Impact evaluation of the large scale integration of electric vehicles in the security of suply

Faculty of Engineering of University of Porto Department of Electrical and Computer Engineering Impact Evaluation of the Large Scale Integration of Electric Vehicles in the Security of Supply Leonardo Elizeire Bremermann Thesis submitted to the Faculty of Engineering of University of Porto in partial fulfilment of the requirements for the degree of Doctor of Philosophy Supervisor: Prof. Manuel Matos, Ph.D. Full Professor at the Department of Electrical and Computer Engineering Faculty of Engineering of University of Porto Co-supervisor: Prof. Jo˜ao Abel Pe¸cas Lopes, Ph.D. Full Professor at the Department of Electrical and Computer Engineering Faculty of Engineering of University of Porto Porto, April 2014 rLeonardo Bremermann “...Let us fight for a new world - a decent world that will give men a chance to work - that will give youth a future and old age a security...Let us fight for a world of reason, a world where science and progress will lead to all men’s happiness...We are coming into a new world, a kindlier world, where men will rise above their hate, their greed and brutality.” The Great Dictator (1940) - directed by Charlie Chaplin This work was supported by the Funda¸c˜ao de Ciˆencia e Tecnologia (FCT) – reference SFRH/BD/65355/2009, by CAPES/CNPq under the Ciˆencia sem Fronteiras programme – reference 237983/2012-5, the MIT Portugal Program on Sustainable Energy Systems, and the Institute for Systems and Computer Engineering of Porto (INESC Porto). The research was developed under the framework of the European project Mobile Energy Resources for Grids of Electricity (MERGE) – contract n◦241399 (FP7) and the Funda¸c˜ao de Apoio `a Inova¸c˜ao (FAI) through the project Redes El´ectricas Inteligentes com Ve´ıculos El´ectricos (REIVE). Acknowledgments People in Portugal have received me with open arms since I arrived in Porto, by the first time, in 2007 and therefore, I have many thanks to say. I hope not to forget any one. First of all, I would like to thank my supervisor Prof. Manuel Matos. I thank for all discussions we had about the thesis subject to make this final version become better. He has taught me not only the matters of interest of this thesis, but also, through his honesty, sincerity, serious work and coordination of the Power Systems Unit of the Institute for Systems and Computer Engineering of Porto (Instituto de Engenharia de Sistemas e Computadores do Porto - INESC Porto), the way to become a researcher and rigorous professional. I am grateful to have been his student throughout these years. For these reasons I would like to express my sincere gratitude. I would like to thank my co-supervisor Prof. Jo˜ao Abel Pe¸cas Lopes, which opened the INESC Porto doors to me, and pointed the subject of this thesis, leading me to pursue it and defend this idea fiercely. He also actively participated in the thesis writing by providing several discussions about the issues addressed in this dissertation. He gave me an opportunity to teach at Faculty of Engineering of the University of Porto (Faculdade de Engenharia da Universidade do Porto - FEUP) guiding me through this task. Thank you Prof. Jo˜ao Abel Pe¸cas Lopes. A special thank to Prof. Mauro Augusto da Rosa, which has followed all the ix popula¸c˜ao e as diferentes estrat´egias de carregamento dos VE. O objectivo ´e incluir estes modelos na avalia¸c˜ao da adequa¸c˜ao da seguran¸ca do abastecimento para sistemas de gera¸c˜ao com alto n´ıvel de integra¸c˜ao de potˆencia e´olica. Os modelos de VE desenvolvidos nesta tese incluem os conceitos de carregamento controlado e n˜ao controlado das baterias, os quais s˜ao divididos em carregamento directo, no vazio, controlado e “vehicle-to-grid” (V2G). Este ´ultimo, ´e desenvolvido sob duas perspectivas: contribui¸c˜ao para aumento da reserva operacional e compensa¸c˜ao na varia¸c˜ao da produ¸c˜ao de electricidade a partir da energia e´olica. ´ E esperado que, sob certas circunstˆancias, as estrat´egias de carregamento controlado criem oportunidades de neg´ocios para os setores de electricidade, fornecendo servi¸cos de sistema para mitigar o impacto dos VE na adequa¸c˜ao da seguran¸ca do abastecimento. Este t´opico ´e explorado, atrav´es desta tese, assumindo a existˆencia de uma entidade agregadora que ser´a respons´avel pelo carregamento dos VE com o objectivo de carregar e gerir a energia el´ectrica armazenada nas baterias. Os modelos de VE desenvolvidos foram inclu´ıdos no m´etodo de Simula¸c˜ao de Monte Carlo Sequencial (SMCS), o qual consiste num m´etodo probababil´ıstico capaz de representar o comportamento estoc´astico dos componentes do sistema, levando em conta a caracter´ıstica temporal dos seus estados de opera¸c˜ao. Os modelos propostos foram avaliados atrav´es da utiliza¸c˜ao do sistema teste de fiabilidade IEEE 1996 e de sistemas reais tais como Portugal, Espanha e Gr´ecia, para a configura¸c˜ao dos sistemas de gera¸c˜ao de 2030. O desempenho dos modelos foi analisado atrav´es de diferentes cen´arios de integra¸c˜ao de VE. A an´alise dos resultados mostra que em caso de integra¸c˜ao em larga escala de VE, o carregamento destes deve ser monitorizado atrav´es de estrat´egias de carregamento controlado para evitar a necessidade de aumentar a capacidade de gera¸c˜ao no futuro. As possibilidades de controlar e injectar energia el´ectrica no sistema tˆem demonstrado ser um suporte efectivo para a reserva operacional, mantendo a adequa¸c˜ao da seguran¸ca do abastecimento nos n´ıveis estimados para cen´arios sem integra¸c˜ao dos VE. Resum´e Le d´eveloppement durable des secteurs du transport et de l’´energie a ´et´e consid´er´e comme ´etant le principal objectif pour r´eduire les ´emissions de gaz `a effet de serre (GES), en g´en´eral, et le dioxyde de carbone (CO2), en particulier. Le d´eploiement massif des v´ehicules ´electriques (VE) et la croissance de l’utilisation des sources d’´energie renouvelable (SER) dans la g´en´eration sont consid´er´es comme les meilleures options pour r´eduire les ´emissions de GES dans les secteurs du transport et de la production d’´energie ´electrique. Cependant, la croissance de l’utilisation des SER et l’int´egration `a grande ´echelle des VE exigent un ensemble de centrales ´electriques avec des g´en´erateurs flexibles pour compenser les variations intrins`eques `a la production ´eolienne et aux besoins du transport ´electrique. Sans consid´erer les ressources hydro-´electriques, l’´energie ´eolienne est la principale SER d´evelopp´ee par l’industrie de production d’Electricit´e. Dans des pays comme le Portugal, l’Espagne et l’Allemagne, les centrales ´eoliennes ont d´ej`a atteint une p´en´etration variant entre 20% et 30% de la puissance totale install´ee. Un certain niveau de r´eserve op´erationnelle est n´ecessaire pour compenser les variations dues aux pannes et `a l’incertitude li´ee `a la pr´evision de la charge et de la production ´eolienne. La s´ecurit´e d’approvisionnement devrait ˆetre garantie par les institutions de r´egulation au niveau europ´een et son ´evaluation `a long-terme peut assurer que la g´en´eration planifi´ee pour le futur soit capable de couvrir la charge pr´evue. xvii Cette th`ese propose une m´ethodologie bas´ee sur le processus de Poisson, en consid´erant les habitudes de mobilit´e de la population et diff´erentes strat´egies de charge des batteries des VE, pour son inclusion dans l’´evaluation de la s´ecurit´e de l’approvisionnement dans les syst`emes dont la g´en´eration connait une forte p´en´etration d’´energie ´eolienne. Les mod`eles de charge des VE d´evelopp´es dans cette th`ese prennent en compte une charge contrˆol´ee et non-contrˆol´ee des batteries des VE qui peut ˆetre caract´eris´ee en diff´erentes strat´egies de charge: charge directe, charge durant la p´eriode de creux du diagramme de la demande globale, charge non-contrˆol´ee et la possibilit´e que les batteries puissent injecter d’´energie dans le r´eseau en cas de besoin. Cette derni`ere strat´egie pr´esente deux perspectives: contribution `a la r´eserve op´erationnelle et compensation des fluctuations de la disponibilit´e de l’´energie ´eolienne. Dans certaines circonstances, la charge contr1ˆol´ee des batteries des VE pourra donc ˆetre un atout, pour le syst`eme ´electrique, en fournissant les services auxiliaires du r´eseau afin de r´eduire l’impact des VE sur la capacit´e du syst`eme `a garantir la s´ecurit´e de l’approvisionnement. Ce th`eme a ´et´e trait´e dans cette th`ese en consid´erant l’existence d’une entit´e responsable pour la gestion de la charge des VE en mesure de contrˆoler l’´energie accumul´ee dans les batteries. Le mod`ele d´evelopp´e pour les VE a ´et´e inclus dans la m´ethode de Simulation S´equentielle de Monte Carlo (SSMC) capable de repr´esenter le comportement stochastique des composants du syst`eme, en prenant en consid´eration la d´ependance temporelle qui caract´erise leurs ´etats d’op´eration. Les mod`eles propos´es ont ´et´e ´evalu´es `a travers l’utilisation d’une configuration modifi´ee du r´eseau test de fiabilit´e IEEE 1996 et des configurations des r´eseaux r´eels du Portugal, de l’Espagne e de la Gr`ece planifi´ees pour 2030. Leur performance a ´et´e analys´ee en utilisant plusieurs sc´enarios d’int´egration avec ou sans d´eploiement des VE. L’analyse des r´esultats a permis de conclure que le d´eploiement massif doit ˆetre g´er´e `a travers une strat´egie de charge contrˆol´ee afin d’´eviter la n´ecessit´e de devoir augmenter la capacit´e de g´en´eration dans les ann´ees prochaines. Le contrˆole de l’´energie emmagasin´ee dans les batteries avec la possibilit´e de pouvoir la r´einjecter dans le r´eseau a d´emontr´e qu’il est possible de fournir une contribution effective en termes de r´eserve op´erationnelle garantissant ainsi le mˆeme niveau de s´ecurit´e d’approvisionnement estim´e dans des sc´enarios sans d´eploiement de VE. Contents Acknowledgement ix Abstract xiii Resumo xv Resum´e xvii List of Figures xxv List of Tables xxix List of Acronyms xxxiii 1 Introduction 1 1.1 Context and Motivation . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1.1 Power System Adequacy Evaluation . . . . . . . . . . . . . . 5 1.1.2 Electric Vehicles in the Power Systems . . . . . . . . . . . . 7 1.2 ResearchQuestions........................... 10 1.3 Main Hypothesis and Assumptions . . . . . . . . . . . . . . . . . . 11 1.4 ThesisObjectives............................ 13 1.5 ThesisStructure............................. 14 1.6 PublicationList............................. 14 2 Background and State of the Art 17 xxi 2.1 Introduction............................... 17 2.2 RegulationReserves .......................... 18 2.3 Adequacy Assessment Methods . . . . . . . . . . . . . . . . . . . . 20 2.3.1 Reliability Indices . . . . . . . . . . . . . . . . . . . . . . . . 20 2.3.2 Analytical Methods . . . . . . . . . . . . . . . . . . . . . . . 21 2.3.3 Simulation Methods . . . . . . . . . . . . . . . . . . . . . . 25 2.3.4 HybridMethods......................... 30 2.3.5 Load and Electric Vehicle Load Modelling . . . . . . . . . . 33 2.3.6 Wind Power Modelling . . . . . . . . . . . . . . . . . . . . . 35 2.4 Related Studies on Electric Vehicles . . . . . . . . . . . . . . . . . . 36 2.5 FinalRemarks.............................. 41 3 Short and Long-Term Risk Analysis 43 3.1 Introduction............................... 43 3.2 Short-term Reserve Evaluation . . . . . . . . . . . . . . . . . . . . . 47 3.2.1 Modelling Generating Unit Outages . . . . . . . . . . . . . . 50 3.2.2 Modelling Demand and Wind Power Forecast Uncertainties . 52 3.2.3 A Simple Example . . . . . . . . . . . . . . . . . . . . . . . 53 3.3 Long-term Reserve Evaluation . . . . . . . . . . . . . . . . . . . . . 55 3.3.1 Modelling Generating Unit Outages . . . . . . . . . . . . . . 62 3.3.2 Modelling Demand Forecast Uncertainty . . . . . . . . . . . 63 3.3.3 Modelling Wind Power Forecast Uncertainty . . . . . . . . . 65 3.3.4 Relationship between Generation and Load Uncertainties . . 66 3.3.5 A Simple Example . . . . . . . . . . . . . . . . . . . . . . . 68 3.4 Electric Vehicle Demand Modelling . . . . . . . . . . . . . . . . . . 69 3.5 FinalRemarks.............................. 70 4 Electric Vehicle Modelling 71 4.1 Introduction............................... 71 4.2 MobilityPattern ............................ 73 4.3 Proposed Counting Process Methodology . . . . . . . . . . . . . . . 79 4.3.1 Homogeneous Poisson Process . . . . . . . . . . . . . . . . . 80 4.3.2 Non-homogeneous Poisson Process . . . . . . . . . . . . . . 83 4.4 Uncontrolled Charging Models . . . . . . . . . . . . . . . . . . . . . 88 4.4.1 Direct Charging Strategy . . . . . . . . . . . . . . . . . . . . 88 4.4.2 Valley Charging Strategy . . . . . . . . . . . . . . . . . . . . 89 4.5 Controlled Charging Models . . . . . . . . . . . . . . . . . . . . . . 90 4.5.1 Controlled Charging Strategy . . . . . . . . . . . . . . . . . 91 4.5.2 Vehicle-to-Grid Charging Strategy . . . . . . . . . . . . . . . 92 4.6 Conventional Load and Proposed EV Load Estimation . . . . . . . 93 4.6.1 EV Load from the Homogeneous Poisson Process . . . . . . 95 4.6.2 EV Load from the Non-homogeneous Poisson Process . . . . 97 4.7 Proposed EV Models Integration on SMCS Method . . . . . . . . . 98 4.7.1 Static Reserve Evaluation with Electric Vehicles . . . . . . . 99 4.7.2 Operating Reserve Capacity Evaluation with Electric Vehicles100 4.8 FinalRemarks..............................104 5 Controlled Electric Vehicle Charging Modelling 109 5.1 Introduction...............................109 5.2 Active Charging Framework . . . . . . . . . . . . . . . . . . . . . . 111 5.3 Proposed Controlled Charging Modelling . . . . . . . . . . . . . . . 113 5.4 Reliability Aspects of Vehicle-to-Grid . . . . . . . . . . . . . . . . . 118 5.5 Proposed Vehicle-to-Grid Charging Modelling . . . . . . . . . . . . 124 5.5.1 Vehicle-to-Grid for Operating Reserve Capacity . . . . . . . 127 5.5.2 Vehicle-to-Grid for Wind Power Generation Balance . . . . . 129 5.6 FinalRemarks..............................131 6 Simulation and Result Analyses 133 6.1 Introduction...............................133 6.2 System Descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . 134 6.2.1 IEEE Reliability Test System - 1996 . . . . . . . . . . . . . 135 6.2.2 Real Generating Systems . . . . . . . . . . . . . . . . . . . . 136 6.3 Scenarios Description . . . . . . . . . . . . . . . . . . . . . . . . . . 