ebatt = 100[M W h]......... 41

xii List of Figures

6.6 DK2 P_H P P _F(t)-P_LIMbatt = 75[M W ]Ebatt = 100[M W h]......... 42

6.7 DK2 P_H P P (t)-P_LIMbatt = 5[M W ]Ebatt = 100[M W h]............ 43

6.8 DK2 P_H P P (t)-P_LIMbatt = 25[M W ]Ebatt = 100[M W h]........... 43

6.9 DK2 P_H P P (t)-P_LIMbatt = 50[M W ]Ebatt = 100[M W h]........... 44

6.10 DK2 P_H P P (t)-P_LIMbatt = 75[M W ]Ebatt = 100[M W h]........... 44

List of Tables

3.1 DK2 and SE2 RES scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3.2

DK2 HPP correlation coeﬃcients - Power production per hour vs Spot market prices

13

3.3 SE2 HPP correlation coeﬃcients - Power production per hour vs Spot market prices 14

3.4 Capacity factor and correlation of wind and solar power plant . . . . . . . . . . . 14

4.1 DK2 and SE2 RES E_RES_F - Scenario 9 - Forecast . . . . . . . . . . . . . . . . 16

4.2 DK2 and SE2 RES E_RES - Scenario 9 - Measured . . . . . . . . . . . . . . . . . 16

4.3 DK2 and SE2 RES CF and FWH - Scenario 9 - Forecast . . . . . . . . . . . . . . 16

4.4 DK2 and SE2 RES CF and FWH - Scenario 9 - Measured . . . . . . . . . . . . . . 16

4.5 DK2 and SE2 RES E_surplus_F - Scenario 9 - Forecast . . . . . . . . . . . . . . 17

4.6 DK2 and SE2 RES E_surplus - Scenario 9 - Measured . . . . . . . . . . . . . . . . 17

6.1 DK2 Ecurtailed_F(t) [GW h/year]............................ 46

6.2 SE2 Ecurtailed_F(t) [GW h/year]............................ 47

7.1

DK2 and SE2 Scenario 9

P_RES_F E

(

t

):

MAE_RES_F E

(

t

)and

RMSE_H P P _F E

(

t

)

49

7.2

DK2 Scenario 9

P_H P P _F E

(

t

):

MAE_H P P _F E

(

t

)and

RMSE_H P P _F E

(

t

)

50

7.3 SE2 Scenario 9 P_H P P _F E(t):MAE_H P P _F E(t)and RMSE_H P P _F E(t)50

7.4 DK2 and SE2 Scenario 9 P_H P P _F E(t)reduction[%] ............... 50

1 DK2 RES Scenarios comparison - Energy Produced and Surplused . . . . . . . . . 65

2

DK2 RES Scenarios comparison - Energy Produced Measured, Surplused and Curtailed

65

3

DK2 RES Scenarios comparison - Energy Produced Measured, Surplused and Curtailed

66

4 SE2 RES Scenarios comparison - Energy Produced and Surplused . . . . . . . . . 67

5 SE2 RES Scenarios comparison - Energy Produced and Surplused . . . . . . . . . 67

6 SE2 RES Scenarios comparison - Energy Produced and Surplused . . . . . . . . . 68

AEP Annual Energy Production [TWh/year]

BESS Battery Energy Storage System

CHP P Hybrid power plant cost[MW/Me]

CHP P _FHybrid power plant forecast cost [MW/Me]

Crate

Rate at which a battery is charged/discharged relative to its maxi-

mum capacity [hour]

CF Capacity Factor measured [%]

C F _FCapacity Factor forecast [%]

CorRES Correlations in Renewable Energy Sources

DA Day Ahead

Ebatt Battery Energy Capacity [MWh]

Ecurt Annual Energy Curtailed measured [GWh/year]

Ecurt_FAnnual Energy Curtailed forecast [GWh/year]

Ecurt_FAnnual Energy Surplus forecast [GWh/year]

Esurplus Annual Energy Surplus measured [GWh/year]

FWH Full work hours [hours/year]

GW EC Global Wind Energy Council

H P P Hybrid Power Plant

H P S Hybrid Power System

MAE Mean Absolut Error [MW]

M AF E Mean Absolut Forecast Error [MW]

P_H P P (t)Hybrid Power Produced [MW]

P_H P P _F(t)Hybrid Power Produced Forecast [MW]

P_H P P _F E(t)Hybrid Power Produced Forecast Error[MW]

Nomenclature

xvi Nomenclature

P_LIM batt Battery charging/discharging capacity [MW]

P_SOC headroom(t)State Of Charge Headroom Power [MW]

P_SOC headroom_F(t)State Of Charge Headroom Power Forecast [MW]

P RES(t)Wind and Solar Power Production Generated [MW]

P RES_F(t)Wind and Solar Power Production Forecast [MW]

Pcharge (t)Power Charged [MW]

Pcurt(t)Power Curtailed Generated [MW]

Pcurt_F(t)Power Curtailed Forecast [MW]

Pdischarge (t)Power Discharged [MW]

Pdischarge max(t)Maximum Power Discharge [MW]

Pgrid Power Grid Connection Constrain[MW]

Pgrid headroom(t)Grid Headroom Power[MW]

Pmissing(t)Power Missing Generated [MW]

Pshortage (t)Power shortage Generated [MW]

Psurp(t)Power Surplus Generated [MW]

Psurp_F(t)Power Surplus Forecast [MW]

P P Power Plant

P V P P Solar Photovoltaic Power Plant

RCorrelation coeﬃcient

RES Renewable Energy Source

RMSE Root Mean Square Error [MW]

RM SF E Root Mean Square Forecast Error[MW]

SOC(t)State of Charge measured [p.u.]

SOC(t)_FState of Charge forecast [p.u.]

SOC0(t)Initial State of Charge measured [p.u.]

SOC0_F(t)Initial State of Charge forecast [p.u.]

T SO Transmission System Operator

V RE Variable Renewable Energy

WPP Wind Power Plant

Contents

Preface i

Abstract ii

Future perspectives iii

Taksigelser v

Acknowledgements vii

Agradecimientos ix

List of Figures xi

List of Tables xiii

Nomenclature xiv

Contents xvii

1Introduction 1

1.1 Aim and motivation of the study . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Objectiveofthestudy................................ 1

1.3 HPPwithBESSdesign ............................... 2

1.4 Advantages/values of hybrid power plants . . . . . . . . . . . . . . . . . . . . . 4

1.5 Scope of this master thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2Background 5

2.1 Introduction ..................................... 5

2.2 What services can energy batteries provide? . . . . . . . . . . . . . . . . . . . 7

3DK2 and SE2 HPP 8

3.1 Introduction ..................................... 8

3.2 DK2powerplantdetails .............................. 9

3.3 SE2powerplantdetails............................... 10

3.4 Hybrid power plant scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3.5 REScorrelationstudy................................ 11

4HPP Case Scenario Study 16

4.1 Studycase:Scenario9 ............................... 16

xviii Contents

4.2 CorREStimeseries ................................. 17

5Bidding, Real and Full model Algorithms 19

5.1 Introduction ..................................... 19

5.2 Bidding model algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

5.3 Real time model algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

5.4 Fullmodelalgorithm ................................ 35

6Data Analysis 37

6.1 Durationcurves ................................... 37

6.2 Spot market bidding methodology . . . . . . . . . . . . . . . . . . . . . . . . . 40

6.3 Hybrid power production real time measured . . . . . . . . . . . . . . . . . . . 43

6.4 Energysurplusforecast ............................... 45

6.5 Energy curtailed forecast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

6.6 Capacity factor forecast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

7Forecast Error 48

7.1 Introduction ..................................... 48

7.2 Forecast error metrics: MAFE and RMSFE . . . . . . . . . . . . . . . . . . . . 49

8Conclusions and Future work 51

8.1 Main important aspects learnt . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

8.2 Contributions .................................... 51

8.3 Conclusion ...................................... 52

8.4 Futurework ..................................... 53

Bibliography 54

Appendix A: Matlab Code 57

Appendix B: Measured Scenarios Comparison 65

DK2HPP-Measured................................... 65

SE2HPP-Measured ................................... 67

CHAPTER 1

Introduction

1.1 Aim and motivation of the study

The aim of this master thesis is to develop an algorithm methodology to minimize the hybrid

power produced forecast error so it’s important to keep in mind that I’m not presenting an

optimization model in this work. I’ll show the dependency of the battery size used for this

purpose so using the forecast error we will reduce its variability. We will see what positive

eﬀects adding a battery within a hybrid wind and solar power plant has, its sensitivity eﬀect,

having as main advantages these:

•reduction of the curtailment produced, bigger battery size less curtailment.

•repair of the forecast error

For that purpose I’m using a 1 Day Ahead, 1DA, power production forecast for wind power

plant, WPP, and for solar photovoltaic power plant, PVPP, too. So we will see how much

storage system I’m including at the end.

This kind of forecasting is a medium term forecast type, being its key applications: scheduling,

reserve requirement, market trading and congestion management,

T. Tian and I. Chernyakhovskiy[16].

We have to take into account how every single scenario is aﬀecting in the curtailment produced

by each one. We must also take into account that the amount of curtailed energy depends

on my algorithm design decision and the amount of energy generated, when more energy is

generated then you also curtail more.

1.2 Objective of the study

To develop a methodology model for the hybrid power plant, HPP, to bid in the spot market

and then minimize the forecast error in the real time model so we can reduce the applied

associated penalties due to this error produced.

When playing with diﬀerent battery sizes we will see how it’s aﬀecting to power curtailed

drawing the important conclusion that the bigger the battery size is the better in terms of

curtailment reduction.

It goes without saying that this increase in battery size will directly incur the consequent

increase in the price of energy storage system. It is important to keep in mind that the cost of

storage continues to fall We only have to look at the storage price evolution.

2 1 Introduction

1.3 HPP with BESS design

Here in Figure 1.1 we can see how the hybrid power plant, HPP, is designed including the

battery energy storage system, BESS:

Figure 1.1: Hybrid power plant + battery energy storage system

Here in Figure 1.2 we can see the AC and DC coupled HPP with BESS designs. Being the

main advantage from AC connected HPP its ease of expansion as it enables more systems to

be added in parallel to either increase the storage power or the overall HPP capacity. Those

designs are out of this master thesis scope but if interested further relevant content can be

found at Kaushik Das et al.[6].

(a) AC connected HPP (b) DC connected HPP

Figure 1.2: Topology of connected HPP

There are mainly two HPP conﬁgurations which are is the mostly developed right one, while

the left one is an alternative version where solar converters can be removed in some situations.

1.3 HPP with BESS design 3

Figure 1.3: Types of HPPs: integration and operation of diﬀerent generating modules

(a) Global map with HPP locations and types

(b) Capacity share of wind, solar and storage tech-

nologies in 2024

Figure 1.4: World HPP and RES capacity share in 2024

All this information presented in Figures 1.3 and 1.4 has been extracted from Wind Europe

report, Renewable Hybrid Power Plants - Exploring the Beneﬁts and Market Opportunities -

July 2019 elaborated by Wind Europe.

4 1 Introduction

1.4 Advantages/values of hybrid power plants

The main advantages are cost reduction and revenue increase.

•Reduction in land cost

•Optical use of electric infrastructure and other infrastructure saves costs.

•Joint permitting process reduces risks and costs

•Shared resources reduce internal costs

•Joint site development reduces costs

•Less ﬂuctuating production increases electrical infrastructure utilization

•

Storage increases ﬂexibility and number of accessible markets (Energy market, ancillary

services market)

•Reduction of Forecast Error using storage

1.5 Scope of this master thesis

This report is divided in the following chapters:

•Chapter 1 presents introduction

•Chapter 2 includes background

•Chapter 3 presents DK2 and SE2 HPP locations

•Chapter 4 presents scenario 9 study

•Chapter 5 includes Spot Market Bidding and Operational methodologies

•Chapter 6 includes data analysis scenario 9 study

•Chapter 7 presents forecast error P_H P P _F E(t)

•Chapter 8 are drawn the main master thesis conclusions

CHAPTER 2

Background

2.1 Introduction

Figure 2.1: Danish energy system - June 29th 2020 18:51 hrs

Energinet.dk website

I’ve decided to start with this fantastic picture above, where the Danish Energy System can be

perfectly understood. It was indeed one of the ﬁrst ﬁgures commented at the very beginning

of this master thesis work in our supervisory meeting. Show us the importance of the energy

connections between neighbouring countries and their energy tradings, with import and export

energy.

Penetration rates of renewable energy sources, RES, expanded during the last years due

to increased environmental conscience and market acceptance. Another beneﬁt of combining

wind and solar photovoltaic power from the TSO’s point of view is a better utilization of grid

infrastructure and reduced congestion: improving security of energy supply and system stability

Kaushik Das et. al.[8].

The main challenge with RES is their inherent variability due to the stochastic nature of

the RES causing some diﬃculties as severe power ramps, up and down, and increased volatility

in the power generation in general. In future WPPs and PVPPs need to participate more in

the energy markets. Adding energy storage we will increase the dispatchability of RES. Wind

6 2 Background

and solar photovoltaic power don’t reliably produce power on demand, so including energy

storage system is a mitigation strategy.

Negative impact associated to stochastic nature of RES can be reduced by coupling WPPs

and PVPPs with BESS. Energy storage will be more relevant even where power prizes are

correlated negatively with renewable power production.

Battery storage is one of several technology options that can enhance power system ﬂexi-

bility and enable high levels of renewable energy integration. Studies and real-world experience

have demonstrated that interconnected power systems can safely and reliably integrate high lev-

els of renewable energy from VRE sources without new energy storage resources

Thomas Bowen

;

Ilya Chernyakhovskiy;Paul Denholm[6].

Energy storage allows RES to increase their revenue providing ﬂexibility and time shift-

ing of the power production. We have to take into account that increasing the integration of

RES also increases energy curtailment situations due to power overproduction compared to

grid connection constrain.

If we store surplus power,

Psurplus

(

t

), at times of high output we can improve the RES

production stabilization.

In order to participate in energy market, VRE sources need to reduce the uncertainty of

forecast errors. Inclusion of storage can be a viable option not only to minimize the penalties

due to forecast uncertainties but also to maximize the revenue generation

Kaushik Das et. al.

[4].

Wind and solar photovoltaic power are used in convex optimisation algorithm for making

day ahead decisions on battery operation. One day ahead optimized results are used as input

for the operating model.

Adding BESS in the WPP and PVPP we have several advantages as:

•minimize the penalties due to forecast uncertainties.

•minimize the excess energy spilt.

•maximize the revenue generation.

Including BESS we will reduce the impact of the forecast error associated to wind and solar

photovoltaic power production forecasts so it’ll reduce balancing penalty too.

It’s important to say that this master thesis study is covering the HPP but not the hy-

brid power system, HPS, as I’m not taking into account

Pdemand

(

t

)neither

Edemand

(

t

)in the

methodology development scope of this master thesis work.

2.2 What services can energy batteries provide? 7

2.2 What services can energy batteries provide?

To answer these questions there is an excellent NREL document called "Grid-Scale Battery Stor-

age Frequently Asked Questions" by Thomas Bowen;Ilya Chernyakhovskiy;Paul Denholm[5].

Including this paragraph extracted from this cited document:

["...Arbitrage: Arbitrage involves charging the battery when energy prices are low and dis-

charging during more expensive peak hours. For the BESS operator, this practice can provide a

source of income by taking advantage of electricity prices that may vary throughout the day.

One extension of the energy arbitrage service is reducing renewable energy curtailment. System

operators and project developers have an interest in using as much low-cost, emissions-free

renewable energy generation as possible; however, in systems with a growing share of VRE,

limited ﬂexibility of conventional generators and temporal mismatches between renewable energy

supply and electricity demand (e.g., excess wind generation in the middle of the night) may

require renewable generators to curtail their output. By charging the battery with low-cost

energy during periods of excess renewable generation and discharging during periods of high

demand, BESS can both reduce renewable energy curtailment and maximize the value of the

energy developers can sell to the market..."]

As we’re going to see all along this master thesis work all this is going to be fully stud-

ied and explained completely in detail. We could summarize it saying that using a really big

battery size we’re making sure that we’re reducing power curtailed but as we’ll ﬁnd included

within the models methodology development presented, it’s not that simple, and many aspects

must be taken into account to get a complete picture of it.

CHAPTER 3

DK2 and SE2 HPP

3.1 Introduction

DK2 and SE2 HPP distance is: 877.06 [Km].

Figure 3.1: Nordic area wind conditions - July 29th 2020 6.51 pm

Source: Windy.com

3.2 DK2 power plant details 9

3.2 DK2 power plant details

DK2 HPP location:

Ringsted, Ringsted Municipality, Region Zealand, 4100, Denmark. The geographic coordinates

for DK2 hybrid power plant are:

Latitude: 55o27’ 8.256" North and Longitude: 11o49’ 25.429" East

Figure 3.2: DK2 HPP location

Source: Global Wind Atlas

(a) Mean power density (b) Wind speed rose (c) Mean wind speed

Figure 3.3: DK2 HPP

Source: Global Wind Atlas

10 3 DK2 and SE2 HPP

3.3 SE2 power plant details

SE2 HPP location:

Storsjön, Östersunds kommun, Jämtland County, Region Norrland, Sweden

The geographic coordinates for SE2 hybrid power plant are:

Latitude: 63o13’ 15.594" North Longitude: 14o20’ 33.474" East

Figure 3.4: SE2 HPP location

Source: Global Wind Atlas

(a) Mean power density (b) Wind speed rose (c) Mean wind speed

Figure 3.5: SE2 HPP

Source: Global Wind Atlas

3.4 Hybrid power plant scenarios 11

3.4 Hybrid power plant scenarios

WPP+PVPP Scenario DK2 and SE2 RES Pgrid [M W ]

Scenario 1 0 WPP+100 PVPP 100

Scenario 2 25 WPP+75 PVPP 100

Scenario 3 50 WPP+50 PVPP 100

Scenario 4 75 WPP+25 PVPP 100

Scenario 5 100 WPP+0 PVPP 100

Scenario 6: Overplanting case 80 WPP+60 PVPP 100

Scenario 7: Overplanting case 100 WPP+40 PVPP 100

Scenario 8: Overplanting case 105 WPP+35 PVPP 100

Scenario 9: Overplanting case 120 WPP+20 PVPP 100

Table 3.1: DK2 and SE2 RES scenarios

The hybrid wind and solar pv power plant scenarios studied for DK2 and SE2 power plant

locations can be seen in Table 3.1 above. Where 4 overplanting scenarios have been also taken

into account for DK2 and SE2 power plant Nordic locations. We can see how in every single

scenario there is a power grid connection constraint,

Pgrid

= 100 [

MW

]which is going to be

one of the input parameters for the bidding model algorithm later presented in Chapter 5.

From all the nine scenarios studied and work all along this master thesis development I’m

presenting mainly scenario 9 in this document as I found it more interesting than the other

eight. I have also seen the eﬀect of increasing the wind ans solar pv installed capacity in every

single one of them and as I’ve been working with two Nordic power plant locations, DK2 and

SE2 the eﬀect of the solar photovoltaic has less eﬀect than it would in other locations like, for

example in Spain or France, where capacity factors for PVPP are higher than in Denmark

or Sweden. It goes without saying that the weight and generation behaviour of wind energy

production is completely diﬀerent from solar pv.

To understand the enhanced value of HPP diﬀerent studies in terms of capacity factors, vari-

ability, curtailment, value of storage etc. It should be noted that the value of HPP is enhanced

when the evacuation capacity to the grid is limited. It can be interpreted that for limited grid

connection, maximum weather resources should be used to generate maximum power close to

the evacuation capacity to maximize revenue.Kaushik Das et. al.[8]

3.5 RES correlation study

I’m presenting here a correlation study between RES power production and spot market prices.

Spot market prices have been obtained from Nordpool website:

https://www.nordpoolgroup.

com/historical-market-data/

, It can be read a complete and detailed correlation study here

Matti Koivisto et. al.[10].

12 3 DK2 and SE2 HPP

This is the block diagram of the main parts of CorRES. Simulated available power gener-

ation means the generation before, for example, curtailment:

Figure 3.6: CorRES block diagram

Matti Koivisto et. al.[10]

CorRES have provided the following time series:

•Wind power production forecast WPP_F(t) [p.u.]

•Solar pv power production forecast P V P P _F(t) [p.u.]

•Wind power production measured WPP(t) [p.u.]

•Solar pv power production measured P V P P (t) [p.u.]

These power production time series will be used as inputs for the bidding and real time models

later in

Chapter 5

. It will be then when I take into account the wind and solar capacity to get

P_RES_F(t)and P_RES(t).

3.5 RES correlation study 13

I’m presenting the study correlation from the time series obtained using

CorRES

software

for year 2016 for DK2 and SE2 wind and solar pv power plants. In this ﬁgure we can see

forecast and measured correlation studies between power production per hour vs spot market

prices. Storage is specially relevant for market where power prices are negatively correlated, as

we are going to see that happens with this two Nordic hybrid power plants.

Figure 3.7: Scatter plot: Power generated per hour vs El. Spot market prices

Here in Table 3.2 I’m including the correlation factor results for DK2 and SE2 for all the 9

scenarios:

HPP correlation coeﬃcient Forecast Measured

Scenario 1 R: 0.1155 0.1153

Scenario 2 R: 0.0217 0.0156

Scenario 3 R: -0.0945 -0.1149

Scenario 4 R: -0.1588 -0.1878

Scenario 5 R: -0.1844 -0.2142

Scenario 6 R: -0.1188 -0.1429

Scenario 7 R: -0.1528 -0.1813

Scenario 8 R: -0.1588 -0.1878

Scenario 9 R: -0.1727 -0.2025

Table 3.2: DK2 HPP correlation coeﬃcients - Power production per hour vs Spot market

prices

14 3 DK2 and SE2 HPP

HPP correlation coeﬃcient Forecast Measured

Scenario 1 R: 0.0746 0.0719

Scenario 2 R: 0.0177 0.0207

Scenario 3 R: -0.0594 -0.0579

Scenario 4 R: -0.1087 -0.1163

Scenario 5 R: -0.1288 -0.1392

Scenario 6 R: -0.0775 -0.0788

Scenario 7 R: -0.1039 -0.1106

Scenario 8 R: -0.1087 -0.1163

Scenario 9 R: -0.1196 -0.1292

Table 3.3: SE2 HPP correlation coeﬃcients - Power production per hour vs Spot market

prices

Reviewing these Tables 3.2 and 3.3 ﬁrstly we can easily see the impact of WPPs installed

capacity when is gaining more participation in the diﬀerent scenarios presented. Secondly there

is another important conclusion we can draw from these tables, and is that PVPPs participation

is relevant too, that can be seen if we compare scenario 5 with 8 and 9, where correlation

coeﬃcients, R, in absolute value are lower than correlation coeﬃcient for scenario 5.

One could say, then why we just pick scenario 3 if is having a lower

R

value? The an-

swer for that is the power generation, that is lower than in scenarios 8 and 9. And this is

precisely why I think is a good idea choosing for this study just Scenario 9.

So taking into account scenario 9 now, if we analyze scenario correlation coeﬃcient result

ﬁrstly we can observe that the correlation is negative, the electric prices are higher when the

renewable generation is lower and upside down, when the power produced is reaching a higher

value then the prices are lower, the law of supply and demand. Secondly the higher the cor-

relation factor R value is the better, in terms of R absolute value, the smaller the value the better.

Here is when the BESS becomes more relevant, in terms of improving its revenue being

able to sell the energy when prices are higher. As it allows us to move the power in time

transferring to the grid at later times when the price is higher Jose L. Crespo-Vazquez.[14].

Another correlation study between DK2 and SE2 wind and solar power plants and capacity

factors can be also found at:

Table.1 Kaushik Das et. al.

[8], where the following results are

presented in Table 3.4:

PP Location Wind Power CF [%] Solar PV Power CF [%] R

DK2 PP: 42 12 -0.1574

SE2 PP: 24 10 -0.1206

Table 3.4: Capacity factor and correlation of wind and solar power plant

3.5 RES correlation study 15

The stronger the negative correlation the better regarding e.g. utilization of the grid

connection and balanced energy output.

CHAPTER 4

HPP Case Scenario Study

4.1 Study case: Scenario 9

As commented previously in

Chapter 3

, I’m focusing my study on scenario 9, as I’ve found

it more interesting, so all the results and analysis that I’m presenting belong to the study of

this scenario. For all calculations power grid connection constraint

Pgrid

= 100 [

MW

]has been

taken into account. Here in the following Tables 4.1, 4.2, 4.3, 4.4, 4.5 and 4.6 I’m presenting

some relevant scenario 9 main ﬁgures:

Scenario 9 DK2 E_RES_F[T W h/y ear]SE2 E_RES_F[T W h/year ]

120 WPP+20 PVPP 0.471 0.303

Table 4.1: DK2 and SE2 RES E_RES_F - Scenario 9 - Forecast

Scenario 9 DK2 E_RES_F[T W h/y ear]SE2 E_RES_F[T W h/year ]

120 WPP+20 PVPP 0.468 0.281

Table 4.2: DK2 and SE2 RES E_RES - Scenario 9 - Measured

Scenario 9 DK2 CF [%] DK2 FWH [h] SE2 [%] SE2 FWH [h]

120 WPP+20 PVPP 53.75 4708.5 34.63 3033.6

Table 4.3: DK2 and SE2 RES CF and FWH - Scenario 9 - Forecast

Scenario 9 DK2 CF [%] DK2 FWH [h] SE2 [%] SE2 FWH [h]

120 WPP+20 PVPP 53.46 4683.1 32.09 2811.1

Table 4.4: DK2 and SE2 RES CF and FWH - Scenario 9 - Measured

4.2 CorRES time series 17

Scenario 9 DK2 Esurplus_F[GW h/year]SE2 Esurplus_F[GW h/y ear]

120 WPP+20 PVPP 30.403 13.273

Table 4.5: DK2 and SE2 RES E_surplus_F - Scenario 9 - Forecast

Scenario 9 DK2 Esurplus [GW h/year]SE2 Esurplus [GW h/y ear]

120 WPP+20 PVPP 25.509 8.997

Table 4.6: DK2 and SE2 RES E_surplus - Scenario 9 - Measured

4.2 CorRES time series

CorRES, Correlations in Renewable Energy Sources, is a simulation tool developed by DTU

Wind Energy department. It’s capable of simulating both wind and solar power generation.