139 6.3.1 IEEE Reliability Test System - 1996 HW . . . . . . . . . . . 139 6.3.2 Real Generating Systems Scenarios . . . . . . . . . . . . . . 142 6.3.3 Electric Vehicle Scenarios . . . . . . . . . . . . . . . . . . . 146 6.4 Validation of the Adequacy Evaluation of Generating Systems Tool 148 6.5 Reference Cases Analyses . . . . . . . . . . . . . . . . . . . . . . . . 149 6.5.1 Reference Case of the IEEE Reliability Test System - 1996 HW ...............................149 6.5.2 Reference Cases of the Real Generating Systems . . . . . . . 150 6.6 Results and Discussions . . . . . . . . . . . . . . . . . . . . . . . . 152 6.6.1 Real Generating Systems . . . . . . . . . . . . . . . . . . . . 153 6.6.2 IEEE Reliability Test System 1996 HW - Uncontrolled ChargingEVModels......................163 6.6.3 IEEE Reliability Test System 1996 HW - Controlled ChargingModels ........................175 6.7 Computational Burden . . . . . . . . . . . . . . . . . . . . . . . . . 180 6.8 FinalRemarks..............................180 7 Conclusions and Future Work 183 7.1 Conclusions ...............................183 7.2 Main Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 7.3 FutureWork...............................189 Bibliography 191 List of Figures 1.1 World GHG Emissions flowchart 2010 [1]. . . . . . . . . . . . . . . . 3 1.2 M1 EV uptake rates for each of three scenarios [2]. . . . . . . . . . 7 2.1 Flowchart of the Sequential Monte Carlo Simulation process. . . . . 28 2.2 Representations of the single load curve. . . . . . . . . . . . . . . . 33 3.1 Unit commitment representation based on technology predominance. 48 3.2 Operating Reserve Capacity evaluation. . . . . . . . . . . . . . . . . 59 3.3 Markov Model representations. . . . . . . . . . . . . . . . . . . . . 62 4.1 Classification of EV Models. . . . . . . . . . . . . . . . . . . . . . . 72 4.2 Sample of the northern regions of Portugal. . . . . . . . . . . . . . . 74 4.3 Weekday arrivals distribution. . . . . . . . . . . . . . . . . . . . . . 76 4.4 Weekend arrivals distribution. . . . . . . . . . . . . . . . . . . . . . 78 4.5 The counting process through an HPP. . . . . . . . . . . . . . . . . 81 4.6 SampleoftheHPP............................ 82 4.7 The counting process through an NHPP. . . . . . . . . . . . . . . . 85 4.8 SampleoftheNHPP........................... 87 4.9 System load profile using EV direct charging strategy. . . . . . . . . 89 4.10 System load profile using EV valley charging strategy. . . . . . . . . 90 4.11 System load profile using EV controlled charging strategy. . . . . . 91 4.12 System load profile using V2G charging strategy. . . . . . . . . . . 93 4.13 Illustration of the EV load calculation through HPP. . . . . . . . . 96 4.14 EV load profile using HPP and direct charging strategy. . . . . . . . 96 xxv List of Acronyms AGC Automatic Generation Control CC Controlled Charging CO2Carbon Dioxide COPFT Capacity Outage Probability and Frequency Table COPT Capacity Outage Probability Table DC Direct Charging DOD Depth of Discharge EENS Expected Energy Not Supplied EPNS Expected Power Not Supplied EU European Union EV Electric Vehicle(s) EV-AL Electric Vehicle Aggressive penetration Level scenario xxxiii EV-LL Electric Vehicle Low penetration Level scenario EV-ML Electric Vehicle Moderate penetration Level scenario F&D Frequency and Duration FCR Frequency Containment Reserve FFT Fast Fourier Transform FM Failure Mode FOR Forced Outage Rate FRR Frequency Restoration Reserve GCE Gram-Charlier Expansion GGS Greek Generation System GHG Greenhouse Gas Emission(s) HEV Hybrid Electric Vehicle(s) HL Hierarchical Level HPP Homogeneous Poisson Process ICE Internal Combustion Engine IEEE Institute of Electrical and Electronics Engineers Li-ion Lithium-ion LOLD Loss Of Load Duration LOLE Loss Of Load Expectation LOLF Loss Of Load Frequency LOLP Loss Of Load Probability MCS Monte Carlo Simulation MERGE Mobile Energy Resources in Grids of Electricity MTTF Mean Time To Failure MTTR Mean Time To Repair NERC North American Electric Reliability Corporation NHPP Non-Homogeneous Poisson Process NMAE Normalized Mean Absolute Error NOD Number of Discharges NSMCS Non-Sequential Monte Carlo Simulation ORC Operating Reserve Capacity ORR Outage Replacement Rate PBM Population-Based Method PGS Portuguese Generation System PHEV Plug-in Hybrid Electric Vehicle(s) PJM Pennsylvania-New Jersey-Maryland RBTS Roy Billinton Test System REIVE Redes El´etricas Inteligentes e Ve´ıculos El´etricos (Smart Grids with Electric Vehicles) RES Renewable Energy Source(s) RR Replacement Reserve RTS Reliability Test System SGS Spanish Generation System SI Swarm Intelligence SMCS Sequential Monte Carlo Simulation SO System Operator SOC State Of Charge STABALID Stationary Batteries Li-ion Safe Deployment US United States V2G Vehicle-to-Grid VC Valley Charging WFE Wind Power Forecast Error Chapter 1 Introduction 1.1 Context and Motivation Over the past two and a half centuries, the societies have burnt increasing amounts of fossil fuels to be used on power machines, generate electricity, heat buildings and transport people and goods. Since the industrial revolution, in 1750, the concentration of carbon dioxide (CO2) in the atmosphere has increased by roughly 40%, and it continues to rise [4]. Around 11% of the greenhouse gases emitted worldwide each year come from within the European Union (EU). In 2011, the latest year with available comprehensive data, EU-15 emissions stood 14.9% below their base year level. Based on estimates for 2012 by the European Environment Agency, EU-15 emissions averaged 12.2% below base-year levels during the 2008-2012 period (the target levels were 8%). This means the EU-15 over-achieved its first Kyoto’s target by a wide margin. The 8% collective reduction commitment has been translated into national emission reduction or limitation for each of the EU-15 Member States under what is known as the “burden sharing” agreement [5]. 1 Chapter 1 For 2020, the EU has made a unilateral commitment to reduce overall greenhouse gas emissions from its 28 Member States by 20% compared to 1990 levels. The EU has offered to increase this figure to 30% if other major economies agree to undertake their fair share of a global emissions reduction effort. In consonance with these objectives, the “2020 climate and energy package” seals the EU commitment on raising the share of EU energy consumption produced from renewable sources to 20%, improving the EU’s energy efficiency to 20% and reducing the greenhouse gas (GHG) emissions by 20% [6]. Portugal intends to have 60% of its generated electricity coming from RES by 2020, in order to satisfy 31% of its final energy consumption of the same year. In addition, Portugal aims at reducing its dependence on energy imports and on the use of fossil fuels [7]. Figure 1.1, which is an update of the 2000 World Resources Institute’s flowchart, makes clear how much CO2was produced in 2010 and by whom. This figure distinguishes between sectors according to their primary energy use (coal, gas and oil), which include mainly industry, transport and energy supply sectors. This flowchart also shows that the GHG emissions are originated mainly from two sources: direct and fossil fuel related emissions. Globally, industry sector accounts for 29% of the GHG emissions, followed by the transport and electrical energy supply sectors which account for 15% and 13% of the GHG emissions, respectively. The reduction mark for the transport sector for 2020 is 10% in EU-28 [8]. The electric vehicles (EV) are the main alternative technologies developed to achieve this goal allowing the reduction of GHG emissions by zero. From the new EV generation perspective, the hybrid electric vehicle (HEV) has been introduced in the market. This type of EV mixes the fossil fuels and electricity power sources reducing significantly, but not totally, the CO2emissions. The second alternative is the plug-in hybrid EV (PHEV), which beyond the fuel mix (fossil fuels and electrical energy) also uses rechargeable batteries, which can be charged from an external power source. The 2 1.1. Context and Motivation Figure 1.1: World GHG Emissions flowchart 2010 [1]. pure EV are the most recent alternative, which also consists of batteries that can be restored to full charge by connecting a plug to an external wall socket, however is the only one that has electricity as its only power source. This kind of vehicle (pure EV) plays a major role in the reduction of GHG emissions. On one hand, the EV has zero CO2emissions while used for mobility purpose. On the other hand, the battery charging can be controlled in order to increase the usage of renewable energy sources (RES), reducing the GHG emissions from the supply energy sector side [9]. The deployment of the RES and EV on the electric power systems will certainly affect the System Operator’s (SO) decision-making in terms of operation and planning. The increase of renewable sources has been included in the power system analysis through suitable models and methodologies. The expected large scale integration of EV in the electric power systems will also require the 3 Chapter 1 development of adequate models to be considered in the power system analysis methodologies. Among other issues, the security of supply at the generation system level is one of the main concerns of the European Community (EC) to achieve the implementation of a sustainable climate change policy. The monitoring of the security of supply is supported by a legal framework at the European level which is comprised in the Directive 2009/72/EC [10]. This Directive foresees that the Member States shall ensure the monitoring of security of supply issues. Such task should be delegated to the regulatory authorities. The monitoring shall, in particular, cover the balance of supply and demand on the national market, the level of expected future demand and envisage additional capacity being planned or under construction, and the quality and level of maintenance of the networks, as well as measure to cover peak demand and to deal with shortfalls of one or more suppliers [10]. Under this context, this thesis is concerned with the analysis of the EV impact on the generation systems with high integration of RES. In generation system analysis, it is usual not to include the transmission and distribution networks. Such evaluation can aid regulatory authorities to determine the need of increasing the generation capacity, identify the incentives for storage facilities to avoid wasting of renewable generation and other decisions which involve the security of supply. The assessment of security of supply may be divided into two perspectives: adequacy and security [11]. The North American Electric Reliability Corporation (NERC) [12] defines adequacy and security as: •Adequacy – The ability of the electric system to supply the aggregate electrical demand and energy requirements of the end-use customers at all times, taking into account scheduled and reasonably expected unscheduled outages of system components. •Security – The ability of the electric system to withstand sudden disturbances 4 1.1. Context and Motivation such as electric short circuits or unanticipated loss of system elements. Given the current context, this thesis proposes the development of EV models to be included in the adequacy of supply assessment for systems with high level of RES in generation portfolio. This analysis is carried out through the adequacy perspective and the evaluations are performed considering a long-term time horizon [13]. Usually, the adequacy of supply is analysed through the reliability evaluation of the generating systems, which in turn may be divided in two concepts: static and operating reserves. The static reserve assessment is performed by evaluating the difference between the total generating capacity and the total system load. This evaluation produces risk indices which are used to measure the adequacy level of the generating systems. Conversely, operating reserve assessment measures the requirements of the generating systems to cover short-term problems which may result from uncertainties of the RES and load forecast, and forced outages of the generating units. These concepts will be further described in detail. 1.1.1 Power System Adequacy Evaluation The continuous electrical energy supply is affected by random failures of the electrical components. The integration of different types of generating sources to provide electricity to a wide range of customers with varying requirements is another problem that affects the continuous supply of electrical energy, mainly because of the variable behaviour of the primary energy resources. Electric power utilities, therefore, should provide an acceptable degree of system reliability in the planning, design and operation of their systems considering the existing economical constraints. The term “reliability” when associated with a power system is a measure of the system’s ability to meet the customer requirements for electrical energy. Power system reliability evaluation has been extensively developed over the last sixty years mainly focusing on the adequacy perspective and there are many publications available on this subject [14]. 5 Chapter 1 The assumptions over which these approaches were developed are: •The numbers of arrivals counted in disjoint intervals are independent of each other. •The probability distribution of the number of arrivals counted in any time interval only depends on the length of the interval. •The probability distribution of the number of arrivals is a Poisson distribution. •No counted arrivals are simultaneous. The main consequences of these assumptions are: •The probability distribution of the waiting time until the next arrival is an exponential distribution. •The arrivals are distributed uniformly on any interval of time. The advantages of the proposed methodology is that the model can be applied to any mobility pattern since the arrival average parameter is known. Regarding the charging strategies, the assumptions are: •Direct charging: the EV battery charging starts after an occurrence of an arrival. •Valley charging: the EV battery charging can only be made during valley hours. •Controlled charging: the EV battery charging can only be made during valley hours. However, it is assumed that EV battery charging can be controlled. The EV respond to a signal, postponing their battery charging if the operating reserve is threatened. 