Correlations in Renewable Energy Sources (CorRES) is a simulation tool developed at Technical

University of Denmark, Department of Wind Energy capable of simulating both wind and

solar generation. It uses a unique combination of meteorological time series and stochastic

simulations to provide consistent VRE generation and forecast error time series with temporal

resolution in the minute scale.

Such simulated VRE time series can be used in addressing the challenges posed by the

increasing share of VRE generation. These capabilities will be demonstrated through three

case studies: one about the use of large-scale VRE generation simulations in energy system

analysis, and two about the use of the simulations in power system operation, planning, and

analysis Matti Koivisto et. al.[10].

Application of VRE generation simulation tools can be used:

•in the estimation of adequacy of reserves in power systems

•in stability analysis

•long-term transmission system planning

•electricity market studies

There are diﬀerent approaches for VRE generation simulation:

•meteorological reanalysis models for generating the underlying meteorological ﬁelds

•stochastic time series models for simulating the time series of interest

18 4 HPP Case Scenario Study

To assess the value of storage for revenue maximization of a wind power plant. The results

show that proposed algorithms can increase the revenue by more than

10[%]

compared to the

operation of WPP without BESS.

CorRES outputs are:

•Forecast RES power production

•Measured RES power production

In this study case, RES are wind and solar photolvotaic.

Forecast and measured wind and solar power production time series obtained with DTU

CorRES tool are the starting point of this master thesis work. These power production time

series belong to years 2015 and 2016 for DK2 and SE2 wind and solar photovoltaic power plant

speciﬁed locations. So I’ve worked with hourly power production data for these two mentioned

years. As I’m not presenting a variability study I’m combining both years in order to work

with only one data for each scenario getting more weightage to the obtained results. I have to

say that in a very beginning stage of this work I used persistent PVPP forecast before using

CorRES forecast. Persistent forecast is a poor forecast which consist of assuming for the next

period that the power will be the same that it was in the last past one. The challenge of using

a bad forecast, as persistent is, battery will be used quite a lot, so a really big battery will be

needed.

We can read more about time series in this really interesting article: "Simulation of transconti-

nental wind and solar PV generation time series" Edgar Nuño et. al.[12].

CHAPTER 5

Bidding, Real and Full model

Algorithms

5.1 Introduction

The criteria for designing the HPP is to minimize the forecast error,

P_H P P _F E

(

t

), not

to have stable power production over the day. So reducing the forecast error means that you

will be penalized less for it, then this is the main improvement of this methodology here presented.

The purpose of including within this hybrid power plant a

BESS

is precisely to reduce

the forecast error

P_H P P _F E

(

t

)as well as to reduce curtailment including an appropiate

battery energy storage system size.

Another important factor is the amount of energy than can be charged and discharged to/from

the battery in one hour time. This is because it is not possible to get more that a limited amount

of energy supply from the battery as well as is not possible to charge the battery all the amount

we would like to charge it. This is where

P_LIMbatt

parameter becomes important as it is

the power battery limitation for charging and discharging operation. In this chapter I am pre-

senting how it is taken into account as a parameter for the bidding and real time modes analyzed.

Power surplus,

Psurplus

(

t

), is used to charge the battery too, so it will reduce forecast er-

ror too. Every single time we have

Psurplus

(

t

)then we take the opportunity to charge the

battery and when we don’t then we just bid what we have, P_HP P _F(t).

And ﬁnally when it is not possible to include all this

Psurplus

(

t

)produced within the bat-

tery then we spill this power, commonly known as

Pcurtailed

(

t

). Curtailment is spilling the

energy, letting it go as we are not able to storage it within BESS. This situation is going to

happen every single time that battery is full charged and from RES we are producing more

power than

Pgrid

then battery won’t be able to accommodate that power produced, then

Pcurtailed

(

t

)will be spilt. Just like I am showing, in this master thesis, increasing the battery

capacity, Ebatt we’ll reduce Pcurtailed (t)value.

One of the main battery parameters is the rate at which a battery is charged/discharged

relative to its maximum capacity, known as Crate.

From this study we can see and understand how important is to have a really good power

production forecast for the hybrid power plant. It will reduce the forecast error and of course

the battery size needed too. In order to participate in energy market, VRE need to reduce the

uncertainty of forecast errors. So this is why storage is a viable option: to minimize penal-

20 5 Bidding, Real and Full model Algorithms

ties due to forecast uncertainties and to maximize the revenue generation

Kaushik Das et. al.

[4].

The hybrid power production forecast

P_H P P _F

(

t

)is going to be our 1DA bid for real

time model, that is going to be one of the inputs for it. This is the main reason for the

bidding model design, as we are trading with this hybrid power production forecast not with

the P_RES_F(t)or with SOC_F(t).

5.2 Bidding model algorithm

In this model I’m working with the actual power production values provided by CorRES, that

is to say:

P_WPP_F(t) + P_P V P P _F(t) = P_RES_F(t) [M W ](5.1)

Bidding model inputs:

•Wind and solar power forecast: P_RES_F(t) [M W ]

•Power grid connection constrain: Pg rid = 100 [M W ]

•Battery capacity: Ebatt [M W h]

•Initial state of charge: SOC0_F= 0.5 [p.u.]

Parameters:

•Power battery limitation for charging and discharging P_LIMbatt [M W ]

•

Rate measure at which a battery is charged/discharged relative to its maximum capacity:

Crate [h]

Assumption:

•Sampling time resolution: 1 [hour]

Bidding model outputs:

•State of charge forecast: SOC_F(t) [p.u.]

•Hybrid power produced forecast: P_H P P _F(t) [M W ]

•Power curtailed forecast: Pcur tailed_F(t) [M W ]

5.2 Bidding model algorithm 21

5.2.1 Bidding model ﬂowchart

Here I’m presenting how bidding model works, where complete methodology can be seen in

the next three ﬁgures. First one, Figure 5.1 is a complete overview showing how it works,

meanwhile Figure 5.2 and Figure 5.3 are showing in detail charging and discharging procedures.

Figure 5.1: Bidding model algorithm ﬂowchart

Figure 5.2: Bidding model algorithm ﬂowchart - Charging detail

22 5 Bidding, Real and Full model Algorithms

Figure 5.3: Bidding model algorithm ﬂowchart - Discharging detail

The dispatch SOC_F(t) for this preliminary bidding model the following equations have

been taken into account:

P_SOC headroom_F(t) = (1 −SOC _F(t−1)) ·Ebatt [M W ](5.2)

Pcharge _F(t) = minimum(Psurplus _F(t), P _SOC headroom_F(t), P _LIMbatt) [M W ]

(5.3)

Pdischarge max_F(t)=(SOC_F(t−1) −0.5) ·Ebatt [M W ](5.4)

SOC_F(t) = SOC_F(t−1) + (Pcharge_F(t)−Pdischar ge_F(t))/Ebatt [M W ](5.5)

Pgrid headroom(t) = minimum(Pgrid −P_RES_F(t),0) [M W ](5.6)

Pdischarge _F(t) = minimum(Pdischarge max_F(t), Pgrid headroom_F(t), P _LIMbatt) [M W ]

(5.7)

It has helped myself to clearly better understand how the bidding model is working and how it

is providing the P_HP P _F(t).

In order to provide more consistency between bidding and real time models this is the deﬁnitive

bidding model implemented, it has already checked and is providing the same results as previous

bidding model version too. And ﬁnally here in Figure 5.4 is the bidding model algorithm which

has also been included within the Matlab code attached in Appendix A:

5.2 Bidding model algorithm 23

Figure 5.4: Bidding model algorithm

We can see here how the SOC_F(t) dispatch is working properly, I’ve included two dynamic

limiters that rules the battery operation completely, letting the power produced getting in when

it’s possible to meet the model design conditions for battery charging and discharging.

To get that battery capacity will always be above the 50 [%] of its capacity is crucial to

set the SOC_F(t)limits have been set correctly, that is to say, between: [0.5 : 1] [p.u.]

Another important thing done is to check in every single moment that we are working with the

same units and not mixing them, which of course, needless to say would aﬀect badly to the

model methodology designed and completely to the results obtained. As it’s commonly said

being sure that we’re not mixing apples with bananas is extremely important here.

5.2.2 Bidding Model Study Cases

I’ve distinguished 2 diﬀerent cases to be studied and included within the bidding algorithm and

Matlab code presented in Appendix A, which are:

1st operational condition:

P_RES_F(t)> Pgr id [M W ](5.8)

then we have:

Psurplus_F(t) [M W ](5.9)

2nd operational condition:

P_RES_F(t)<=Pgr id [M W ](5.10)

then we have:

Psurplus_F(t) = 0 [M W ](5.11)

In Table 4.5 have been already presented

E_RES_F

and

Esurplus_F

values for scenario 9 in

DK2 and SE2 hybrid power plants.

24 5 Bidding, Real and Full model Algorithms

5.2.3 Bidding model algorithm behaviour

It’s really important to be aware that battery charging and discharging can not happen at the

same time, so if you charge then you don’t charge and vice versa. Basically we will charge

when we will have available

Psurplus_F

(

t

)due to RES overproduction and will discharge when

we won’t have it, of course we’ll be able to do it having energy stored previously taking into

account SOCF(t)limitation that must always be [0.5 [p.u.] : 1 [p.u.]].

When do I charge the battery?

Every single time I have

Psurplus_F

(

t

)and it’s room in the battery for it, otherwise power

produced will be curtailed partially or entirely depending on the available room for energy

storing. It is important to take into account that

P_LIMbatt

will limit the charge and discharge

operation. Taking a look to Figure 5.4 we can see that we’ll charge when we’re not discharging,

in other words, when

Pbatt

is negative then is meeting the limiter with down limit zero, so it’s

not discharging.

When do I discharge the battery?

Pdischarge _F

(

t

)is calculated as we want to discharge what it takes to get. As we have

just seen it’s important to take into account +

/−Pbatt_max_F

limit values that will limit the

charge and discharge operation. Eventually I’m taking the minimum value just to really know

what power is discharged and supplied to grid. Ultimately in the next adder

Pdischarge _F

(

t

)

is compared to

Pdischarge _F_req

(

t

), fruit of both is

P_missing_F

(

t

)that is aggregated to

Pgrid , then we have obtained our bid for real time model, P_H P P _F(t).

What is Pcurtailment_F_total(t)?

Power curtailment is the power spilt because there is no room for it within the battery,

this power produced is not going into the grid. It could be used for selling it later. We have to

take into account a really important aspect, curtailment power is can come from two places

within the Figure 5.4, which are both limiters, when signal is overlimits. So to calculate

Pcurtailment_F_total(t)we have to apply the following formula:

Pcurtailment_F_total(t) = Pcurtailment _F(t) + P0

curtailment_F(t) [M W ](5.12)

5.2 Bidding model algorithm 25

5.2.4 Bidding model results

Bidding model results are widely presented in Chapter 6, but here we can see a quick plotting

comparison for all the forecast time series taking into account 2015 and 2016 years, for several

P_LIMbatt parameter values: 5, 10, 50 and 100 [MW].

(a) P_LIMbatt = 5MW (b) P_LIMbatt = 10MW

Figure 5.5: DK2 Bidding - Scenario 9

(a) P_LIMbatt = 50MW (b) P_LIMbatt = 100MW

Figure 5.6: DK2 Bidding - Scenario 9

To facilitate its better understanding what I’m going to present now is only two days

detailed study so we’ll be able to fully understand the forecast results provided. Firstly we will

see a clear case of charging operation as we have plenty of Psurplus_F(t).

I have chosen two consecutive days, June 17

th

and 18

th

from the power production time

series provided by CorRES, where is a clear example of Psurplus(t)existence.

I’m presenting a complete study including diﬀerent

P_LIMbatt

values: 5, 10, and 50 [MW]

and Crate = 1 [hour].

26 5 Bidding, Real and Full model Algorithms

(a) P_LIMbatt = 5MW (b) P_LIMbatt = 10MW

Figure 5.7: DK2 forecast - Scenario 9

Figure 5.8: DK2 forecast - Scenario 9 - P_LIMbatt = 50MW

Reviewing Figures 5.7 and 5.8 we can check how

Pcurtailed_F

(

t

),

SOC_F

(

t

)and

Pdischarge _F

(

t

)

is varying according to the

P_LIMbatt

parameter value taken in every single case. I’m present-

ing here in this work speciﬁcally in Chapter 6 a complete data analysis using the duration curves.

Secondly now we will see a clear case of discharging operation providing its energy to the grid

as we don’t have much of

Psurplus_F

(

t

)produced this time. I have chosen for showing this two

consecutive days in June 2015, June 8th and 9th for DK2 HPP.

5.2 Bidding model algorithm 27

(a) P_LIMbatt = 5MW (b) P_LIMbatt = 10MW

Figure 5.9: DK2 forecast - Scenario 9

Figure 5.10: DK2 forecast - Scenario 9 - P_LIMbatt = 50MW

28 5 Bidding, Real and Full model Algorithms

Figure 5.11: DK2 forecast - Scenario 9 - P_LIMbatt = 100MW

In these Figures 5.9, 5.10 and 5.11 we can check how

Pcurtailed_F

(

t

),

SOC_F

(

t

)and

Pdischarge _F

(

t

)is varying according to the

P_LIMbatt

value taken in every single case. In this

second case as advanced can be clearly seen how low is

Psurplus_F

(

t

)compare to the ﬁrst case

above presented. It can be also seen clearly the power generation intermittency of RES.

5.3 Real time model algorithm 29

5.3 Real time model algorithm

In this model I’m working with the actual power production values provided by CorRES, that

is to say:

P_WPP(t) + P_P V P P (t) = P_RES(t) [M W ](5.13)

taking into account the WPP and PVPP capacity installed, as CorRES is providing time series

in per units, so for scenario 9. Another input for this real time model will be hybrid power

production forecast,

P_H P P _F

(

t

), output from the previous model algorithm, bidding model.

Real time model output is the

P_H P P

(

t

), so after obtaining this value we are able to calcu-

late the

P_H P P _F E

(

t

)and quantify how good our hybrid power production forecast has been.

As we have also seen within the bidding model that the battery charging and discharging

operation can not happen both at the same time. So if you charge then you don’t charge and

vice versa. So basically we will charge when we’ll have

Psurplus

(

t

)and will discharge when we

won’t and, of course, having stored energy previously within the BESS to be provided to the grid.

It’s really important to take into account that our bid is:

P_H P P _F

(

t

), 1 DA, so it means

that we are saying that our forecast obtained the previous day, is what is going to happen in

the actual day, that could be fully understood checking the real time model ﬂowchart presented

in this ﬁfth chapter, point 5.3.2.

Once we have run both models, bidding and real time model, then hybrid power produced

forecast error, P_H P P _F E(t), can be calculated as:

P_H P P _F E(t) = P_H P P (t)−P_HP P _F(t) [M W ](5.14)

5.3.1 Inputs, outputs, parameters and assumptions

Real time model inputs:

•Wind and solar power production measured: PRES (t) [M W ]

•Power grid connection constrain: Pg rid = 100 [M W ]

•Hybrid power production forecast: PH P P _F[M W ]

•Battery capacity: Ebatt [M W h]

•Initial state of charge: SOC0[p.u.]

Parameters:

•P_LIMbatt [M W ]

•Crate [h]

30 5 Bidding, Real and Full model Algorithms

Assumption:

•Sampling time resolution: 1 [hour]

Real time model outputs:

•State of charge: SOC(t) [p.u]

•Hybrid power produced measured: P_H P P (t) [M W ]

•Power curtailed: Pcur tailed_total(t) [M W ]

5.3.2 Real time model ﬂowchart

Figure 5.12: Real time model algorithm ﬂowchart

5.3.3 Real Time Model Study cases

I’ve distinguished 3 diﬀerent cases to be studied and included within the real time algorithm

and Matlab code, which are:

1st operational condition:

P_RES(t)> P _H P P _F(t) [M W ](5.15)

then we have:

Pbatt(t)_req < 0 [M W ](5.16)

2nd operational condition:

P_RES(t)< P _H P P _F(t) [M W ](5.17)

then we have:

Pbatt(t)_req > 0 [M W ](5.18)

5.3 Real time model algorithm 31

3rd operational condition:

P_RES(t) = P_H P P _F(t) [M W ](5.19)

then we have:

Pbatt(t)_req = 0 [M W ](5.20)

So we are calculating Pbatt(t)_req as:

Pbatt(t)_req =P_HP P _F(t)−P_RES(t) [M W ](5.21)

and taking into account that the corresponding operational condition

Pbatt

(

t

)

_req

will be

positive, negative or zero.

Our priority is to minimize the hybrid power plant forecast error,

P_H P P _F E

(

t

). So

this time we are not keeping the 50 [%] of the battery capacity all the time as I have presented

in bidding model algorithm. As soon as we get power we will try to provide it to the grid if it’s

possible. I’m remarking again that it is not possible to charge and discharge at the same time,

this has been also taken into account in the real time model algorithm.

5.3.4 Real time model algorithm behaviour

It is really important to be aware that battery charging and discharging can not happen at the

same time, as happened with previous model presented, bidding model. So if you charges then

you can’t charge and vice versa.

When do I charge the battery?

Every single time I have

Psurplus

(

t

)and it’s room in the battery for it, otherwise it will be

curtailed. It is important to take into account

P_LIMbatt

that will limit the charge and

discharge operation. Taking a look to Figure 5.12 we can see that we’ll charge when we are not

discharging, in other words, when

Pbatt

is negative then is meeting the limiter with down limit

with zero value, so it’s not discharging.

When do I discharge the battery?

Pdischarge

(

t

)is calculated as we want to discharge what it takes to get. As we have just seen it’s

important to take into account +

/−Pbatt_max

limits that will limit the charge and discharge

operation. I’m taking the minimum value just to really know what power is discharged and

supplied to grid. Finally in the next adder

Pdischarge

(

t

)is compared to

Pdischarge _req

(

t

), fruit

of both is

P_missing

(

t

)that is aggregated to

P_H P P _F

(

t

), our bid from bidding model,

then we have obtained our bid for real time model, P_H P P (t).

What is Pcurtailment_total(t)?

It’s the power spilt because there is no room for it within the battery, this power produced is

not going into the grid. It could be used for selling it later. We have to take into account a

really important aspect, curtailment power is can come from two places within the Figure 5.12,

which are both limiters, when signal is overlimits. So to calculate

Pcurtailment_total

(

t

)we have

to apply the following formula:

Pcurtailment_total(t) = Pcurtailment (t) + P0

curtailment(t) [M W ](5.22)

32 5 Bidding, Real and Full model Algorithms

5.3.5 Real time model results

Measured results are widely presented in Chapter 6, but here we can see a quick comparison

plotting for all the forecast time series taking into account 2015 and 2016 years, for several

P_LIMbatt values.

Figure 5.13: DK2 measured - Scenario 9 - P_LIMbatt = 5MW

Figure 5.14: DK2 measured - Scenario 9 - P_LIMbatt = 100MW

5.3 Real time model algorithm 33

To facilitate its understanding what we are going to do now is only a two days detailed

study so we’ll be able to fully understand the forecast results provided.

I have chosen two consecutive days, June 17

th

and 18

th

2016 from the time series provided,

where is a clear example of Psurplus(t)existence.

I’m presenting a complete study including diﬀerent

P_LIMbatt

values: 5, 10, 50 and 100

[MW] and Crate = 1 [hour].

(a) P_LIMbatt = 5MW (b) P_LIMbatt = 10MW

Figure 5.15: DK2 measured - Scenario 9 -

P_LIMbatt

= 5

MW

and

P_LIMbatt

= 10

MW

Figure 5.16: DK2 measured - Scenario 9 - P_LIMbatt = 50MW

In these ﬁgures above we can check how

Pcurtailed

(

t

),

SOC

(

t

)and

Pdischarge

(

t

)are varying

according to the

P_LIMbatt

value taken in every single case. I’m presenting here in this work

34 5 Bidding, Real and Full model Algorithms

speciﬁcally in Chapter 6 a complete data analysis using the duration curves.

Secondly we will see a clear case of discharging operation providing its energy to the grid as we

don’t have much of

Psurplus

(

t

)power produced this time. I have chosen for showing this two

consecutive days in June 2015, June 8th and 9th for DK2 HPP.

(a) P_LIMbatt = 5MW (b) P_LIMbatt = 10MW

Figure 5.17: Days with low Psurplus(t)−Pbatt = 5 and 10 [M W ]

(a) P_LIMbatt = 50MW (b) P_LIMbatt = 100MW

Figure 5.18: Days with low Psurplus(t)−Pbatt = 50 and 100 [M W ]

Finally in these Figures 5.17 and 5.18 we can see how

Pcurtailed

(

t

),

SOC

(

t

)and

Pdischarge

(

t

)

are varying according to the

P_LIMbatt

value taken in every single case. In this second case as

advanced can be clearly seen how low is

Psurplus

(

t

)compare to the ﬁrst case above presented.

It can be seen also clearly the intermittency of RES power generation again, as it has already

happen with the forecast too.

5.4 Full model algorithm 35

5.4 Full model algorithm

Here full model algorithm is presented, where forecast and real time models are integrated

together, taking into account the time execution sequence for both.

Full model inputs:

•Wind and solar time series: WPP_F(t),P V P P _F(t),WPP(t)and P V P P (t) [p.u.]

•

Wind and solar capacities

P_WPP_F

,

P_P V P P _F

,

P_WPP

and

P_P V P P

[

MW

]

Parameters:

•Wind power capacity P_WPP [MW ]

•Solar pv power capacity P_P V P P [M W ]

Assumption:

•Time resolution: 1 [hour]

Full model outputs:

•HPP power produced: P_H P P (t) [M W ]

•Hybrid power plant forecast error: P_H P P _F E(t) [M W ]

5.4.1 Full model ﬂowchart

Figure 5.19: Full model algorithm ﬂowchart

36 5 Bidding, Real and Full model Algorithms

5.4.2 Full model algorithm behaviour

CorRES software is providing us the wind and solar production time series, for bidding and for

real time models, in [p.u.]. As we’re having diﬀerent wind and solar capacities for every single

one of the scenarios, e.g. for scenario 9, having as results the following:

P_RES_F(t)=(WPP_F(t)) ·P_WPP + (P V P P _F(t)) ·P_P V P P [M W ](5.23)

P_RES(t)=(WPP(t)) ·P_WPP + (P V P P (t)) ·P_P V P P [M W ](5.24)

Then, ﬁrstly

P_RES_F

(

t

)is taken into account within the bidding algorithm meanwhile

P_RES(t)will be entering within the real time model.

Secondly once we have already obtained

P_H P P _F

(

t

), as bidding model output, that will be

one of the inputs for the real time model together with the previously mentioned

P_RES

(

t

).

Finally we’re obtaining the full model outputs, which are:

P_H P P

(

t

)supplied to the grid and

P_H P P _F E

(

t

). As we have seen during this master thesis hybrid power plant forecast error

will tell us how good the forecast, which was our 1 day ahead bid. All the hybrid power plant

forecast error metrics calculated are provided in detail in Chapter 7

CHAPTER 6

Data Analysis

6.1 Duration curves

Duration curves is another way of plotting the ramp rates, which haven’t been used here,

but this time the results are plotted in a sorted manner, e.g. based on its value and not

chronologically. Then we are able to check for a certain number of hours how much the power

variation is. The maximum and minimum values are exactly the same as in the time series

plotting as I haven’t changed anything about how I’m calculating from time series plotting

method to this one. What the duration curve is telling us is for how long we can expect what

kind of changes, so we can see the power variability.

Cumulative distribution function:

A cumulative frequency distribution is the sum of the

class and all classes below it in a frequency distribution which means that is the sum of the

events when the signal has a certain value. So in cumulative distribution function we are

counting how many times the signal has every single value. It can be also done for diﬀerent

bins or changes in the power production.

What I’m including here in the ﬁrst 2 ﬁgures, Figures 6.1a and 6.1b, following below is

the main relevant aspects for the forecast as:

•P_RES_F(t)

•Psurplus_F(t)

•Pcurtailed_F(t)

•Pcharge _F(t)

•Pdischarge _F(t)

•SOC_F(t)·Ebatt

•P_H P P _F(t)

I’ve changed the values of

P_LIMbatt

parameter taking: 5, and 50 [

MW

]values maintaining

E_batt size to 100 [

M W h

], these ﬁgures have been plotted using the bidding model presented

in Chapter 5, which are:

38 6 Data Analysis

(a) P_LIMbatt = 5MW (b) P_LIMbatt = 50MW

Figure 6.1: DK2 Forecast duration curves

Some of the observation that I can get from these ﬁgures reviewing is how is reduced

Pcurtailed_F

(

t

)when

P_LIMbatt

is increased from 5 [

MW

]to 50 [

MW

]. As it was expected

and have been explained in this master thesis work one of the two main advantages of using

energy storage is minimizing the curtailment, the bigger the better in terms of reducing power

curtailed.

Secondly, another important aspect of plotting these duration curves is to check

SOC_F

(

t

),

then we can see that takes a little more time to get the 100 [%] of its capacity when

P_LIMbatt

value is 50 [

MW

]. It is understandable as battery size will be bigger too. Also it can be seen how

the

SOC_F

(

t

)is behaving as expected in this bidding model due to the design methodology,

maintaining 50 [%] as minimum value of its capacity all the time.