12 1.4. Thesis Objectives •V2G charging: the EV battery charging starts after an arrival occurs. However, the EV respond to a signal, postponing their battery charging to inject electric energy back to the grid if the operating reserve is threatened. The SOC of the batteries is taken into account. The main hypothesis raised is that an adequate battery charging may contribute to the adequacy of the security of supply in order to mitigate the EV impact on the power systems. This hypothesis will be verified in this thesis throughout the risk indices, which also allow to assess the average load curtailment for each EV penetration scenario, estimated by the simulation process. 1.4 Thesis Objectives The main objective of this thesis is to develop EV models capable of representing the EV charging behaviour in order to measure the EV load impact on the adequacy of the security of supply. In order to reach this objective the following goals were persecuted: •The development of a methodology to build EV models capable of estimating the EV load and chronologically representing its behaviour for different charging strategies, which, in turn, affect the total system load profile. •To estimate the power availability in the batteries that can be injected into the grid (V2G model), considering the battery SOC. •To include the EV models in the SMCS method in order to evaluate the EV impact on the security of supply. 13 Chapter 1 1.5 Thesis Structure This thesis is structured as follows. Chapter 1 presents the context and motivation of the problem under research as well as the assumptions and main objectives to be achieved with this thesis. Chapter 2 presents a background and literature review about the reliability evaluation and the integration of EV in the power systems. In order to draw a framework of this thesis, a relevant discussion about short and long-term risk analysis is presented in Chapter 3. The development of the EV models are divided between Chapters 4 and 5which describe the EV charging strategy modelling and the methodology to include the EV impact on the SMCS method. Chapter 6 presents the results and discussions, through the use of test and real generating systems, about the performance of the proposed EV models. Finally, this dissertation ends with Chapter 7, where the main contributions, conclusions and future work are presented. 1.6 Publication List International Journals •(Chapter 3) - R.J. Bessa, M.A. Matos, I.C. Costa, L. Bremermann, I.G. Franchin, R. Pestana, N. Machado, H. Waldl, and C. Wichmann, ”‘Reserve Setting and Steady-State Security Assessment Using Wind Power Uncertainty Forecast: A Case Study,”’ Sustainable Energy, IEEE Transactions on, 3(4), 827-836, 2012. •(Chapters 4 and 5) - L. Bremermann, M.A. Matos, J.A.P. Lopes, and M.A. 14 1.6. Publication List da Rosa, ”‘Electric Vehicle Models for Evaluating the Security of Supply,”’ Electrical Power System Research, Article accepted for publication on February 2th, 2014. DOI: 10.1016/j.epsr.2014.02.001. Conferences •(Chapter 3) - R.J. Bessa, L. Bremermann, M.A. Matos, R. Pestana, N. Machado, H. Waldl, C. Wichmann, “Reserve and Congestion Management Using Wind Power Probabilistic Forecasting: A Real Case-Study,” EWEA - European Wind Energy Association, Brussels, Belgium, 2011. •(Chapters 4 and 5) - L. Bremermann, M.A. da Rosa, M.A. Matos, J.A.P. Lopes, and J. Sumaili, “Operating Reserve Assessment Incorporating a Stochastic Electric Vehicle Model,” Proceedings of PMAPS 2012 - International Conference on Probabilistic Methods Applied to Power Systems, Istanbul, Turkey, June, 2012. •(Chapters 4 and 5) - L. Bremermann, L.M. de Carvalho, M.A. da Rosa, M.A. Matos, J.A.L. Lopes, “Avalia¸c˜ao do Risco da Integra¸c˜ao de Ve´ıculos El´etricos na Adequa¸c˜ao da Reserva Operacional,” ENRSF - Encontro Nacional de Riscos, Seguran¸ca e Fiabilidade, Lisboa, Portugal, May, 2012. •(Chapters 4 and 5) - I.C. Costa, M.A. da Rosa, L.M. de Carvalho, L. Bremermann, J.P. Iria, “Identifying Benefits Between the Integration of Electric Vehicles and Renewable Power Usage,” PEOCO 2014 - IEEE 8th International Power Engineering and Optimization Techniques , Maro, 2014. •(Chapters 4 and 5) - L. Bremermann, M.A. da Rosa, M.A. Matos, J.A.P. Lopes, L.M. de Carvalho and I.C. Costa, “Adequacy of the long-term operational reserve of a system with wind power and electric vehicles under severe scenarios,” Accepted to PMAPS 2014 - International Conference on 15 Chapter 1 Probabilistic Methods Applied to Power Systems, Durham, England, July, 2014. •(Chapter 5) - I.C. Costa, L.M. de Carvalho, L. Bremermann, M.A. da Rosa and J. Soares, “Stationary batteries performance assessment to deal with renewable intermittency,” Accepted to PMAPS 2014 - International Conference on Probabilistic Methods Applied to Power Systems, Durham, England, July, 2014. 16 Chapter 2 Background and State of the Art 2.1 Introduction This chapter presents a literature review about the adequacy of security of supply and the impact of EV integration in the generating system. The EV deployment will create an additional electric load that might require more generating installed capacity. However, active demand side management can be provided by an aggregation entity that is able to control the EV charging rate or even postpone the battery charging to hours where the conventional load is lower. Moreover, this entity can also manage the charged batteries of the EV in order to contribute with power injection into the grid (V2G concept) [28]. A sustainable electric system must exploit these control opportunities in order to increase the integration of RES without compromising the adequacy of the security of supply. The integration of EV on the power system analysis has been addressed in the literature mainly through two different paths: EV as uncontrollable and 17 Chapter 2 controllable load. The integration of EV as uncontrollable load has generated negative effects in the power system operation mostly due to the increase in the load consumption in peak load hours. Under a controlled EV battery charging, the EV becomes a flexible load capable of storing electrical energy. The EV integration under a controlled scheme has been seen as a positive way to contribute to the operational reserve, improving RES integration and avoiding, at least momentarily, system reinforcement. In the long-term planning perspective, the monitoring of the adequacy of security of supply should ensure that the installed generating capacity meets the load forecast and the generating unit outages with an adequate level of risk. This risk is assessed through the static reserve evaluation, which consists of measuring the system balance. A complementary assessment, the focus of this thesis, relies on the long-term evaluation of the operational reserve, which is concerned with the flexibility of the generating systems to cope with the short-term uncertainties. 2.2 Regulation Reserves The SO has the responsibility of managing the balance between generation and load to ensure a supply of energy to the final consumer with quality and continuity. Due to the removal of facilities for regular scheduled maintenance, unexpected generating unit outages and the uncertainty related to the load forecast, a reserve level must be kept in order to ensure an adequate and acceptable continuity of supply during the events with capacity shortage. In terms of short-term system operation, the effect of these events is translated in frequency imbalances and, generally, these imbalances are solved automatically and/or by giving set points to the generating units and flexible loads. The EV, under adequate charging schemes, represents a flexible load capable of contributing to the systems. This thesis will exploit this concept in Chapter 5. 18 2.2. Regulation Reserves The European Network of Transmission System Operators for Electricity (ENTSOE) defines the reserve levels as follows [29]: •Frequency Containment Reserve (FCR) - The FCR aims at stabilising the frequency after a system disturbance. For such task, synchronised generating units must respond to this imbalance. •Frequency Restoration Reserve (FRR) - The FRR is responsible to offset the frequency deviation caused by the system disturbance. While the FCR acts to stabilise the frequency, the FRR corrects the frequency deviation to the nominal value considering one or more load frequency control areas. Generally, the FRR is given by an automatic generation control (AGC) system, which consists of setting the operating points of the generating units or change the state of flexible loads. •Replacement Reserve (RR) - The RR is usually activated to reset the FRR level. Generally, the RR amount is composed by generating units that are able to start up quickly. The reserve levels aforementioned are related to the short-term operation of the system. The SO should define the reserve levels required in a market environment via suitable operating planning. The increasing integration of RES has created an additional interest on the performance assessment of the operational reserve in a long-term time frame [13]. The reserve that is spinning, synchronized and ready to take load up, is generally known as spinning reserve. When quick start generating units, such as gas turbines and hydro plants, interruptible loads and assistance from interconnected systems are taken into account and added to the spinning reserve, then the total capacity is known as operational reserve. This long-term evaluation measures the adequacy of the security of supply considering that the operating reserve should meet the uncertainties related to the load and RES forecast errors (mainly wind power) and the generating unit 19 Chapter 2 outages. Hence, this thesis also focuses on the probabilistic assessment of the operational reserve in a long-term perspective. The framework of this analysis will be described in Chapter 3. 2.3 Adequacy Assessment Methods The adequacy assessment of the security of supply can be performed using either deterministic or probabilistic methods. Deterministic methods have been widely used in the past and consider only specific configurations of the system which ignore the stochastic and probabilistic nature of the system’s components. The probabilistic methods can provide risk values taking into account several operation scenarios like, for instance, the probability of the forecast load becomes greater than the generating capacity. The probabilistic methods are, usually, divided into two methods: analytical and simulation. 2.3.1 Reliability Indices The main outcome of the probabilistic methods used to assess the adequacy of the security of supply is the reliability indices. The reliability indices have different designations regarding the system’s hierarchical level (HL) involved in the adequacy assessment [30]. This thesis is concerned to the adequacy evaluation of generating systems, which is usually known as HL-1 system. Reliability indices are, generally, categorised as probability, energy, frequency and duration indices [30]. This thesis uses the traditional reliability indices, in order to analyse the adequacy of the security of supply, which are: •Loss of Load Probability (LOLP) - this index gives the probability of the load curtailment. •Loss of Load Expectation (LOLE) - this index represents the average of load 20 2.3. Adequacy Assessment Methods curtailed during the evaluation period. It can be expressed in hours/year, days/year or weeks/year. •Expected Power Not Supplied (EPNS) - this index represents the average power curtailed during the evaluation period. It is expressed in MW. •Expected Energy Not Supplied (EENS) - this index represents the average energy curtailed during the evaluation period. It is expressed in MWh/year. •Loss of Load Frequency (LOLF) - this index gives the average number of load curtailment events during the evaluation period. It is expressed in occurrences/day,occurrences/week or occurrences/year. •Loss of Load Duration (LOLD) - this index represents the average duration of load curtailment events during the evaluation period. It can be expressed in hours/occurrence,days/occurrence or weeks/occurrence. 2.3.2 Analytical Methods Analytical techniques represent the system by a mathematical model and calculate by obtaining the probability mass function of the system states. Equation (2.1) presents a general formulation to calculate a given reliability index. E[G(X)] = X x∈A G(x)p(x) (2.1) where xis the current state of the random variable X,Ais the set of all system states, p(x) is the probability of the system state x,G(x) is the outcome of the test function H, which is a mathematical formulation of a given reliability index, for the system state x.E[G(X)] is the calculated reliability index. The main advantage of analytical methods is that they, usually, provide reliability indices in a relatively short computing time. However, the analytical model of the system requires the use of assumptions in order to simplify the problem. This is the case of almost all 21 Chapter 2 Figure 2.1: Flowchart of the Sequential Monte Carlo Simulation process. the component’s state duration and is calculated as follows: Ti=−1 αi ln(Ui).(2.4) Where Uiis a uniformly distributed random number between [0,1], istands for the component number. The MTTF and MTTR values are represented by α, and are used according to the current system’s state. The load transitions occur in an hourly basis with 8760 load points. 4. Update the simulation clock t, according to the selected state transitions. 5. In order to obtain yearly reliability indices, evaluate the test function over the accumulated values. 6. Update the outcome of reliability test functions and the corresponding indices. 28 2.3. Adequacy Assessment Methods 7. If the simulated year is not in the end, then return to step 4. Otherwise, go to step 8. 8. Estimate the expected mean values of the yearly indices as the average over the results for each simulated sequence. 9. Test the stopping criteria according to their definitions in the beginning of the simulation process. Usually, the number of sampled years and the convergence index βare the selected criteria to end the simulation process. 10. If the stopping criteria is not reached, then, repeat step 2 each time span and record the results of each duration sampled for all components. Otherwise, go to step 11. 