Finally we can see what happens to

P_H P P _F

(

t

)in both cases, where for a higher

P_LIMbatt

value and a bigger battery size we can provide during more time the power grid connection

constrain value,

Pgrid

= 100 [

MW

]. This would be the main conclusion, the bigger the better,

in terms of battery size, as we can provide and comply with

Pgrid

during more time and this

exactly one of the actions we want to obtain from this model. It is important to remark that

bidding model presented is not the optimized model. I’m sure that can be presented other

model versions which better optimize their results, but what it’s for sure is that the bidding

model presented is working with consistency and reliability providing good results as we can

see.

Now I’m doing the same for the real time model, the next two ﬁgures are plotted using it

and the main relevant aspects are:

•P_RES(t)

•Psurplus(t)

•Pcurtailed(t)

•Pcharge (t)

•Pdischarge (t)

6.1 Duration curves 39

•SOC(t)·Ebatt

•P_H P P (t)

(a) P_LIMbatt = 5MW (b) P_LIMbatt = 50MW

Figure 6.2: DK2 Measured duration curves

Some of the observations that I can get from these ﬁgures reviewing is how is reduced

Pcurtailed_F

(

t

)when

P_LIMbatt

is increased from 5 [

MW

]to 10 [

MW

]. Here we can also

see another great diﬀerence compared to bidding model, which is that for lower

P_LIMbatt

values curtailment is really high, increasing dramatically 20 [%] of the time. It’ seen also that

curtailment is happening approximately 50 [%] of the time.

If we increase

P_LIMbatt

for a same

Ebatt

value we can easily see how the direct conse-

quences of doing it are. There are mainly two: we’ll reduce dramatically curtailment till just

having less than 5 [%] of the time and these curtailment values will be really high. Here we

can see another big diﬀerence compared to bidding model, and this is due to our main model

strategy which is minimizing the forecast error, P_HP P _F E (t).

Again as expected one of the two main advantages of using energy storage is minimizing

the curtailment, the bigger the better but for real time model we can see that this also aﬀects

to the high value of it when we increase P_LIMbatt values.

Here, in the real time model, it’s important to say that the design methodology for the

SOC_F

(

t

)is diﬀerent to the previous model, it can be easily seen when checking the limits

of the dynamic limiters included within the real time model in the SOC dispatch. Secondly,

another important aspect of plotting these duration curves is to check

SOC_F

(

t

), then we can

see that takes more time to get the 100 [%] of its storing capacity when

P_LIMbatt

= 50 [

MW

],

more than it does with bidding model too. Thirdly, battery is remaining completely empty more

than 35 [%] of the time for lower

P_LIMbatt

values, 25[%] of the time for higher

P_LIMbatt

values as it happens with P_LIMbatt = 50 [M W ].

Again we can see what happens to

P_H P P

(

t

)in both cases, for a higher

P_LIMbatt

value,

we can provide during more time the power grid connection constrain value,

Pgrid

= 100 [

MW

].

This would be the main conclusion, the bigger the better as we can provide and comply with

the

Pgrid

during more time and this exactly one of the actions we want to obtain from this

40 6 Data Analysis

model. It is important to remark that real time model presented is not the optimized model

neither. I’m sure that can be presented other model versions which better optimize the results,

but what it’s for sure is that this real time model presented is working with consistency and

reliability providing good results, as we can see.

Finally if we take into account bidding and real time model ﬁgures we can conclude that

we are reducing the forecast error of the hybrid power plant,

P_H P P _F E

(

t

). This is going

to be studied in detail in the next chapter, Chapter 7 where I’m presenting the forecast error

metric calculated and their results.

6.2 Spot market bidding methodology

I have taken into account several diﬀerent battery limits,

P_LIMbatt

[

MW

]in both HPP loca-

tions for this scenario 9 detailed study. I have also taken into account that

Ebatt

= 100 [

M W h

]

and Crate = 1 [hour].

Changing

P_LIMbatt

because this parameter will aﬀect to the forecast

Pcharge _F

(

t

),

Pdischarge _F

(

t

),

Pcurtailed_F(t), and of course P_H P P _F(t)too.

6.2.1 Hybrid power production forecast

Figure 6.3: DK2 P_H P P _F(t)-P_LIMbatt = 5[M W ]Ebatt = 100[M W h]

6.2 Spot market bidding methodology 41

Figure 6.4: DK2 P_H P P _F(t)-P_LIMbatt = 25[M W ]Ebatt = 100[M W h]

Figure 6.5: DK2 P_H P P _F(t)-P_LIMbatt = 50[M W ]Ebatt = 100[M W h]

42 6 Data Analysis

Figure 6.6: DK2 P_H P P _F(t)-P_LIMbatt = 75[M W ]Ebatt = 100[M W h]

In these four Figures above, 6.3, 6.4, 6.5 and 6.6, we can clearly observe two main things: Firstly

SOC_F

(

t

)

·Ebatt

which is reaching full charge just 15 [%] of the time for

P_LIMbatt

= 5 [

MW

]

which is only 12

.

5 [%] approximately, here we can see that in terms of battery full capacity

charged we need to increase

P_LIMbatt

values otherwise most of the time we’ll be below full

charge situation.

Secondly, how

Pcurtailed

(

t

)is behaving when I’m increasing

P_LIMbatt

from 5 [

MW

]till

75 [

MW

], also it can be seen that there is no change detected from 50 to 75 [

MW

], having the

same

Pcurtailed

(

t

)value. In this case I would say that there is no justiﬁed reason for increasing

P_LIMbatt

parameter value, but we have to take into account that the other main reason of

including BESS is minimizing forecast error. So we’ll need to check ﬁrst about it and after to

make the right decision looking at the whole picture.

Finally we can see how is

P_H P P _F

(

t

)just reaching

Pgrid

between the: [25% : 27

.

5%]

of the time depending on the

P_LIMbatt

value taken. Here you can see another diﬀerence with

the real time model where

P_H P P

(

t

)can’t manage to reach

Pgrid

during as much time as

bidding model does.

6.3 Hybrid power production real time measured 43

6.3 Hybrid power production real time measured

Figure 6.7: DK2 P_H P P (t)-P_LIMbatt = 5[M W ]Ebatt = 100[M W h]

Figure 6.8: DK2 P_H P P (t)-P_LIMbatt = 25[M W ]Ebatt = 100[M W h]

44 6 Data Analysis

Figure 6.9: DK2 P_H P P (t)-P_LIMbatt = 50[M W ]Ebatt = 100[M W h]

Figure 6.10: DK2 P_H P P (t)-P_LIMbatt = 75[M W ]Ebatt = 100[M W h]

As it has happened in the last point these four Figures above, 6.7, 6.8, 6.9 and 6.10, are showing

mainly two aspects: Firstly

SOC

(

t

)

·Ebatt

which is reaching full charge just [57

.

5%] of the time

less for

P_LIMbatt

= 5 [

MW

]which is only [67

.

5%] approximately, here we can see that in

6.4 Energy surplus forecast 45

terms of battery full capacity charged we need to increase

P_LIMbatt

values otherwise most of

the time we’ll be below full charge situation. If we compare to forecast ﬁgures from last point

there is another aspect, the behaviour of

SOC_F

(

t

)and

SOC

(

t

)is diﬀerent, and this because

bidding and real time model have diﬀerent limits within the second limiter placed in SOC

dispatch. Bidding model is keeping all the time this established range between: [0

.

5 : 1] [

p.u.

]

meanwhile real time model isn’t.

Secondly, how

Pcurtailed

(

t

)is behaving when increasing

P_LIMbatt

from 5 [

MW

]till 75 [

MW

].

But here it is a great diﬀerence with last point as I have detected a diﬀerent behaviour

Pcurtailed

(

t

). This is due to the real time methodology algorithm design which is slightly

diﬀerent as it can be seen in Chapter 5.

Finally we can see how is

P_H P P

(

t

)just reaching

Pgrid

between the: [17

.

5% : 21%] of

the time depending on the

P_LIMbatt

value taken. Here you can see another diﬀerence with

the bidding model where P_H P P _F(t)manage to reach Pgrid during a little more time.

6.4 Energy surplus forecast

Pgrid connection constrain is constant, its power value is: 100[M W ].

Psurplus(t)_Fis deﬁned as:

P_RES_F−Pgr id [M W ](6.1)

when:

P_RES_F > Pgrid [M W ](6.2)

So you can obtain Esurplus(t)_Fjust summing up its Psurplus(t)_Ffor all the time series.

DK2 Esurplus_F(t)

Esurplus_F(t) = 30.404 [GW h/year](6.3)

SE2 Esurplus_F(t)

Esurplus_F(t) = 13.273 [GW h/year](6.4)

6.5 Energy curtailed forecast

Pcurtailed

(

t

), measured and forecasted, is deﬁned as the power spilt that you can storage within

the battery because it’s fully loaded. As I have said before one of the main eﬀects of including a

BESS is reducing

Pcurtailed

(

t

), storaging all the energy we can within the battery and supplying

it later to the grid.

So you can obtain

Ecurtailed

(

t

)

_F

just summing up its

Pcurtailed

(

t

)

_F

for all the time se-

ries. This is one of the outputs of the spot market bidding model algorithm.

46 6 Data Analysis

6.5.1 DK2 Ecurtailed_F(t)

Ecurtailed_F

(

t

)will vary from one battery size chosen to other, also as we can see here in the

table below, will be

Crate

dependent. Concluding the much bigger battery size we include

within HPP the better as we will reduce

Pcurtailed

(

t

)and

Ecurtailed

(

t

)so hybrid power plant

forecast error, P_H P P _F E(t).

We are going to see here some results obtained with the bidding model presented in

Chapter 7

using Matlab code included in the

Appendix A

. It has been taken into account

Ebatt

= 100 [

MW

]

and Crate = 1 [hour].

P_LIMbatt [MW] Ecurtailed(t) [GW h/year]

10 46.797

20 6.685

30 0.236

40 0.236

50 0.236

Table 6.1: DK2 Ecurtailed _F(t) [GW h/year]

So we can see how increasing 5 times P_LIMbatt we can highly reduce Ecurtailed_F(t).

6.6 Capacity factor forecast 47

6.5.2 SE2 Ecurtailed_F(t)

P_LIMbatt [MWh] Ecurtailed(t) [GW h/year]

10 20.102

20 5.834

30 0.095

40 0.095

50 0.095

Table 6.2: SE2 Ecurtailed _F(t) [GW h/year]

So we can see how increasing 5 times P_LIMbatt we can highly reduce Ecurtailed_F(t).

6.6 Capacity factor forecast

DK2 C F _F(t)Capacity factor forecast for DK2 HPP is:

DK2C F _F= 53.75 [%]

and full work hours forecast for DK2 HPP is:

DK2FWH_F= 4708.50 [hours]

SE2 C F _F(t)

Capacity factor forecast for SE2 HPP is:

SE2C F _F= 34.62 [%]

and full work hours forecast for SE2 HPP is:

SE2FWH_F= 3032.12 [hours]

CHAPTER 7

Forecast Error

7.1 Introduction

We have to say that when we’re talking about minimizing the forecast error, as it’s our main

criteria and strategy followed in the presented methodology included in

Chapter 5

, we have to

take into account the two dimensions of it, the power and the energy.

If we’re talking about power dimension, using the duration curves then what we can see

is having one determined battery size how much power I’ll be able to minimize in terms of the

time, in [MW] units.

On the other side if we take into account the developed algorithm then we can talk about the

energy side, the state of charge, SOC (t)in [MWh] units.

It’s already known that a storage unit consists in two parts:

•electrical infrastructure mainly with the converter, [MW ]

•battery bank, which is placed behind, [M W h]

The battery bank, energy side, is the most critical, which means how much battery and how

many battery hours we have. Both, power and energy side are duable, but second is more

costly. In this master thesis I’m not going through a cost study of the battery as it is out of

the scope but it is important to know that it has changed dramatically during the last ten

years, nowadays it’s about

P ricebattery bank

= 230 [$

/M W h

]. What we have seen during the last

decade is that this battery energy price for battery bank has been decreasing dramatically and

it’s expected that this will happen even more. It should not be forgotten in cost calculations

the battery converters price either.

So it will be a compromise between reducing the hybrid power plant forecast error,

P_H P P _F E

(

t

)

and the overcost added buying more battery.

We don’t have to forget, as commented previously in this master thesis, we’re using this

battery to utilize the curtailed power, that would also mean we have a little value of reduction

of cost of energy, CHP P [e/M W h].

In this master thesis a complete study taking into account diﬀerent energy capacities is

presented.

7.2 Forecast error metrics: MAFE and RMSFE 49

7.2 Forecast error metrics: MAFE and RMSFE

With the forecast error metrics we can see easily how relevant the eﬀects of including an

BESS

within the wind and solar photovoltaic power plant are.

Forecast error metrics formulas:

Mean Absolute Forecast Error:

M AF E =1

n·

n

X

n=1

|(yj−ˆyj)|(7.1)

Root Mean Square Forecast Error:

RM SF E =v

u

t

1

n·

n

X

n=1

(yj−ˆyj)2(7.2)

and this equation must always be fulﬁlled:

RM SF E ≥M AF E (7.3)

where:

•yiis measured hybrid power production P_HP P (t) [M W ]

•ˆyjis forecast hybrid power production P_HP P _F(t) [M W ]

being the hybrid power plant forecast error, P_HP P _F E (t), which can be calculated:

P_H P P _F E(t) = P_H P P (t)−P_HP P _F(t)(7.4)

being:

Pgrid

= 100 [

MW

],

Ebatt

= 100 [

M W h

]and

Crate

= 1 [

hour

]. Previously using the time

series provided by CorRES and taking into account the wind and solar capacity installed for

scenario 9, I’ve calculated MAE and RMSE for P_RES_F E(t), where this is:

P_RES_F E (t) = P_RES(t)−P_RES_F(t)(7.5)

having the following results for DK2 and SE2 RES locations:

RES MAE_H P P _RES(t)[M W ]RMSE_RES_F E(t)[M W ]

DK2 18.1 26.5

SE2 14.9 22.9

Table 7.1: DK2 and SE2 Scenario 9

P_RES_F E

(

t

):

MAE_RES_F E

(

t

)and

RMSE_HP P _F E(t)

Knowing this two forecast error metrics then now I’m presenting how the inclusion of a

BESS aﬀects to the reduction of the P_H P P _F E(t)for DK2 HPP:

50 7 Forecast Error

P_LIMbatt [M W ]MAE_H P P _F E(t)[M W ]RM SE_H P P _F E(t)[M W ]

5 7.0 16.3

25 6.7 16.6

50 6.7 16.8

75 6.7 16.8

Table 7.2: DK2 Scenario 9

P_H P P _F E

(

t

):

MAE_H P P _F E

(

t

)and

RMSE_HP P _F E(t)

And for SE2 HPP:

P_LIMbatt [M W ]MAE_H P P _F E(t)[M W ]RM SE_H P P _F E(t)[M W ]

5 7.0 16.0

25 6.7 16.3

50 6.7 16.5

75 6.7 16.5

Table 7.3: SE2 Scenario 9

P_H P P _F E

(

t

):

MAE_H P P _F E

(

t

)and

RMSE_HP P _F E(t)

We can easily observe that including BESS has a real eﬀect on the reduction of the forecast

error, between: [2

.

5 : 2

.

7] times for DK2 HPP and [2

.

1 : 2

.

21] times for SE2 HPP. Another

interesting conclusion that we can draw from these results is that we will have more success

when we have better power production. With the ﬁgures I’ve calculated it can be seen that we

can reach 0.5 points of diﬀerence if we look at the reduction range.

Another important conclusion can be drawn from the results and is that as much bigger

value of

P_LIMbatt

the better for reducing forecast error. Please remember that I have used

for this calculation an Ebatt = 100 [M W ].

Finally answering to the question How much can we reduce forecast error adding BESS

in Scen. 9? here it’s the answer and results:

P_LIMbatt [M W ]DK2P_HP P _F E(t)reduction[%] SE2P_H P P _F E(t)reduction[%]

5 61.28 53.11

50 63.12 54.82

Table 7.4: DK2 and SE2 Scenario 9 P_H P P _F E(t)reduction[%]

CHAPTER 8

Conclusions and Future work

8.1 Main important aspects learnt

During the master thesis work all along this ﬁve months period I’ve learnt the following aspects:

•Relevance of battery energy storage system inclusion in wind and solar power plants.

•Study of the proper strategy to be applied to get a forecast error reduction.

•Developing of model methodology: bidding and real time algorithms.

•

The importance of the optimization of both algorithms if we want to have success reducing

penalties.

•To work with forecast and real time data provided by CorRES.

•

The eﬀect of choosing a good forecast to be taken into account 1 Day Ahead as our bid.

•Relevance and impact of battery energy storage size in the forecast error reduction.

8.2 Contributions

Reviewing the most relevant results obtained during this master thesis work I’ve have clearly

presented the following aspects:

•Bidding model algorithm development and methodology

•Real time model algorithm development and methodology

•Forecast error metric calculations where can be checked the eﬀectiveness of the models

•A complete data analysis of the SOC(t),Pcurtailed(t)and P_H P P (t)

52 8 Conclusions and Future work

8.3 Conclusion

As RES are totally weather dependent so are intermittent and stochastic resources and there

is an essential need of incorporate battery energy storage systems. It is known that if we

want to complete and allow them to be fully integrated within the energy market then BESS

participation is absolutely necessary. RES are very valuable and adding energy storage systems

to the hybrid wind and solar plants if integrated in large scale. Including BESS we’re boosting

and allowing them to provided system services and congestion reduction for power systems.

During this master thesis work reviewing all the data analysis provided I can conclude the high

impact that battery energy storage has on two main aspects, which are:

•minimizing the forecast error, P_H P P _F E(t)

•minimizing the power curtailment, Pcurtailed(t)

Another important aspect to be taken into account is the relevance of

P_LIMbatt

parameter

value, it has been fully reviewed the eﬀect of its increase and how it aﬀects to curtailment and

forecast error reduction too. To have success with both aspects battery size is decisive, being

a rule of thumb, the bigger the better in terms of forecast error and curtailment reduction.

Of course, needless to say that the model methodology algorithm design is crucial to achieve

excellent results. For that purpose it is necessary to establish correctly which is going to be

your strategy and priority. In this work it has been to minimize the forecast error.

Then BESS becomes more relevant, in terms of improving its revenue being able to sell

the energy when prices are higher. As it allows us to move the power in time transferring to

the grid at later times when the price is higher.

8.4 Future work 53

8.4 Future work

Further research aim to cover the following aspects:

•

To Develop model methodology algorithm to reach the optimal forecast error reduction.

•

RES power generation forecasting using ensemble approach based on deep learning and

statistical methods.

•

To include within the model methodology development the spot market prices as one of

the inputs.

•

To include new energy storage system combinations: electric + thermal and chemical

storage to be added to the hybrid wind and solar pv plant.

•Integration of Power-to-X technologies storage in energy islands.

Bibliography

[1] IEA – International Energy Agency "Renewable Energy Market Update Outlook for 2020

and 2021", May 2020.

[2] Katherine Dykes, Jennifer King, Nicholas Di Orio, Ryan King, Vahan Gevorgian, Dave

Corbus, Nate Blair, Kate Anderson, Greg Stark, Craig Turchi, and Patrick Moriarty, "Oppor-

tunities for Research and Development of Hybrid Power Plants", National Renewable Energy

Laboratory, May 2020.

[3] Wind Europe report "Wind energy in Europe in 2019 - Trends and Statistics", February 2020.

[4] Kaushik Das, Anatole Grapperon, Poul Ejnar Sørensen and Anca Daniela Hansen, "Optimal

operation of battery for wind-storage hybrid power plant", HYBRIDize project, December 2019.

[5] Thomas Bowen, Ilya Chernyakhovskiy, Paul Denholm, NREL "Grid-Scale Battery Storage:

Frequently Asked Questions", September 2019.

[6] Anca Daniela Hansen, Poul Ejnar Sørensen, Kaushik Das, Edgar Nuño, Jayachandra

Naidu Sakamuri and Muﬁt Altin, "Dynamic modelling of Wind-Solar-Storage Based Hybrid

Power Plant", Project: HYBRIDize, 18th International Wind Integration Workshop, Dublin,

Ireland, October 2019.

[7] Wind Europe report, "Renewable Hybrid Power Plants - Exploring the Beneﬁts and Market

Opportunities", July 2019.

[8] Kaushik Das et al., “Enhanced features of wind based hybrid power plants”, in Pro-

ceedings of the 4th International Hybrid Power Systems Workshop, May 2019.

[9] Edgar Nuño,Matti Koivisto, Nicolaos A. Cutululis and Poul Ejnar Sørensen, "On the

Simulation of Aggregated Solar PV Forecast Errors" October 2018.

[10] Matti Juhani Koivisto, Kaushik Das, Feng Guo, Poul Sørensen, Edgar Nuño, Nicolaos

Bibliography 55

Cutululis and Petr Maule, "Using time series simulation tool for assessing the eﬀects of variable

renewable energy generation on power and energy systems", Wiley Interdisciplinary Reviews:

Energy and Environment, Volume 8, Issue 3, pp. e329, 2018.

[11] Paulina Asbeck (Vattenfall), “Next-Gen Generation System: The symbiotic relation-

ship of solar, wind storage hybrid power plants", 17th Wind Integration Workshop, October

2018.

[12] Matti Koivisto et al.,"Using time series simulation tools for assessing the eﬀects of variable

renewable energy generation on power and energy systems", August 2018.

[13] Lennart Petersen, Antonio Martínez, R.M Borsotti Andruszkiewicz, G.C. Tarnowski,

N. Steggel, D. Osmond (Vestas), "Vestas Power Plant Solutions - Integrating Wind, Solar PV

and Energy Storage", 3rd International Hybrid Power Systems Workshop, May 2018.

[14 ] Jose L. Crespo-Vázquez, C. Carrillo, E. Díaz-Dorado, Jose A. Martínez-Lorenzo, Md

Noor-E-Alam , "Evaluation of a data driven stochastic approach to optimize the participation

of a wind and storage power plant in day-ahead and reserve markets", May 2018.

[15] Edgar Nuño, Petr Maule, Andrea Hahmann, Nicolaos Cutululis, Poul Sørensen ,Ioanna

Karagali, "Simulation of transcontinental wind and solar PV generation time series", / Renew-

able Energy 118 425-436 (2018), November 2017.

[16] Wind Europe report, "Wind energy and on-site energy storage", November 2017.

[17] Ekström, J., Koivisto, M., Mellin, I., Millar, R. J., Lehtonen, M., "A statistical model for

hourly large-scale wind and photovoltaic generation in new locations.", IEEE Transactions on

Sustainable Energy, 8(4), 1383–1393, 2017.

[18] Tian Tian and Ilya Chernyakhovskiy, NREL "Forecasting Wind and Solar Generation:

Improving System Operations", January 2016.

[19] Quan Minh Duong, Marco Mussetta, F. Grimaccia, Emanuele Ogliari and S. Leva, "Hybrid

Model for Hourly Forecast of Photovoltaic and Wind Power", IEEE International Conference

on Fuzzy Systems, July 2013.

[20] Yann Riﬀonneau, Seddik Bacha, Franck Barruel and Stephane Ploix, "Optimal Power

Flow Management for Grid Connected PV Systems With Batteries", IEEE Transactions on

Sustainable Energy, Vol. 2, No. 3, July 2011.

[21] Sercan Teleke, Mesut E. Baran, Subhashish Bhattacharya, Alex Q. Huang, "Optimal

Control of Battery Energy Storage for Wind Farm Dispatching", IEEE Transactions on Energy

Conversion, Vol. 25, No. 3, September 2010.

[22] Josef Kallrath, Panos M. Pardalos, Steﬀen Rebennack, Max Scheidt, "Optimization in the

Energy Industry", Ed. Springer, 2009.

56 Bibliography

[23] Luís M. Costa, Franck Bourry, Jérémie Juban and George Kariniotakis, "Management of

Energy Storage Coordinated with Wind Power under Electricity Market Conditions", Proba-

bilistic Methods Applied to Power Systems Conference, May 2008.

[24] Jorge Marquez Angarita and Julio Garcia Usaola, "Combining Hydro-Generation and Wind

Energy Biddings and Operation on Electricity Spot Markets", Electric Power Systems Research

77 393–400, June 2006.

[25] G.N. Bathurst and G. Strbac "Value of combining energy storage and wind in short-

term energy and balancing markets", Electric Power Systems Research 67, 1-8, 2003.

[26] Magnus Korpaas, Arne T. Holen and Ragne Hildrum, "Operation and sizing of energy

storage for wind power plantsin a market system", Electrical Power and Energy Systems 25,

599–606, 2003.

[27] Jorge Marquez Angarita and Julio Garcia Usaola, "Combining Hydro-Generation and Wind

Energy Biddings and Operation on Electricity Spot Markets", Electric Power Systems Research

77 393–400, 2007.

[28] Danish Energy Agency - Energistyrelsen, "Denmark’s Energy and Climate Outlook

(DECO19)",

https://ens.dk/en/our-services/projections-and-models/denmarks-energy-

and-climate-outlook.

[29] Swedish Energy Agency - Energimyndigheten,"Energy in Sweden 2019: An overview",

https://energimyndigheten.a-w2m.se/Home.mvc?ResourceId=5794.

[30] Nord Pool Marke data - Day Ahead Prices,

https://www.nordpoolgroup.com/Market-

data1/#/nordic/table.

[31] IEA – International Energy Agency "Renewable Energy Market Update Outlook for

2020 and 2021", May 2020.

Appendix A: Matlab Code

This Matlab code presented has not been optimized but implemented following a clarity criteria,

although I have already veriﬁed that the code is correct and the contribution of this master

thesis is valid, this code must be used for reference purposes only, without any warranty or

liability for my part.