11. End the process if the desired degree of confidence is achieved. If not, return to step 2. The advantages of the SMCS are: •It can easily calculate the actual frequency index. •It considers any state duration distribution, exponential or non-exponential distributions. •It is the only method able to calculate the statistical probability distributions of the reliability indices in addition to their expected value. •The method is also able to represent hydro and wind series chronologically. This series bring an important season component that affects the power output of such generation technologies. The SMCS method is the one used to evaluate the adequacy of the generating systems performed in this thesis. Therefore, more attention is given for this simulation method in Chapter 3. 29 Chapter 2 2.3.4 Hybrid Methods Hybrid methods have been developed to improve some characteristics of classic methods, and since simulation methodologies have already a complete and realistic approach, the main objective is to improve performance. Even so, the reduction of computational effort originates an almost unavoidable loss of information and of its quality. Analytical/Simulation Methods Reference [47] merges the Monte Carlo methodology with the LOLP technique for evaluating the reliability of an hydrothermal generating system. This approach takes into account the effect of reservoir depletion on the output capacities of the hydroelectric units. The simulation method is used to acquire the energy states whilst the analytical technique calculates the reliability indices. An alternative hybrid method, presented in reference [48], still uses an analytical/simulation scheme, in which the recursive method is used to incorporate wind generation and hydro depletion through its hourly usage over an year interval. The presented method has the aforementioned advantages of the analytical methods, but loses, for instance, the possibility of computing the probability distributions of the reliability indices, as computed in the SMCS method. The discussion of a new way of examining the probability distributions of the reliability indices based on an hybrid method is presented in reference [49]. The test system includes the EV impact on such analysis. The EV representation follows the basic wind power representation, i.e., they are added as positive capacities on the system load. 30 2.3. Adequacy Assessment Methods Simulation/Simulation Methods The quasi-sequential Monte Carlo simulation is one of the methods that combine different concepts from the Monte Carlo theory. Reference [50], presents a method based on the NSMCS. The system’s components’ state are sampled according to the state space representation. However, the chronology of the load is kept through the use of the multilevel non-aggregate Markov load model, instead of the traditional multi-state Markov load model that transforms the hourly chronological peak load levels into a state space representation. The quasi-sequential MCS creates a connection with the chronology aspect, by sampling the availability of the system components for each load level, which allows the inclusion of other time-dependent characteristics, like the capacity fluctuation of generating units or scheduled maintenance. A different load model (Multi-Level Non-Aggregate Markov Load Model) is able to restore the chronological aspect and is implemented through a pseudo-chronological simulation [44]. The purpose of this approach is to use the non-sequential method to select the failure states of the system, and the sequential method is only used when there is a complete interruption of the system. The pseudo-sequential Monte Carlo simulation is described in detail in reference [44]. Despite being one of the most robust methods for power system analysis, especially for large systems, the Monte Carlo methods retain considerable difficulties in the probing process of very rare events. Pseudo-Sequential Monte Carlo Method [51] combines the state system sampling of the NSMCS with the chronology simulation of the SMCS, processing only the failure sequences. Prior to the application of the method, a considerable amount of yearly sequences is simulated using a similar process to the one used in SMCS. A small difference occurs on Combined Pseudo-Sequential and State Transition Method [52], sequential simulation is processed through System State Transition Sampling, which leads to a lower computing time and a loss in the capture of some chronological events. 31 Chapter 2 In order to suppress this problem, a Cross-Entropy based Monte Carlo method is proposed in [53]. Several reduction techniques have already been applied to power systems, with some presenting better results than others in what regards real power systems. However, rare events were still a problem deserving little discussion on reliability related literature. The proposed method in [53] serves as an optimisation algorithm for the selection of distorted parameters. The main idea is to use an auxiliary importance sampling density function, whose parameters are obtained from an optimization process that minimizes the computational effort of the MCS estimation approach. The Cross-Entropy concept is further applied to the SMCS method in [54] in order to evaluate generating capacity reliability indices. A slightly different optimisation process (still based on Cross-Entropy Method) is used. The proposed methodology suitably modifies the chronological evolution of the system in order to improve its statistical efficiency and convergence properties. In order to decrease the computational effort, the Sequential Population Based Monte Carlo Simulation method is proposed in [41]. This method aims to create a generating states list, in which the total capacity is lower to the system’s peak load, using the PBM. Then, the generation state sampled from the SMCS method are compared to those from generated list. Instead of performing the traditional SMCS procedure of composition to test the G−L≤0 to all sampled generation states, the list created in the first phase is used to identify which of them will proceed to the SMCS composition and evaluation stage. Basically, the insight was coding the generation states to build a vector with generating units that has the same capacity and the same stochastic parameters. Therefore, each entry of this vector is an integer value between zero and the number of “equal” generating units. The detailed pseudo-codes of the first and second phases can be found in [41]. The reduction of the computational effort is given, mainly, because additional time to compose and evaluate all system states is avoided. 32 2.3. Adequacy Assessment Methods 2.3.5 Load and Electric Vehicle Load Modelling Conventional Load Modelling Usually, load models are developed through historical observation of the electrical demand. Depending on the power system analysis method and the necessary accuracy of the load representation, four models may be used to obtain the reliability indices. The load model is combined with the COPT or assessed by the simulation techniques. These representations are transversal for the aforementioned methods, however its usage depends on the accuracy needed. Figure 2.2: Representations of the single load curve. Figure 2.2 presents four different load models. One considers the annual peak load for all load cycle. This means that a constant load is compared to the generation model and then, the LOLP index is determined. This method introduces excessive errors leading to a too conservative approach (see Figure 2.2 a)). Another methodology, but still conservative, is the linearisation of a load diagram 33 Chapter 2 using maximum and minimum load points. This method could not capture the frequency, since, transitions disappear with the linearisation approach (see Figure 2.2 b)). Other technique is based on the daily or hourly peak demands re-ordering them to obtain a descendent curve (see Figure 2.2 c)). Nonetheless, this approach loses the chronological behaviour of the system demand. The most common representation to index evaluation is modelling load in an hourly resolution, that is composed by load constant steps through the complete period (see Figure 2.2 d)). This model, makes possible to build a load representation with the same parameters as the generation model [31]. The load representation must be well defined, mainly because different load model representations lead to different reliability index meanings. For instance, a LOLE of 1.0 days/year obtained through the peak load demand in a daily resolution, which corresponds to a 365 points in a year, means that the generation capacity is not sufficient to meet the peak load demand in an average of 1 day in 365 days of the year. On the other hand, a LOLE of 1.0 hours/year obtained through the peak load demand in an hourly resolution, which corresponds to a 8760 points in a year, means that the generation capacity is not sufficient to meet the peak load demand in an average of 1 hour of the 8760 hours of the year. Electric Vehicle Load Modelling The EV brings some different approaches to modelling EV load. Reference [55] presents a methodology of optimizing power systems demand due to EV charging load. The EV load is calculated a priori and added (in a distributed way) to the conventional load points, maintaining it fixed during the process. The results demonstrate that EV charging load has significant potential to improve the daily load profile of power systems if the charging loads are optimally distributed. The SMCS method is used in [56] to evaluate the effects of different EV types, locations and penetration levels on the IEEE-Roy Billinton Test System (RBTS) Bus-6 distribution test system. More references will be given in the next section which 34 2.3. Adequacy Assessment Methods is related to the literature review of the EV. This thesis presents some EV models based on the mobility patterns and expected EV charging behaviour. Characteristics such as travel distances, battery SOC, and average time from parked vehicles are also addressed in such approaches. The resultant EV load is added to the conventional system demand to be evaluated. Then, the battery charging strategies define the EV load profile. If no charging control is provided, the EV load increases the conventional load amount requiring more capacity or operating reserve from the system. On the other hand, if charging control is provided, the EV load becomes a flexible load with storage capacity, which supports the opportunities that will be discussed and addressed throughout this thesis. 2.3.6 Wind Power Modelling Due to the increasing integration of wind power generation in the grid, it has been addressed in the probabilistic methodologies of adequacy assessment. Wind Power modelling for Analytical Methods References [57, 58] have addressed the wind power generation in analytical methods. Reference [58] divides the overall system into two subsystems, containing the conventional and wind units, and a generation system model is built using a Recursive Algorithm for each of these two subsystems. The power output of the wind subsystem is calculated for each hour under study and a vector containing the hourly output of the wind power generation unit subsystem is created. The probability model of the wind generation subsystem is modified to take into account the effect of the fluctuating energy generation. Then, the two subsystems are combined to calculate the LOLE for the desired hour. Therefore, the reliability 35 Chapter 2 index for the entire period is computed by the summation of all hourly values of LOLE. Wind Power modelling for Simulation Methods In simulation methods [13,59–61], the wind power has been modelled through the multi-state Markov model, which is used to represent the failure/repair cycle of the wind farm’s generating units. Usually, this model addresses the transitions between states as following an exponential distribution and calculated by Equation (5.1). The wind variability produces fluctuations in the power output of wind turbines, which is also represented through wind power series (generally in a hourly basis). These series capture the production of the wind farms in percentage of their total capacities. Therefore, the maximum capacity of a given wind farm is multiplied by the correspondent value of the wind power series, according to the simulation time, and taking into account the generating units’ state of the wind farm, in order to produce the total wind power generation amount. 2.4 Related Studies on Electric Vehicles Historically, the electric vehicles appeared in the beginning of the 20th century. Reference [62] presents a discussion about the use of EV to correct load factor during the valley period. Curiously, the advantages of the EV had been compared with the use of horses. Instead of occupying the street and having to spend money to feed the horses, the EV owners could charge the EV batteries cheaper, during the night. Nowadays, it is expected that in a near future the EV deployment will take the place of ICE vehicles. Therefore, the EV assessment has been included in general power system references. 36 2.4. Related Studies on Electric Vehicles Market Environment Recently, the technological improvement on the batteries and elements of the network, concepts as controlled/smart charging and V2G are discussed. In this context, the EV are able to provide electrical energy to the systems. The accumulated energy from a large group of vehicles, can be bid in the market through an aggregation entity. The reference [63] presents equations developed for calculating the capacity for grid power from three types of electric vehicles (hybrid, battery, and fuel cell vehicles). These equations are applied to evaluate revenues and costs for vehicles that are used to supply electricity to three electric markets (peak power, spinning reserves, and regulation). The results suggest that vehicles probably will not generate bulk power, both because of their fundamental engineering characteristics and because their calculated per kWh cost of energy from vehicles is higher than bulk electricity from centralized generators. A commercial value of V2G for ancillary services is analysed in reference [64]. It is described the infrastructure considered to support the integration of EV on the distribution level. The methodology presented is used to model and analyse the load demand in a distribution system due to EV battery charging. It is stated the random characteristic of the EV behaviour and the load demand is calculated taking the SOC into account. Furthermore, the engineering rationale and economic motivation for V2G power are compelling. The societal advantages of developing V2G include additional revenue stream for cleaner vehicles, increased stability and reliability of the electric grid, lower electric system costs, and eventually, inexpensive storage and backup for renewable electricity. According to [65] to properly estimate the cost of charging/discharging EV battery, it is necessary not only to assess the operational costs, but also the impacts of the new consumption patterns in the development of long-term power plant portfolio. The impact of EV over the market perspective is also presented in reference [28], 37 Chapter 3 what cost?” This issue is a widely recognized problem and several criteria and techniques have been developed as an attempt to answer that. Those first used were mainly deterministic based, where typical deterministic criteria are: •Planning generating capacity: installed capacity equals the expected maximum demand plus a fixed percentage of the expected minimum demand. •Operating capacity: spinning capacity equals expected load demand plus a reserve which is equal to one or more largest units. Some of that are still used in planning phase studies, however, the essential characteristic of deterministic criteria is that it does not account for the stochastic nature of the system behaviour, of customer demands and of component failures. Typical probabilistic aspects are: •Forced outage rates of generating units. •All planning and operating decisions are based on load forecasting techniques. Then, the uncertainties are inherent to the forecast methods which cannot be characterised in deterministic criteria. Combining conventional generation, unconventional generation with forecast properties and consumption variability has made the task of fitting large amounts of wind generation into unit commitment procedures even more complex. Power system planners and operators are already familiar with a certain amount of variability and uncertainty, particularly because variability and uncertainty are related to the system demand. Assuming that the output from wind generation is not as dispatchable as conventional sources, the level of unit commitment uncertainty is increased, consequently, making the task of setting reserve levels more challenging [80–82]. 44 3.1. Introduction Risk-based methodologies such as the PJM method [83] are adequate to assess short-term unit commitment risks considering intervals up to a few hours. Such evaluation is conditioned to a short period of time, and it is essentially dependent on the quality of load and wind forecasts. Usually, these short-term concerns have been seen as a way of controlling the amount of spinning reserve, providing operators with information on operation system risks, taking the generating units available at the operation moment into consideration. For the medium and long-term assessment, the risk evaluation must account for the available system capacity performance [13, 60] in order to meet the expected demand growth, assuring that investment options will result in more robust and flexible generating configurations that are consequently more secure. From a technological perspective, the design characteristics of conventional hydro and thermal generators already enable the generating units to contribute to system support services, such as voltage and frequency regulation [84]. Recently, new technologies have been massively connected to the system, such as: wind and solar power. Although, the current technology allows providing a certain level of ancillary services, the level of wind unpredictability is still significant and it is not capable of fully providing the same system support as hydro and gas technologies. Moreover, these inherent unpredictable and volatile characteristics of the wind impose additional requirements to the reserve level. Firstly, the necessary reserve to deal with the uncertainty that comes from the wind production may increase due to the fluctuating characteristic of this primary resource. Secondly, this fluctuating characteristic may also require more flexible conventional generators (hydro and combined cycle gas turbines) in order to cope with system support services [84]. To deliver both flexibility and system support services, large conventional plants could be desirable, however they usually require expensive investments [85]. 45 Chapter 3 Assessing reserve requirements to ensure an adequate level of energy supply is an important aspect for both expansion and operation planning of the generating systems. In the past, the planning phase concern was related to prepare the generating systems to meet the long-term load forecast, whereas the operating phase concerns were related to dealing with short-term load forecast, where sufficient generation should be scheduled in order to account for load uncertainties and sudden loss of generating units. With the massive usage of wind power technology, another set of uncertainties have been introduced on planning and operating phases. From the short-term reserve evaluation perspective, the uncertainty linked to the wind power fluctuations brings huge difficulties to the unit-commitment and dispatch procedures of the generating systems. From the long-term reserve evaluation perspective, the uncertainty linked to the massive usage of wind power makes it difficult to prepare the future generating systems so that they can deal with large levels of uncertainty (mainly wind power and load forecasting errors) and thus meet the load forecast for the future. In summary, the main concern of short-term reserve evaluation is measuring the unit commitment risk level. Complementary, the decisions on long-term generation management are essentially around reinforcing bulk generation. In fact, it is a common knowledge that increasing the participation of renewable power, mainly wind power, in the total generation mix means that operating and planning methodologies and standards must be revisited [86]. As the scope of this thesis is within the adequacy of power systems context, this chapter discusses the short and long-term risk analysis of the generating systems. The discussion is conducted through the use of two approaches: an analytical technique that performs an evaluation on unit commitment risk (short-term), and a SMCS method, which assesses the performance of the long-term operating reserve [13]. This chapter is organized as follows. Section 3.2 introduces short-term operating 46 3.2. Short-term Reserve Evaluation reserve evaluation. Section 3.2.1 presents the modelling of the generation unit outages. The demand and wind power forecast uncertainty models, under the short-term perspective, are presented in Section 3.2.2. The long-term operating reserve capacity evaluation is introduced in Section 3.3. Section 3.3.1 presents the modelling of the generation unit outages under the long-term assessment perspective. The long-term demand representation is addressed in Section 3.3.2. Section 3.3.3 presents the modelling of the wind power uncertainties for the long-term assessment. 3.2 Short-term Reserve Evaluation One of the first methods that included the idea of risk to calculate generating reserve was the PJM [83]. The basic aim is to evaluate the probability of the committed generation to meet or fail to meet the expected demand during a period of time [30]. The PJM method is rooted in short-term concerns where the main uncertainties involved are load forecast errors and forced generating units outages. The concepts are based on the assumptions that failures and repairs are exponentially distributed. The measurement obtained is a system risk index that outlines the probability which the existing generation capacity has of not satisfying the expected load demand, during time period T(lead time) and/or the probability of the operator not reacting to replace any damaged unit or using new ones [30,60]. Therefore, the index represents a measurement of the loss of load associated with the scheduled generating reserve [30,60]. For a single unit, the probability of failure at interval [0, T], i.e.Pdown(T), can be calculated by Pdown(T)≈P(tup ≤T) = 1 −e−λT (3.1) where λrepresents the failure rate of a given generating unit. If T << 1, for short 47 Chapter 3 lead times up to some hours, then Equation (3.1) becomes Pdown(T)≈λT = ORR. Consequently, it is also possible to build an analytical generation model [30], with high efficiency and mainly compatible with the operating expectations in terms of time response. Figure 3.1, shows two possible commitments in accordance with their technology predominance. The use of a merit order, to represent the generation commitment, is a common practice on the adequacy evaluation of the generating system. Even though, the generation and load are distributed over the power system network, the merit order, in a single bus representation, is based on the generating unit cost and technology in order to give a guidance of the dispatch procedure. Figure 3.1: Unit commitment representation based on technology predominance. Currently, to deal with variable generation from the operating perspective two different categories of generating systems were identified from the technological point of view: predominantly thermal generation and predominantly hydro generation. In the predominantly thermal system, the commitment of generating units starts 48 3.2. Short-term Reserve Evaluation using less flexible technologies, such as Coal and Nuclear turbines (see Figure 3.1). These generating units deliver inertia (stability) to the system and meet a major portion of the system load. However, these units are running in the low uncertainty zone (see Figure 3.1) of the unit commitment basis due to the low costs and their inability to deliver flexibility. In the predominantly hydro system, the commitment of generating units starts using large hydro plants which have fast response to ensure system stability. This system has high level of flexibility, however the variability of its primary resources affects the produced power requesting more installed power capacity. The reserve capacity, which is synchronized (spinning) to take load up, is based on the generating units with more flexibility, such as gas and hydro turbines, lying in the high uncertainty zone (see Figure 3.1) of the unit commitment basis. From a generation assessment perspective, it is usual not only to consider the synchronised units in their assessment, but non-synchronised units such as hydro and gas turbines as fast tertiary reserve. Furthermore, for both examples and from the operating reserve evaluation perspective, the challenge is to verify whether these generating units are enough to deal with fast and enormous power variations. Usually, these power variations require a certain level of quick generation response, which are addressed by hydro and gas technologies. The flexibility of the generating system is also an issue that should be evaluated in the planning phase of the generating systems. In this thesis, flexibility is considered the feature ascribed to a system capable of accommodating generation variation. Nowadays, there are a considerable number of flexibility sources to deal with the variability of RES [12]. Hydro predominant systems, generally have enough flexibility to integrate large amounts of variable wind power. In these cases, the main concern is to coordinate hydro and wind power in order to avoid loss of wind production, since water is storable whereas wind is not. Procedures based 49 Chapter 3 on water storage involving pumping facilities by wind power during convenient operation hours is one of the solutions applied to coordinate hydro and wind production from the operating perspective. Once the system technology category is defined, unit commitment or dispatch procedures may be organized as operational decisions. In general, the operational decisions involve technical constraints related to the reserve needs and economic issues. Depending on the method used to define the level of the synchronized reserve and/or fast tertiary reserve, these operational decisions may lead to overscheduling, which could be more reliable but also costly, or may lead to underscheduling, which could be more affordable and yet unreliable. Analytical-based approaches are essentially mathematical formulations based on enumeration methods. The aim of these approaches is to calculate probability density functions using generation and load risk models. In general, the reserve risk model is built considering system generation and load as independent variables. Therefore, the reliability indices are calculated through a simple mathematical manipulation. This type of approach is mostly applied for operational purposes due to its computational efficiency and simple implementation. In this context, the following sections will present an analytical method to model generating unit outages as well as the usual approaches to model load demand and wind power forecast uncertainties. 3.2.1 Modelling Generating Unit Outages One of the analytical method advantages is to reduce the modelling dependence, where the stochastic behaviour of system’s components is defined through mathematical enumerative procedures. The system load and wind power forecasts are, in this context, treated independently through dedicated systems and consequently modelled outside of the analytical model. The system risk index is calculated using a Capacity Outage Probability and Frequency Table 50 3.2. Short-term Reserve Evaluation (COPFT) [53]. Another issue regarding this type of evaluation is related to the Outage Replacement Rate (ORR) parameter, which is similar to the Forced Outage Rate (FOR) [30] used in planning studies. The main difference between FOR and ORR is that the latter is not simply a fixed characteristic of a unit, but it is a time-dependent quantity affected by the value of the lead time considered. Hence, it is possible to build a generation model equal to the capacity and frequency outage probability table [53] in order to assess risks on hourly basis. While the COPFT is built, it is possible to follow an intuitive process based on decoupling G(system capacity) in different subsystems, mainly to convolve all stochastic capacities Gand L(system demand) at an appropriate moment of the evaluation. The information linked to resource fluctuation, such as water inflows or wind variability, is complementary to the stochastic model and, most of the time it is applied to the calculation of G[49]. To cope with short-term concerns, a COPFT is built including all committed generating units, which follows a two-state Markov Model, resulting from simple information on capacities (cg), probabilities (pg) and frequencies (fg) linked to the unit commitment decision, represented as follows: G={cg, pg, fg}(3.2) Time-dependent power sources, such as wind power, are rarely included in the conventional generation model. Although wind turbines behave as hydro or thermal units from the stochastic point of view, the key factors linked to wind capacity are wind speed and direction, which behave differently, for instance, from hydro depletions. Usually, wind power generation, addressed on short-term operating reserve evaluation, is a quantity provided by wind power forecast tools and, at the moment, it is considered as a complementary input in unit commitment risk evaluation. In the same way, the value of the system load Lis also provided by load forecast 51 Chapter 3 tools, and it is considered as a complementary input in unit commitment risk evaluation. Consequently, the reserve model will be an appropriate combination of generation model Gwith system load Land system wind power generation, in order to set the spinning reserve based on the pre-established risk criterion. Most importantly, this pre-established risk criterion should be followed in accordance with the operator experience. 