1

2% === = = = = = = ====== = = = = = = == ===== = = = = = = = ====== = = = = = = ====== = = = = = = ====== = = = = %

3% Emilio Barrachina Gasco %

4% s1 8 2 80 4 @ s tu d e n t . d tu . d k %

5% Master thesis %

6% Wind Energy Master %

7% Te c h n i c a l U n i v e r s i t y o f D e n m a r k %

8% J u ly 1 s t 2 0 20 %

9% === = = = = = = ====== = = = = = = == ===== = = = = = = = ====== = = = = = = ====== = = = = = = ====== = = = = %

10

11 c le a r a ll ;c l os e a l l ;cl c ;

12

13 %% CorRES TIME SERIES INPUTS

14

15 %% INPUTS

16

17 P_RES_F = xlsread(’ DK 2 R ES _F ’ ,’ S ce n a ri o 9 ’ ,’ B 2 : B 17 5 4 5 ’ );

18 f re q u en c y = x ls r e ad ( ’ DK 2 R ES _F ’ ,’ S ce n a ri o 9 ’ ,’ C2 : C 1 7 54 5 ’ ) ;

19 P _R E S = x l sr e a d ( ’ DK 2 R ES ’ ,’ S ce n a ri o 9 ’ ,’ A 2 : A 17 5 4 5 ’ );

20

21 % Zo o m s t u dy for on l y t wo days

22 %{

23 P_RES_F = xlsread(’ DK 2 R ES _F ’ ,’ S ce n a ri o 9 ’ ,’ B 2 : B4 8 ’ ) ;

24 f re q u en c y = x ls r e ad ( ’ DK 2 R ES _F ’ ,’ S ce n a ri o 9 ’ ,’ D2 : D 49 ’ );

25 P _R E S = x l sr e a d ( ’ DK 2 R ES ’ ,’ S ce n a ri o 9 ’ ,’ C 2 : C4 9 ’ ) ;

26 %}

27 t_range = 1:1 7 5 4 4 ;

28 tim e = 1 : length(t_range);

29

30

31 SOC_pre_F = 0.5; % [ p . u . ]

32 SOC_pre = 0. 5 ; % [ p . u .]

33 P_grid = 100; % [ MW ] , g r i d c o nn e c t io n p o we r c on s t ra i n

34

35 C_rate = 1; % [ h o u rs ] , we can c h a n g e it to 2, 4 , 6 , 8 h o u r s

36 P_LIM_batt = 50; % [ MW ] , w e c an c h oo s e a h i gh e r v al u e , 2 5 , 50 , 7 5 ,. . .

37 % P _b a tt _ ma x _F = P _ LI M_ b at t / C _r at e ;

58 Appendix A: Matlab Code

38 % P _b a tt _ ma x = P_ L IM _ ba t t / C_ ra t e ;

39 P _b a tt _ ma x _F = P _L I M_ b at t ;

40 P _b a tt _ ma x = P_ L IM _ ba t t ;

41

42 E_batt = 10 0; % [ M Wh ]

43

44 %% YE A R L Y R ES ENERGY GENERATED AND C A P A C I T Y F A C T O R : FO R E C A S T A ND MEASURED

45

46 for t = ti m e

47 P _R E S_ F ( t) = P _R ES _ F (t ) ;

48 E_RES_F=su m (P_RES_F );

49 E _R E S_ F _y e ar = E _ RE S _F / 2 ;

50

51 P _R ES ( t ) = P_ R ES ( t );

52 E _R E S = su m ( P _R ES ) ;

53 E _R E S_ y ea r = E _R E S /2 ;

54 end

55

56 E_RES_F_year = E_RES_F_year /1000000 % [ TW h / y ea r ]

57 E_RES_year = E_RES_year / 1 0 0 0 0 0 0 % [ TW h / y ea r ]

58

59 C F_ F = (( E _ RE S _F _ ye a r ) /( 8 76 0 ) ) *1 00 0 00 0 % [%]

60 CF = ( ( E_ R ES _ ye a r ) /( 8 76 0 ) ) *1 00 0 00 0 % [ %]

61

62 % FU L L WORK HOU R S : FO R E C A S T A N D MEASURED

63 FWH_ F = ( CF _F * 87 60 ) /1 00 % [ h ou r s /y e ar ]

64 F WH = ( C F * 8 76 0 ) / 1 00 % [ h ou rs / y e ar ]

65

66

67 %% BIDDING M ODEL

68 for t = ti m e

69

70 P _d i sc h ar g e _r e q_ F ( t) = - P_ R ES _ F (t ) + P_ g ri d ;

71

72 if P _ RE S _F ( t ) > P _g ri d

73 P _s u rp _ F (t ) = P _ RE S _F ( t ) - P _g ri d ;

74

75 elseif P _R E S_ F ( t) <= P _g ri d

76 P _s u rp _F ( t ) = 0;

77 end

78

79 if P_discharge_req_F (t)<0

80 P_discharge_req_F(t)=0;

81 end

82 P _b a tt _ re q _F ( t )= - P _R E S_ F ( t) + P _g r id ;

83 if P_batt_req_F (t)>P_batt_max_F

84 D el t a_ S OC _ re q _F ( t )= P _ ba t t_ ma x _F * ( -1 / E_ b at t );

85 P _c u r ta i l me n t _F ( t ) =0 ;

86 elseif P_batt_req_F(t)<-P_batt_max_F

87 D el t a_ S OC _ re q _F ( t )= - P _b a tt _ ma x_ F * ( -1 / E_ b at t );

88 P _c u rt a il m e nt _ F (t ) = P_ b at t _m a x_ F - P _ ba t t _r e q_ F ( t) ;

89 else

Appendix A: Matlab Code 59

90 D el t a_ S OC _ re q _F ( t )= P _ ba t t_ re q _F ( t ) *( - 1/ E _b a tt ) ;

91 P _c u r ta i l me n t _F ( t ) =0 ;

92 end

93 if D e l ta _ SO C _ re q _F ( t ) + SO C _p re _ F > 1

94 S OC _ F (t ) = 1;

95 P _c u rt a il m en t _F _ ( t) = D el t a_ SO C _r e q_ F ( t) + S OC _p re _ F - 1;

96 elseif D el t a_ S O C_ r eq _ F ( t) + S OC _ pr e _F <0 . 5

97 S OC _F ( t ) = 0. 5;

98 P_curtailment_F_(t)=0;

99 else

100 S OC _F ( t )= D e lt a _ SO C _r e q_ F ( t) + S OC _ pr e _F ;

101 P_curtailment_F_(t)=0;

102 end

103 D el t a_ S OC _ F ( t) = S OC _ F (t ) - S OC _ pr e _F ;

104 S OC _ pr e _F = S O C_ F ( t) ;

105 P _b a t t _F ( t ) = D e l ta _ S O C_ F ( t ) *( - E _b a t t / 1 ) ;

106 if P _ ba t t_ F ( t) > =0

107 P _d i sc h ar g e_ F ( t) = P _b a tt _ F (t ) ;

108 P _c h a rg e _F ( t ) =0 ;

109 else

110 P _d i s ch a r ge _ F ( t ) =0 ;

111 P _c h ar g e_ F ( t) = - P_ b at t _F ( t );

112 end

113 P _m i ss in g _F ( t )= - P _d i sc h ar g e_ r eq _ F (t ) + P_ d is c ha r ge _F ( t ) ;

114 P _H PP _ F (t ) = P_ g ri d + P_ m is s in g _F ( t );

115 P _c u rt a il m en t _t o ta l _F ( t )= P _ cu r ta i lm e nt _ F (t ) + P_ c ur t ai lm e nt _ F_ ( t );

116 end

117

118 %% REAL T I ME M O D E L

119 for t = ti m e

120

121 P _d i sc h a rg e _r e q ( t) = - P _R ES ( t ) + P_ H PP _ F (t ) ;

122 if P _ RE S ( t) > P _g r id

123 P _s ur p ( t ) = P _ RE S ( t) - P _ gr i d ;

124 elseif P _R ES ( t ) < = P _g ri d

125 P _s ur p ( t ) = 0 ;

126 end

127

128 if P _ di s ch a rg e _r e q (t ) <0

129 P _d i s ch a r ge _ r eq ( t ) =0 ;

130 end

131 P _b a tt _ re q ( t )= - P _R E S (t ) + P _H P P_ F ( t) ;

132 if P _ b a tt _ r eq ( t ) > P _b a t t_ m a x

133 D el t a_ S OC _r e q (t ) = P _b at t _m a x *( - 1/ E _ ba tt ) ;

134 P _c u r ta i l me n t ( t ) =0 ;

135 elseif P _b a tt _ r eq ( t ) <- P _ b at t _m a x

136 D el t a_ S OC _r e q (t ) = - P_ ba t t_ m ax * ( -1 / E _b at t ) ;

137 P _c u rt a il m en t ( t) = P _b at t _m ax - P _ ba t t_ r eq ( t );

138 else

139 D el t a_ S OC _ re q ( t) = P _b a tt _ re q ( t) * ( -1 / E_ b at t );

140 P _c u r ta i l me n t ( t ) =0 ;

141 end

60 Appendix A: Matlab Code

142 if D e l ta _ S OC _ re q ( t ) + SO C_ p re >1

143 S OC ( t ) =1 ;

144 P _c u rt a il m en t_ ( t ) = De l ta _S O C_ r eq ( t )+ S OC _ pr e - 1;

145 elseif D el t a _S O C_ r e q ( t) + S OC _ pr e < 0

146 S OC ( t ) =0 ;

147 P_curtailment_(t)=0;

148 else

149 S OC ( t) = D e lt a _S O C_ r eq ( t ) + SO C_ p re ;

150 P_curtailment_(t)=0;

151 end

152 D el t a _S O C ( t) = S OC ( t ) - S OC _ pr e ;

153 S OC _ pr e = S OC ( t ) ;

154 P _b at t ( t) = D el t a_ S OC ( t ) *( - E_ b at t / 1) ;

155 if P _ ba t t (t ) > =0

156 P _d i sc h ar g e ( t) = P _b a tt ( t );

157 P _c h ar g e ( t ) =0 ;

158 else

159 P _d i s ch a rg e ( t ) = 0;

160 P _c h ar g e (t ) = - P_ ba t t (t ) ;

161 end

162 P _m i ss in g ( t) = - P_ d is c ha r ge _ re q ( t) + P _d i sc ha r ge ( t );

163 P _H PP ( t ) = P_ HP P _F ( t ) + P_ m is s in g ( t) ;

164 P _c u rt a il m en t _t o ta l ( t) = P _c u rt a il m en t ( t) + P _c u rt a il m en t_ ( t ) ;

165 end

166

167 %% FULL M O D EL

168

169 for t = ti m e

170 P _H P P_ F E (t ) = P _H PP ( t ) - P_ H PP _ F (t ) ;

171 G RI D (t ) = P _H PP ( t ) ;

172 P _c u rt ai l ed ( t )= P _ cu r ta i lm en t _t o ta l ( t) ;

173 P _R ES ( t ) = P_ R ES ( t ) -P _ RE S _F ( t ) ;

174 end

175

176 % % E RR OR M E TR IC S : P _ HP P _F E ( t) M AF E a nd R SM FE

177

178 fprintf(’ P _ H PP _ F E E R RO R M ET R I CS % s \ n ’) ;

179

180 MAE_P_HPP_FE = mean(abs ( P _H P P_ F E ))

181 RMSE_P_H P P _ F E = sqrt(mean( ( P _ HP P _F E ) . ^2 ) )

182

183 % % E RR OR M E TR IC S : P _ RE S _F E ( t) M A E an d R S ME

184 fprintf(’ P _ R ES _ F E E R RO R M ET R I CS % s \ n ’) ;

185

186 MAE_P_RES_FE = mean(abs ( P _R E S ))

187 RMSE_P_R E S _ F E = sqrt(mean( ( P _R E S ) .^ 2 ) )

188

189 % % PL O TT I NG S

190

191 figure

192 plot( t im e , P _R E S_ F )

193 t it le (’ D K 2 H PP S c en a r io 9 - T i me s e ri e s - P \ _ RE S \ _ F ( t) ’ )

Appendix A: Matlab Code 61

194 xli m ([ 1 1 75 4 4] )

195 xlabel(’ T im e [ h ou r s ] ’)

196 ylabel(’ P \ _ RE S \ _ F [ MW ] ’ )

197 grid on

198

199 figure

200 plot( t im e , P _R E S )

201 t it le (’ D K 2 H PP S c en a r io 9 - T i me s e ri e s - P \ _ RE S ( t ) ’ )

202 xli m ([ 1 1 75 4 4] )

203 xlabel(’ T im e [ h ou r s ] ’)

204 ylabel(’ P \ _ RE S [ M W ] ’ )

205 grid on

206

207

208 figure

209 plot( t im e , P _s ur p _F )

210 t it le (’ D K 2 H PP S c en a r i o 9 - T im e s er i e s - P _ { su r p } \ _F ( t ) ’ )

211 xli m ([ 1 1 75 4 4] )

212 xlabel(’ T im e [ h ou r s ] ’)