3.2.2 Modelling Demand and Wind Power Forecast Uncertainties Usually, conventional generation models do not comprise generation variability, like wind power. While large hydro and thermal power plants have no significant variability in their power production, the wind power generation varies with its primary resource, which in turn is different throughout time. The variations of the wind power are often represented through a sequence of percentage values (wind series), and generally, it is given in the same basis resolution of the load, e.g., 8760 capacity points. The wind series represent the seasonal wind behaviour and different annual scenarios. The probability indexed in each series gives the occurrence chance of each one. In order to combine the probabilities and frequencies of the wind power with conventional generation ones, the wind series of each wind farm is converted into impulses to perform a convolution. At least, three different manners were identified to consider the wind power’s effect on reserve evaluation [31]: •The wind capacity of each wind farm is convolved with the conventional generation. •The summation of the total wind farm capacities is convolved, in an hourly basis, with the conventional generation. •The wind capacity of each wind farm is added, in an hourly basis, as a 52 3.2. Short-term Reserve Evaluation negative load on the load points. The latter case follows an ordinary procedure of the load model, which is described in [31]. During the operation, one has to make decisions in accordance with the last information on what has occurred in the system, and taking the remaining load and wind power forecasting uncertainties into account. The demand forecast uncertainty, close to the real-time operation (up to 2 hours), is usually negligible, while the wind power forecast uncertainty is still relevant. The Normalized Mean Absolute Error (NMAE), of the very-short wind power forecasting varies from 1-2% (10-15 min. ahead) to a maximum of 10% around 2 hours ahead, considering a time resolution of 10 min. However, it is hard to beat the persistence model performance for such a short time horizon [87]. When the time resolution increases to up to 1 hour, the improvement with respect to the persistence model is even more difficult, so it is a common practice to assess unit commitment risk using the last known wind power occurrence as the forecast for the next period, given by c Wi(t+k) = Wi(t) (3.3) where, Wi(t) is the last known wind power generation at instant t, and c Wi(t+k) is the forecast wind power generation launched at instant tfor the horizon k. Note that, the wind forecast error, from a short-term reserve evaluation is still hard to define. Therefore, the persistence method provides guidance when assessing unit commitment risk. 3.2.3 A Simple Example From an operating perspective, one of the main concerns of a system operator is regarding the composition of the spinning reserve. This decision is usually made according to economic aspects combined with a unit commitment risk 53 Chapter 3 uncertainties [84]. Due to the uncertainty of the system load forecast and the variable primary resources, the task of evaluating the operating reserve capacity using probabilistic methods is advisable [30]. To cope with uncertainties from a planning perspective, the Monte Carlo simulation methods are still the standard to assess the adequacy of power systems. Besides that, the SMCS, specifically, has the advantage of providing probability distribution functions associated to reserve requirements related to a set of uncertainties linked to the power balance problem. The SMCS makes keeping track of several features related to the operating history of system states possible. Its flexibility also makes modelling uncertainty details possible, which is a valuable information for generating systems with large renewable sources in their energy mix. After evaluating each system state, performance indices are estimated using the expected value equation. As described in [15], let Qdenote the unavailability (failure probability) of a system and xibe a zero-one indicator variable which states whether or not a value equation is true. The estimate of the system unavailability is given by b E[G] = 1 N N X u=1 G(yu) (3.11) where yuis the sequence of system states in year u,G(yu) is the reliability test function evaluated at yu,Nis the number of simulated years (samples) and Gis the random variable which maps G(yu) values. The uncertainty surrounding the estimated indices is given by the variance through V(b E[G]) = b E[G]−b E[G]2 N(3.12) The stochastic process convergence is tested using the coefficient of variation βas follows. β=qV(b E[G]) b E[G]·100% (3.13) 60 3.3. Long-term Reserve Evaluation Following the SMCS procedure, the conventional system reliability indices may be estimated. This traditional view can provide important information on loss of load events and, in this case, the simulation will monitor the success and failure states of the static and operating reserves, where the following reliability indices will be calculated [15]: LOLP, LOLE, EENS, LOLF and LOLD. In order to illustrate the estimate of a reliability index, using Equation (3.11), the formulation of the LOLE index is exemplified. Let the LOLP of a given state xbe described as LOLP(x) = (1, if xis FAILURE 0, if xis SUCCESS.(3.14) The LOLE function can be written by GLOLE(yu) = 1 8760 X n∈S d(xi)LOLP(xn) (3.15) where xis a given state nof the system states set Sin a year u.d(xu) is the duration of a given system state and 8760 is the period under analysis. Therefore, following the Equation (3.11) the LOLE index can be estimated as follows LOLE =1 N N X u=1 GLOLE(yu).(3.16) The LOLE is the average number of hours in a given period (in this thesis this period corresponds to one year) in which the hourly load is expected to exceed the available generating capacity. By using the same concept presented in Equations (3.11), (3.12) and (3.13), the uncertainties surrounding the operating reserve concept might be monitored to investigate, in detail, the uncertainty impacts on the performance of the operating reserve from a long-term perspective. The following sections will introduce all of these deviations as random variables. Although the LOLE index does not indicate the severity of the deficiency nor the frequency nor the duration of loss of load, it is the most widely used probabilistic criterion in generating capacity planning studies [15]. 61 Chapter 3 3.3.1 Modelling Generating Unit Outages The SMCS method is generally used to accurately reproduce the entire cycle of generating unit outages [13]. The failure/repair cycle of the generating units is represented by two-state and multi-state Markov Models (as seen in Figure 3.3) and their transitions are usually approached by an exponential probability distribution [13,60]. Figure 3.3: Markov Model representations. The representation of hydro and thermal power plants follow, in general, the twostate Markov Model, as illustrated in Figure 3.3 a). Using a simplified model, hydro plants have their maximum output multiplied by their corresponding value of the hydro series [13, 60]. The hydro series are the available volume of water in the reservoir for each hydro plant region. This information can be given as a monthly average value, which will represent the seasonal water variability or even in a smaller resolution as a weekly basis. This latter, makes introducing a set of rules to the scheduling procedure possible in order to guarantee a maximum energy usage of the hydro units, for instance. The residence time of the hydro and thermal plants representation in each state 62 3.3. Long-term Reserve Evaluation is calculated according to T=−1 αln(U) (3.17) where Tis the residence time of each generating unit and αassumes λ, which is the mean time to failure (MTTF), if the current state is “Up” or it assumes µ, which is the mean time to repair (MTTR), if the current state is “Down”. Uis a uniformly distributed random number which is sampled in the interval [0,1]. Instead to model the wind power capacity to convolve with the load (as showed in Section 3.2.2), here, it is modelled to convolve with the conventional generation. The capacity of wind turbines might be represented by a multi-state Markov Model (as illustrated in Figure 3.3 b), to characterise the failure/repair cycle of a wind farm. Then, the capacity associated to the kth state is given by: Ck= (N−k)C, k = 0,1, ..., N (3.18) where Cis the capacity of a single generating unit. Nis the number of wind farm’s generating units and kis the generating unit state. With the objective to reduce the number of states during the SMCS procedure, a simple truncation process defines the desired order of accuracy. Therefore, instead of N+1 states, a smaller number up to the capacity CLwill limit this model [43]. Similarly to the hydro depletion, the maximum output of a wind turbine is multiplied by the corresponding value of the wind series. 3.3.2 Modelling Demand Forecast Uncertainty The seasonal characteristic of the load shapes the load profile throughout the year. For instance, due to the high temperature of the summer season, the electrical load is impacted mainly by air conditioners, whereas the electrical load in the winter season is characterised by electric heating resulting in different load profiles. In order to achieve good results in the reliability assessment of generating systems, the detailed modelling of the load is desired, however this accuracy depends on the amount and the quality of available data [45]. 63 Chapter 3 The load model, usually consists of 8760 load steps. In order to track the chronological characteristic of the load profile, a state space representation can be approached by Markov model. Some models keep the chronological aspects of the load tracked, contributing to the reduction of the computational effort, however the SMCS method sequentially follows these load steps during the simulation procedure process. The error between the load forecast and the actual load is included in the load modelling, as showed in Equation (3.19). From this definition two uncertainty levels may be represented: short and long-term load forecasting errors, which can be simulated through the SMCS process. In the short-term representation, an hourly uncertainty is calculated during the simulation, whereas the long-term load uncertainty is calculated once a simulated year. The latter, causes an effect over all load profile while the short-term load forecasting error inserts a noise in the chronological representation of the load. La(t) = Lf(t)+∆L(t) (3.19) where Lf(t) is the forecast load in the hour tthat comes from the chronological modelling, La(t) is the actual load at hour tand ∆L(t) is the short-term uncertainty assumed to follow a Gaussian distribution [13] with zero mean and a standard deviation proportional to the load. The standard deviation of this normally-distributed error in Equation (3.19) is assumed to be a percentage of the load. The main implication of this short-term load forecast is that it directly affects the decisions related to the amount of spinning reserve, as well as fast tertiary reserve. 64 3.3. Long-term Reserve Evaluation 3.3.3 Modelling Wind Power Forecast Uncertainty The state of the art in day-ahead wind power forecasting provides a normalized root mean square error between 15 and 20% [88] of installed capacity, which in fact has an impact on the task of fitting large amounts of wind generation into unit commitment or dispatch procedures. Some researchers have assumed normally-distributed errors for short-term forecasting [86], mainly based on the amount and the geographical dispersion of wind power [89]. Nevertheless, other researchers affirm that the distribution of the actual wind power forecasting error is not normally-distributed [90]. While wind speed distribution is generally approached through the Weibull and Beta distributions [91, 92], the analysis made over the wind speeds data for six onshore sites in Germany [91] showed that for relevant wind speed range (where the wind is useful for wind power generation), the probability density functions can also be Gaussian. On the other hand the non-linear relation between the wind speed and wind power leads the wind power prediction errors to non-Gaussian distributions. In fact, wind speed, and therefore wind generation, is usually modelled using stochastic processes due to the complexity of wind behaviour. A simplified method based on the persistence method is presented to characterise the error of wind power due to the short-term wind forecasting procedure. ∆Wi(t) = Wi(t)−Wi(t−τ) (3.20) where Wi(t) is the wind production for an individual wind farm at hour t,Wi(t−τ) is the last known wind production for such wind farm, and ∆Wi(t) is the wind power deviation of this individual wind farm at hour tdue to the error in the wind forecasting procedure. After identifying the wind power deviation for each wind farm at hour t, it is necessary to define the system deviation, which will be the sum of all individual 65 Chapter 3 wind farm deviations. ∆WS(t) = k X i=1 ∆Wi(t), i = 1,2, ..., k (3.21) where ∆WS(t) is the system wind deviation. The persistence method is the largest one used to represent the uncertainty of the wind power forecast, since it has a simple implementation and a adequate level of accuracy, as mentioned in Section 3.3.3. 3.3.4 Relationship between Generation and Load Uncertainties The aforementioned system uncertainties may be viewed as a positive or negative impact on the system balance, depending on how suddenly they occur and on the magnitude and direction of the variations. Table 3.2 shows a summary of consequences (upward or downward reserve) that should be monitored during the adequacy evaluation of the generating systems concerned with these variations. The concept of net load forecast is usually applied in the literature [93] to refer to the imbalance caused by load and wind power forecast deviations. This concept is not used here, mainly because all variation models are considered statistically independent events and an additional major variation has been modelled using the generating outages. The effects shown in Table 3.2, reinforce the need for assessment of long-term operating reserve. These events involve not only the load and wind power forecasting errors, but also the generating outage variations. Therefore, the power balances may have negative or positive effects to the system with upward or downward reserve as consequence. As the adequacy evaluation of the generating system, performed in this thesis, is concerned to the long-term power 66 3.3. Long-term Reserve Evaluation Table 3.2: Relationship among the system uncertainties. Variables Conditional Deviation Reserve Effects Need Id. ∆L∆G∆WSif d up or down A + - - - upward B- - - (∆G+ ∆WS)>∆L+ downward - - - (∆G+ ∆WS)≤∆Lupward C- - + (∆G+ ∆WS)>∆L+ downward - - + (∆G+ ∆WS)≤∆Lupward D+ - + (∆G+ ∆WS)>∆L+ downward + - + (∆G+ ∆WS)≤∆Lupward E+ + + (∆G+ ∆WS)>∆L+ downward + + + (∆G+ ∆WS)≤∆Lupward F- + - (∆G+ ∆WS)>∆L+ downward - + - (∆G+ ∆WS)≤∆Lupward G+ + - (∆G+ ∆WS)>∆L+ downward + + - (∆G+ ∆WS)≤∆Lupward H - + + + downward balance, the downward reserve can be disregarded since the approach used is based on the system capacities. The first analysis from Table 3.2 is the collective event A. This event shows that the load forecast error is positive (i.e., the actual load is greater than the expected one) and both wind power forecast error and the synchronised generating units are negative (i.e., the amount of generating power is lower than the expected one). This collective event has as consequence an upward requirement of reserve. The event H, shows the opposite situation. The system uncertainties collaborate with the system requesting downward reserve. The ∆Gvariable is the balance between the synchronized generating units and the summation of the forecast system load, primary and secondary reserves, as seen in Equation (3.9). From A to D cases of Table 3.2, ∆Gassumes negative values. This means that all the synchronised generating units available are not able to meet 67 Chapter 3 the operating system requirements plus the system load forecast. However, this does not mean a failure state of the operating reserve capacity. For instance, this lack of capacity might be compensated by the load and wind power forecasting errors which might contribute to the system. Therefore, these possibilities lead to a downward or upward reserve, according to Table 3.2. On the contrary, from E to H cases, the ∆Gassumes positive values. This means that there is enough synchronised capacity to deal with the system requirements. Nonetheless, this situation does not imply on success state of the operating capacity reserve evaluation. As matter of fact, if the surplus of the synchronised capacity plus the fast tertiary reserve (operating reserve capacity) are not enough to meet the load and wind power forecasting errors, assuming that these uncertainties require more electric power generation, then a failure state is established. The above analysis can be expanded to the ∆Land ∆WSuncertainties, according to Table 3.2. 3.3.5 A Simple Example This section presents an illustrative example of the previous discussion highlighting the impact of the system uncertainties in the operating reserve capacity evaluation. Table 3.3 presents the reliability indices of the static and operating reserve capacity evaluations, which were performed through the use of a modified version of the IEEE RTS 1996 [60]. Table 3.3: Example of the SMCS method application. LOLE (h/y) EENS (h/y) LOLF (h/y) LOLD (h/y) β(%) β(%) β(%) β(%) Static Reserve 0.3456 66.49 0.1360 2.5398 (3.30) (4.99) (3.90) - Operating Reserve 0.7679 129.30 0.4785 1.6047 (4.08) (4.97) (4.41) - 68 3.4. Electric Vehicle Demand Modelling The LOLE index of 0.34 hours per year, of the static reserve evaluation, gives an idea of the amount of time that the load is expected to exceed the total available generating capacity. From this perspective the test system can be considered quite reliable. From the operating reserve capacity evaluation, the LOLE index of 0.7679 h/y gives an idea about the flexibility of the generating system configuration facing the system deviations. In other words, it shows the amount of time that the deviations of the system are expected to exceed the available committed capacity plus the fast tertiary reserve. 3.4 Electric Vehicle Demand Modelling The estimation of EV load based on mobility patterns is an adequate approach, since no EV data is available, to include this type of load in the static reserve evaluation. From the operating reserve capacity perspective, the uncertainty related to the EV load estimation is not significant, but the opportunity that they arise to contribute to the operational reserve should also be taken into account. Under an adequate charging strategy the EV charging rate can be controlled or even postponed. Analogously to Equation (3.19), the total EV load is given by LEV total =LEV −∆LEV (3.22) where LEV is the estimate EV load, LEV total is the actual EV load, and ∆LEV is the portion of EV load which is able to contribute to the system if the operating reserve capacity is threatened. This action is named as controlled charging strategy. This new variable ∆LEV , directly affects the reduction of LEV total and indirectly affects the increase of operating reserve Rop. This action differentiates the 69 Chapter 4 where kis the number of sub-samples, ris the r−th random group r= 1, ..., k,xr represents the total of a feature in group r, obtained through the sample values. yris the total individuals in group rof the sample, and ˆ Ycis the independent estimate of the population in county c. In order to control the quality of the survey, 5% of the sample was reinterviewed being obtained a consistency rate of 98% of the answers. Taking advantage of the data separation related to the transport mode, Table 4.1 presents the expected arrivals of the motorised population, during an ordinary weekday, divided through different reasons of making a trip in the north region of Portugal. This table relates the citizen mobility when people go to work, leisure, shopping or home in an hourly basis. The associated percentages are the density information of the mobility and the expected arrivals parameter per hour. Note that, during the overnight, almost no vehicle arrivals happen. On the other hand, the peak periods at work hour, lunch time and return home revealed the daily Portuguese habits. Assuming that EV owners are able to charge the EV batteries wherever they are parked, the total expected number of arrivals is the parameter used as input to the counting process. Figure 4.3 represents this arrival distribution of the total motorised population in an ordinary weekday. Figure 4.3: Weekday arrivals distribution. 76 4.2. Mobility Pattern Note that the three peak periods related to work hour, lunch time and return home, appear in the figure from 7 to 9 h., from 12 to 15 h. and from 17 to 19 h. Table 4.1: Distribution of trips throughout the day - per reason. Hour Arrivals Total Labour Shopping Leisure Return Others Home Percentage 0 0,1 0,0 0,1 0,1 0,3 0,1 1 0,0 0,0 0,0 0,0 0,0 0,0 2 0,0 0,0 0,0 0,0 0,0 0,0 3 0,0 0,1 0,0 0,0 0,0 0,0 4 0,0 0,1 0,0 0,0 0,0 0,0 5 0,3 1,1 0,0 0,1 0,0 0,2 6 0,9 2,6 0,3 0,3 0,2 1,0 7 5,8 16,6 1,2 2,6 0,2 8,6 8 10,9 27,4 5,7 4,4 0,6 16,3 9 7,7 10,6 11,6 4,7 0,9 10,8 10 5,3 3,2 12,8 4,7 1,6 4,4 11 2,9 1,7 4,9 2,5 3,0 2,3 12 6,7 2,0 8,9 4,8 13,9 4,0 13 7,5 13,7 4,9 6,1 6,4 6,5 14 7,6 10,5 8,8 9,2 2,3 7,3 15 5,8 2,9 9,1 8,1 2,8 6,0 16 4,4 1,8 6,9 4,9 4,1 4,1 17 5,8 1,5 6,1 4,1 10,9 6,2 18 8,0 1,3 5,8 5,7 18,8 8,3 19 6,0 0,9 4,4 5,6 13,9 5,2 20 5,1 0,7 3,7 9,8 7,6 3,8 21 4,9 0,7 3,3 14,0 3,7 2,9 22 2,5 0,4 1,2 5,8 3,8 1,2 23 1,7 0,3 0,3 2,4 4,9 0,8 Total 100.0 100.0 100.0 100.0 100.0 100.0 77 Chapter 4 Analogously, Figure 4.4 represents the arrivals distribution of the motorised population, during an ordinary weekend. The figure presents an increase in the number of arrivals during the valley period in comparison with Figure 4.3. On the other hand, the decrease in the number of arrivals in the morning period confirms the expected weekend habits. Moreover, three peak periods were identified: from 23 to 1 h., from 12 to 14 h. and from 19 to 21 h. Figure 4.4: Weekend arrivals distribution. Note that in both figures, 4.3 and 4.4, the evening period, which is related to return home, can be related to the peak demand period of the day. Therefore, assuming, for instance, that the population will start to charge the EV batteries as soon as they arrive home, then, the battery charging will increase the traditional peak period of the conventional system load. The survey presented through this section generated the arrivals distributions showed in Figures 4.3 and 4.4. From these ones, the arrivals averages, in an hourly basis, can be used as the only parameter needed to estimate the number of EV, which will proceed to battery charging. 78 4.3. Proposed Counting Process Methodology 4.3 Proposed Counting Process Methodology Different EV load representations have been mentioned in the literature [55,73,75– 77,96–99]. This thesis proposes a probabilistic technique to estimate the number of EV arrivals based on mobility pattern information. It is assumed that the EV arrivals occur randomly, therefore, the Poisson process can be used to estimate these arrivals based only on the expected arrival parameter. The reference [100] also attests the importance of using Poisson process in many stochastic models, such as: calls arriving at a help desk centre, traffic light problems, car arrivals at a fuel station, and so on. This thesis, aims at using arrival distributions data, generally provided by mobility surveys, which can be viewed from two different perspectives leading to the development of two counting process approaches. These approaches were developed, by the author, within the framework of two research projects, the European project MERGE (Mobile Energy Resources in Grids of Electricity) [101] and the Portuguese project REIVE (Smart Grids with Electric Vehicles) [102]. One perspective is based on the homogeneous Poisson process and its major characteristic is counting the EV arrivals using one single variable: the expected arrival rate λ. The second one was developed in order to increase the detailed representation of the EV arrivals by the inclusion of an arrival time variable in the problem. This perspective is based on the non-homogeneous Poisson process, which counts the EV arrivals continuously, using a time dependent expected arrival rate λ(t). The principle behind these proposed approaches, considers that cars arrive at different charging points at random moments of the day in accordance with a Poisson distribution with λrate. It is also considered that each arrival would 79 Chapter 4 correspond either to different types of events, where the EV needs to connect to the grid and proceed to battery charging or to a type of event where the EV do not need to proceed to battery charging. Both types of events may be viewed as a Poisson process. Next sections present the methodology and clarify the differences between these two counting process approaches which are the ones that define the number of EV arrivals during the SMCS procedure. 4.3.1 Homogeneous Poisson Process The homogeneous Poisson process is characterised by a constant rate parameter λ, also known as intensity, such that the number of arrivals in time interval (t, t +τ] follows a Poisson distribution with associated parameter λτ. This relation is given by: P[(N(t+τ)−N(t)) = k] = e−λτ (λτ)k k!k= 0,1, ..., (4.6) where N(t+τ)−N(t) = kis the number of arrivals in time interval (t, t +τ]. The rate parameter λis the expected number of arrivals that occur per unit time. The HPP assumes a counting measure where the number of arrivals in an interval (t, t+τ] is independent of the number of arrivals in other interval as (t+1, t+1+τ]. Figure 4.5 depicts the HPP. In this thesis, the HPP is performed for each hour during an yearly sample of the simulation process. It means that for each hour of different days, the same λ parameter is expected. For instance, hour 1 of day 1 has the same λparameter as the hour 1 of the day 2, and of the day 3 until the end of the year. As the EV arrivals are clustered in an hourly basis (τ= 1), in order to calculate the EV load in the same time horizon of the conventional load, it is assumed that the remaining time in charging state is fixed and dependent on the technical characteristic of the 80 4.3. Proposed Counting Process Methodology Figure 4.5: The counting process through an HPP. battery, as capacity and charging rate. In this approach it is also assumed that the EV charging happens once a day for a fixed battery charging requirement and this happens in the end of the first trip. These assumptions can be parametrised in the inputs of the model. Using Algorithm 1, which is based on the inverse-transform method, Figure 4.6 is generated and gives the number of arrivals per hour during the chronological simulation. In this sample, those three peak periods (see Figure 4.3) are highlighted, although the peak period of the morning presented a lower value than the expected. where nis the number of experiments, Unis a uniformly distributed random variable and Nis a Poisson random variable which represents the number of arrivals. For fixed intervals e−λλn n!is the probability mass function of Nand while ais greater than e−λ, the number of arrival is increased by one. Equation (4.7) gives the departure time for a fixed requirement of the battery capacity. Assuming a certain level of proportionality among the different types of EV for all samples under analysis, the total EV fleet will be proportionally divided into different EV types [2]. The departure time Tdn is then calculated through: 81 Chapter 4 input : Expected number of arrivals λat hour t output:X←n−1 // Number of arrivals at hour t n←0; a←1; while a≥e−λdo Un←U(0,1); a←aUn; n←n+ 1; end Algorithm 1: HPP based on the Inverse-Transform method. Tdn =Tan +Cbj Rcj ·κ(4.7) where Tan is the EV arrival time. Cbj is the battery capacity and Rcj is the charging rate and κis a battery charging coefficient which ranges from [0,1]. The departure time calculated through Equation (4.7) is according to the EV Figure 4.6: Sample of the HPP. 82 4.3. Proposed Counting Process Methodology charging strategy. For instance, the arrival and departure time for valley charging strategy is into the valley hours range. It means that if a vehicle arrives at 9 o’clock and departures at 13 o’clock, its battery charging time will occur from 23 to 3 h, considering 23 h the start battery charging hour for this strategy. Simple Numerical Example Supposing an ordinary week day and a fleet of 20 EV. At 20 h, using the rate parameter λof Figure 4.3 in the Algorithm 1, the sampled number of EV arrivals which proceeds to battery charging is achieved. Table 4.2: Example of the HPP application. a n U(0,1) a=aU λ e−λN 1.000 0 0.99 0.990 2.2 0.01 − 0.990 1 0.90 0.891 2.2 0.01 − 0.891 2 0.98 0.873 2.2 0.01 − 0.873 3 0.30 0.261 2.2 0.01 − 0.261 4 0.03 0.007 2.2 0.01 − 0.007 − − − − − 3 Table 4.2 shows 4 iterations of the counting process method until achieving the stopping criterion. At the end, 3 EV are expected to arrive at 20 h to charge their batteries according to their charging strategies. 