213 ylabel(’ P _ { s ur p l us } \ _F [ M W ] ’ )

214 grid on

215

216 figure

217 plot( t im e , P _ su r p )

218 t it le (’ D K 2 H PP S c en a r i o 9 - T im e s er i e s - P _ { su r p } ( t) ’ )

219 xli m ([ 1 1 75 4 4] )

220 xlabel(’ T im e [ h ou r s ] ’)

221 ylabel(’ P _ { s ur p l us } [ MW ] ’ )

222 grid on

223

224 figure

225 plot(time,P_curtailment_total_F)

226 t it le (’ D K 2 H PP S c en a r i o 9 - T im e s er i e s - P _ { cu r t a il e d } \ _ F (t ) ’ )

227 xli m ([ 1 1 75 4 4] )

228 xlabel(’ T im e [ h ou r s ] ’)

229 ylabel(’ P _ { c ur t a i le d } \ _ F [ MW ] ’ )

230 grid on

231

232 figure

233 plot(time,P_curtailment_total)

234 t it le (’ D K 2 H PP S c en a r i o 9 - T im e s er i e s - P _ { cu r t a il e d } ( t ) - J un e 8 t h a nd

9 th 2 01 5 ’ )

235 xli m ([ 1 1 75 4 4] )

236 xlabel(’ T im e [ h ou r s ] ’)

237 ylabel(’ P _ { c ur t a i le d } [ M W ] ’)

238 grid on

239

240 figure

241 plot( t im e ,( S OC _ F * P_ LI M _b a tt ) )

242 t it le (’ D K 2 H PP S c en a r i o 9 - T im e s er i e s - S OC \ _ F ( t ) - J un e 8 t h a nd 9 t h

2015’)

243 xli m ([ 1 1 75 4 4] )

62 Appendix A: Matlab Code

244 xlabel(’ T im e [ h ou r s ] ’)

245 ylabel(’ S O C \ _F * P _ L I M_ b a tt [ M W ] ’)

246 grid on

247

248 figure

249 plot( t im e ,( S O C* P _ LI M _b a tt ) )

250 t it le (’ D K 2 H PP S c en a r io 9 - T i me s e ri e s - S OC ( t ) - J un e 8 t h a nd 9 t h 2 0 15 ’

)

251 xlim([1 48])

252 xlabel(’ T im e [ h ou r s ] ’)

253 ylabel(’ S O C * P _L I M _b a t t [ M W ] ’)

254 grid on

255

256

257 figure

258 plot( t im e , P _c ha r ge _ F )

259 t it le (’ D K 2 H PP S c en a r i o 9 - T im e s er i e s - P _ { ch a r ge d }\ _ F (t ) - J u n e 8 th

a nd 9 t h 2 0 15 ’ )

260 xlim([1 48])

261 xlabel(’ T im e [ h ou r s ] ’)

262 ylabel(’ P o we r C h ar g e d [ MW ] ’ )

263 grid on

264

265 figure

266 plot( t im e , P _c ha r ge )

267 t it le (’ D K 2 H PP S c en a r i o 9 - T im e s er i e s - P _ { ch a r ge d }( t ) - J un e 8 t h a nd 9

th 2 01 5 ’)

268 xlim([1 48])

269 xlabel(’ T im e [ h ou r s ] ’)

270 ylabel(’ P o we r C h ar g e d [ MW ] ’ )

271 grid on

272

273 figure

274 plot( t im e , P _d is c ha r ge _ F )

275 t it le (’ D K 2 H PP S c en a r i o 9 - T im e s er i e s - P _ { d is c h ar g e d } \ _F ( t ) - J u ne 8 th

a nd 9 t h 2 0 15 ’ )

276 xlim([1 48])

277 xlabel(’ T im e [ h ou r s ] ’)

278 ylabel(’ P o we r d i sc h a r ge d [ M W ] ’ )

279 grid on

280

281 figure

282 plot( t im e , P _d is c ha r ge )

283 t it le (’ D K 2 H PP S c en a r i o 9 - T im e s er i e s - P _ { d is c h ar g e d } ( t ) - J un e 8 t h

a nd 9 t h 2 0 15 ’ )

284 xlim([1 48])

285 xlabel(’ T im e [ h ou r s ] ’)

286 ylabel(’ P o we r d i sc h a r ge d [ M W ] ’ )

287 grid on

288

289 figure

290 plot( frequency ,sort( P _ su r p_ F ) )

Appendix A: Matlab Code 63

291 t it le (’ D K 2 H PP S c en a r i o 9 - D u ra t i on C u rv e - P_ { s ur p } \ _ F (t ) - J u ne 1 7 th

a nd 1 8 th 2 01 6 ’ )

292 xli m ([ 0 1 ])

293 xlabel(’ C u m ul a te d f r eq u en c y ’ )

294 ylabel(’ P _ { s ur p l us } \ _F [ M W ] ’ )

295 grid on

296

297 figure

298 plot( frequency ,sort( P_surp))

299 t it le (’ D K 2 H PP S c en a r i o 9 - D u ra t i on C u rv e - P_ { s ur p } ( t ) - J un e 1 7 t h a nd

18 t h 2 01 6 ’ )

300 xli m ([ 0 1 ])

301 xlabel(’ C u m ul a te d f r eq u en c y ’ )

302 ylabel(’ P _ { s ur p l us } [ MW ] ’ )

303 grid on

304

305 figure

306 plot( frequency ,sort( P_curtailment_total_F))

307 t it le (’ D K 2 H PP S c en a r i o 9 - D u ra t i on C u rv e - P_ { c u rt a i le d }\ _ t ot a l \ _ F (t ) -

J un e 1 7 th a nd 1 8 th 2 0 16 ’ )

308 xli m ([ 0 1 ])

309 xlabel(’ C u m ul a te d f r eq u en c y ’ )

310 ylabel(’ P _ { c ur t a i le d } \ _ F [ MW ] ’ )

311 grid on

312

313 figure

314 plot( frequency ,sort( P_curtailment_total))

315 t it le (’ D K 2 H PP S c en a r i o 9 - D u ra t i on C u rv e - P_ { c u rt a i le d }\ _ t ot a l ( t ) -

J un e 1 7 th a nd 1 8 th 2 0 16 ’ )

316 xli m ([ 0 1 ])

317 xlabel(’ C u m ul a te d f r eq u en c y ’ )

318 ylabel(’ P _ { c ur t a i le d } [ M W ] ’)

319 grid on

320

321 figure

322 plot( frequency ,sort( S O C_ F ) )

323 t it le (’ DK 2 HPP S c e n a r i o 9 - D u r a t i o n C u r ve - S OC \ _ F ( t) - Ju n e 17 th and 18

th 2 01 6 ’)

324 xli m ([ 0 1 ])

325 yli m ([ 0 1 .2 ])

326 xlabel(’ C u m ul a te d f r eq u en c y ’ )

327 ylabel(’ S O C \ _F [ p . u .] ’ )

328 grid on

329

330 figure

331 plot( frequency ,sort( S O C ))

332 t it le (’ D K2 H PP S ce n ar i o 9 - D u ra t io n C u rv e - S OC ( t ) - J un e 1 7 th a nd 1 8 th

2016’)

333 xli m ([ 0 1 ])

334 yli m ([ 0 1 .2 ])

335 xlabel(’ C u m ul a te d f r eq u en c y ’ )

336 ylabel(’ S O C [ p . u . ] ’)

64 Appendix A: Matlab Code

337 grid on

338

339 figure

340 plot( frequency ,sort( P _ di s ch a rg e _F ) )

341 t it le (’ D K 2 H PP S c en a r i o 9 - D u ra t i on C u rv e - P_ { d i sc h a rg e d } \ _ F ( t) - J un e

17 t h a nd 1 8 th 2 0 16 ’ )

342 xli m ([ 0 1 ])

343 yli m ([ 0 3 ])

344 xlabel(’ C u m ul a te d f r eq u e nc y ’ )

345 ylabel(’ P o we r d i sc h a r ge d [ M W ] ’ )

346 grid on

347

348 figure

349 plot( frequency ,sort( P_discharge))

350 t it le (’ DK 2 HPP S c e n a r i o 9 - D u r a t i o n C u r ve - _ { di s c h a r g e d }( t ) - J u ne 17 t h

a nd 1 8 th 2 01 6 ’ )

351 xli m ([ 0 1 ])

352 xlabel(’ C u m ul a te d f r eq u e nc y ’ )

353 ylabel(’ P o we r d i sc h a r ge d [ M W ] ’ )

354 grid on

Appendix B: Measured

Scenarios Comparison

DK2 HPP - Measured

DK2 Energy surplused measured for hybrid power plant

scenarios comparison

RES Scenarios ERE S (t)[T W h/y]Esurplused (t)[GW h/y]Ecur tailed(t)[GW h/y ]

Sc.1: 0WPP+100PVPP 0.129 0.000 0.000

Sc.2: 25WPP+75PVPP 0.189 0.000 0.000

Sc.3: 50WPP+50PVPP 0.249 0.000 0.000

Sc.4: 75WPP+25PVPP 0.309 0.000 0.000

Sc.5: 100WPP+0PVPP 0.369 0.000 0.000

Sc.6: 80WPP+60PVPP 0.372 2.525 2.525

Sc.7: 100WPP+40PVPP 0.420 4.984 4.984

Sc.8: 105WPP+35PVPP 0.432 6.347 6.347

Sc.9: 120WPP+20PVPP 0.468 25.505 25.505

Table 1: DK2 RES Scenarios comparison - Energy Produced and Surplused

RES Scenarios ERE S (t)[T W h/y]Ecurtailed (t)[GW h/y]Ecur tailed(t)[%]

Sc.1: 0WPP+100PVPP 0.129 0.000 0.000

Sc.2: 25WPP+75PVPP 0.189 0.000 0.000

Sc.3: 50WPP+50PVPP 0.249 0.000 0.000

Sc.4: 75WPP+25PVPP 0.309 0.000 0.000

Sc.5: 100WPP+0PVPP 0.369 0.000 0.000

Sc.6: 80WPP+60PVPP 0.372 2.525 0.68

Sc.7: 100WPP+40PVPP 0.420 4.984 1.19

Sc.8: 105WPP+35PVPP 0.432 6.347 1.47

Sc.9: 120WPP+20PVPP 0.468 25.505 5.45

Table 2: DK2 RES Scenarios comparison - Energy Produced Measured, Surplused and

Curtailed

66 Appendix B: Measured Scenarios Comparison

RES Scenarios ERE S (t)[T W h/y]Ecurtailed (t)[GW h/y]Ecur tailed(t)[%] C F [%]

Sc.1: 0WPP+100PVPP 0.129 0.000 0.00 14.70

Sc.2: 25WPP+75PVPP 0.189 0.000 0.00 21.55

Sc.3: 50WPP+50PVPP 0.249 0.000 0.00 28.40

Sc.4: 75WPP+25PVPP 0.309 0.000 0.00 35.25

Sc.5: 100WPP+0PVPP 0.369 0.000 0.00 42.10

Sc.6: 80WPP+60PVPP 0.372 2.525 0.68 42.50

Sc.7: 100WPP+40PVPP 0.42 4.984 1.19 47.98

Sc.8: 105WPP+35PVPP 0.432 6.347 1.47 49.35

Sc.9: 120WPP+20PVPP 0.468 25.505 5.45 53.46

Table 3: DK2 RES Scenarios comparison - Energy Produced Measured, Surplused and

Curtailed

SE2 HPP - Measured 67

SE2 HPP - Measured

SE2 Energy Surplused for hybrid power plant scenarios

comparison

RES Scenarios ERES_F(t)[T W h/y]Esurplused_F(t)[GW h/y]Ecurtailed_F(t)[GW h/y]

Sc.1: 0WPP+100PVPP 0.118 0.000 0.000

Sc.2: 25WPP+75PVPP 0.142 0.000 0.000

Sc.3: 50WPP+50PVPP 0.166 0.000 0.000

Sc.4: 75WPP+25PVPP 0.190 0.000 0.000

Sc.5: 100WPP+0PVPP 0.215 0.000 0.000

Sc.6: 80WPP+60PVPP 0.242 1.228 1.228

Sc.7: 100WPP+40PVPP 0.262 2.029 2.029

Sc.8: 105WPP+35PVPP 0.267 2.445 2.445

Sc.9: 120WPP+20PVPP 0.281 8.997 8.997

Table 4: SE2 RES Scenarios comparison - Energy Produced and Surplused

RES Scenarios ERES_F(t)[T W h/y]Ecurtailed_F(t)[GW h/y]Ecurtailed_F(t)[%]

Sc.1: 0WPP+100PVPP 0.118 0.000 0.000

Sc.2: 25WPP+75PVPP 0.142 0.000 0.000

Sc.3: 50WPP+50PVPP 0.166 0.000 0.000

Sc.4: 75WPP+25PVPP 0.190 0.000 0.000

Sc.5: 100WPP+0PVPP 0.215 0.000 0.000

Sc.6: 80WPP+60PVPP 0.242 1.228 0.51

Sc.7: 100WPP+40PVPP