4.3.2 Non-homogeneous Poisson Process The non-homogeneous Poisson process consists of continuously counting the number of arrivals where the expected ones may change over time. This is characterised by a continuous rate parameter λ(t). The NHPP has been used to 83 Chapter 4 describe numerous random phenomena [103] including cyclone prediction [104], arrival times of aircraft to airspace around an airport and database transaction times [105]. The NHPP assumes N={Nt, t ≥0}for which the number of points in nonoverlapping intervals are independent, but the rate parameter at which points arrive is time dependent. If λ(t) denotes the rate parameter at time t, the number of points in any interval (a, b] has a Poisson distribution with mean λa,b =Zb a λ(t)dt. (4.8) Thus, the number of arrivals in time interval (a, b], given as N(b)−N(a), follows a Poisson distribution with associated parameter λa,b, and is calculated through: P[(N(b)−N(a)) = k] = e−λa,b (λa,b)k k!k= 0,1, ..., . (4.9) Note that, an homogeneous Poisson process is a special case of NHPP which consists of the rate parameter λ(t) becomes constant. Figure 4.7 illustrates the construction of an NHPP based on the acceptance-rejection method. The idea of this method is to first find a constant rate function λ(t) = maxλ which dominates the desired rate function λ(t), next generates from the implied HPP with rate λ(t) = maxλ, and then rejects an appropriate fraction of the generated arrivals so that the desired λ(t) is achieved. Formally a two-dimensional HPP is generated on the strip {(t, x), t ≥0,0≤x≤λ}, with constant rate, and then all points are projected bellow the graph of λ(t) onto the t-axis. The points of the two-dimensional process can be viewed as having a time and space dimension. The arrival epochs form one-dimensional Poisson process with rate parameter maxλ and the positions are uniform in interval [0, maxλ]. This suggests the following alternative procedure for generating NHPP: each arrival epoch of the one-dimensional HPP is rejected with probability 1 −λ(Tn) maxλ where Tn 84 4.3. Proposed Counting Process Methodology Figure 4.7: The counting process through an NHPP. is the arrival time of the n-th event. The epochs not rejected define the NHPP. The basic difference between the counting process approaches is regarding the different time resolutions. While the HPP model clusters the vehicle arrivals in an hourly basis, the NHPP model counts the EV arrivals continuously. The Equation (4.10) generates the EV departure time. The NHPP is performed in the beginning of the SMCS procedure generating an annual arrival and departure times vector. The characteristic of observing the individual arrivals makes sampling the SOC of each vehicle possible, as can be seen in the right side of Equation (4.10). The departure time Tdn of each vehicle is then calculated through: Tdn =Tan +Wn·Cbj Rcj (4.10) where Wnis a uniformly distributed number between [0,1]. Cbj is the battery capacity and Rcj is the charging rate. These electrical parameters are divided into different EV categories j, as can be seen in [2]. The period between the arrival and departure of an EV is the battery charging time of the battery charging requirement. 85 Chapter 4 described in Chapter 5. 4.5.2 Vehicle-to-Grid Charging Strategy The V2G concept is the ability of combining both battery charging and injection of electric energy to the system. This is possible due to a bi-directional converter which allows this level of charging controllability. From an ORC perspective, the V2G is viewed as an opportunity to increase the renewable energy source integration in the generating systems by both smoothing the wind power variability and/or providing secondary reserve to the system. Both concepts are better described in Chapter 5 where the controlled charging strategies are exploited in detail. Figure 4.12 illustrates one probable situation of increasing the operating reserve. The “Lf+LEV (V2G)” line represents the V2G charging strategy impact on the total system load while the “G+V2G” line is the V2G impact on the available generating capacity. Assuming that the Gand Lf+LEV lines represent the normal operation of the system, it is possible to identify a deficit of generating capacity at 17 h. Mobilising EV to inject stored electrical energy back to the grid, the system operation passes to be represented by the G+V2Gand Lf+LEV (V2G) lines. Note that taking the EV SOC into account, the vehicles under a V2G charging strategy can effective increase the operating reserve capacity available, in order to solve the possible system failure state. The amount of injected capacity starts to charge at 23 h. Chapter 5 will present the V2G charging strategy in more detail. 92 4.6. Conventional Load and Proposed EV Load Estimation Figure 4.12: System load profile using V2G charging strategy. 4.6 Conventional Load and Proposed EV Load Estimation Load shapes provide a means of understanding of how much energy is being used at different times of the day, week, season, or throughout a complete year. When the energy use patterns are being developed for groups of equipments with similar functions, the results are commonly referred to as end-use load shapes which represent the habit of the citizens about energy use. The most common end-use categories for the commercial, residential, and industrial classes are presented in Table 4.4. In order to include some new end-use categories for the current classes, it is necessary to approach EV as a new load trend, which will influence the future load shapes. From these main categories, the EV charging might impact on the Residential and Industrial, since the EV owners probably will charge the EV batteries in the own place. On the other hand, the Commercial category, will probably use public or private infrastructure to charge the EV batteries, once they, generally, do not have an infrastructure that allows them to charge the EV battery in their own place. 93 Chapter 4 Table 4.4: Common end-use categories by customer class End-Use Category Commercial Residential Industrial Air conditioning • • • Space heating • • • Interior lighting • • • Miscellaneous equipment • • Domestic hot water • • • Computers • Cooking • Refrigeration equipment • Ventilation • • Exterior lighting • • Process equipment • Motors • Stoves / ovens / ranges • Refrigerators / freezers • Televisions / stereos • Dishwasher • Clothes washer / dryer • Electric Vehicles Charging • • Electric utilities have historically made large use of end-use load shapes in the energy and load forecasting. For these purposes, utilities are commonly faced to predict what their future capacity will be considering factors as: the current base load, the expected change in the number of residential homes, commercial stores, and industrial facilities, or even more, the change in equipment efficiencies over time. However, the massive EV load integration on the power systems could be viewed as a break of paradigm changing the ordinary load shapes and end-use patterns. Based on the forecast supply and demand curves, a utility usually wishes to examine its current and future state of generation capacity under short and long-term perspectives. This thesis is proposing an alternative way to estimate EV load through the use of a counting process, based on the expected number of 94 4.6. Conventional Load and Proposed EV Load Estimation arrivals (mobility pattern). The EV load estimation can be calculated according to the following procedures. 4.6.1 EV Load from the Homogeneous Poisson Process The EV counting process method gives the number of EV which proceeds to battery charging. It is assumed that all connected vehicles will charge their batteries if allowed to the chosen charging strategy. In fact, the counting process does not change with different charging strategies; on the contrary, the EV load profile changes according to the charging strategy. In order to calculate the EV load, the battery electric parameters and the different types of EV must be taken into account. This being defined, the following procedure may be adopted LEV (t) = X j LEV j (t)·Nj(t)j= 1,2, ... (4.11) where Nj(t) is the number of EV arrivals of the type j, in hour t.LEV j(t) is the EV load in hour t. This EV load, calculated for each hour of the simulation process, remains in charging mode until the departure time, calculated in Equation (4.7), is reached. Therefore, the adequacy evaluation of the generating systems is analysed for each transition (hourly) of the EV charging mode. Figure 4.13 illustrates the procedure to calculate the EV load using the HPP. In this illustrative example, the different colours are the EV arrivals counted through the HPP. At 19 h. one EV arrival is identified and it is assumed that this vehicle connects to the grid to charge its battery during 3 hours. At hour 20 h, two EV arrive in a certain place to be charged and their consumption is added to the demand of the last vehicle already in charging mode. At 21 h, four EV are in charging mode. 95 Chapter 4 Figure 4.13: Illustration of the EV load calculation through HPP. Note that for illustrative purposes, a fixed battery charging requirement needs a charging time of three hours. Then, because of the chronological characteristic of the EV arrivals, the load increases as the arrivals happen, assuming that the EV owners will charge the EV batteries as soon as they park. However, after three hours the EV is already charged and is disconnected to the grid, decreasing the EV load as showed in hour 22 and 23 h of Figure 4.14. The Figure 4.14 gives the total EV load profile for each hour, according to the direct charging strategy. Moreover, this load is added in each system load state transition of the SMCS, chronologically. Figure 4.14: EV load profile using HPP and direct charging strategy. 96 4.6. Conventional Load and Proposed EV Load Estimation Figure 4.15: EV load profile using HPP and valley charging strategy. Now, imagine the same previous illustrative example. However, let the direct charging strategy be changed by the valley charging strategy. Figure 4.15 gives an idea of the EV load. The EV arrivals occurred during the daylight, are connected to charge the EV batteries in the valley hours (for instance, at 23 h). Then, this EV load is added on the system load, chronologically. As the system load is lower in the valley hours, it is expected a low EV impact when this charging strategy is applied. 4.6.2 EV Load from the Non-homogeneous Poisson Process The EV load curve of the NHPP, is performed for all samples under analysis integrating the area formed by the arrival and departure times (see Figure 4.16). Therefore, the EV load is given by the following equation LEV (t) = X j=1 ZTdn Tan LEV j (t)dt (4.12) where Tan and Tdn are the arrival and departure times, respectively. As mentioned before, the arrival and departure times form a sorted vector in time. Therefore, the adequacy evaluation of the generating systems is analysed in each transition 97 Chapter 4 Figure 4.16: Illustration of the EV load calculation through NHPP. (arrival or departure) of the EV charging mode. Figure 4.16 illustrates an EV arrival which proceeds to charge the battery at Ta1 and left the charging mode at Td1when the SOC is achieved. A second EV arrival happened at Ta2and left the charging mode at Td2. The integration of the hatched curve is the EV load. In this case, the SMCS evaluates each transition instant, in order to keep a continuous tracking of the load. As mentioned before, the detailed description of the active charging models is given in Chapter 5 under the contexts of the homogeneous and non-homogeneous Poisson processes. 4.7 Proposed EV Models Integration on SMCS Method As the electric system components, the EV load is chronologically represented through the SMCS method. This method relies on repeated random sampling to obtain numerical results by running simulations many times over in order to calculate those same probabilities heuristically. The main advantage of this method is the possibility to generate probability distribution of the events. The 98 4.7. Proposed EV Models Integration on SMCS Method Figure 4.17: Chronology of the Sequential Monte Carlo Simulation. chronological procedure of the generating system adequacy evaluation is illustrated in Figure 4.17. From the static reserve evaluation perspective, Figure 4.17 shows system failure state in three moments. As previously mentioned, the evaluation is conducted from two different perspectives: static reserve and operating reserve capacity evaluations, as in [13]. The objective is to evaluate the impact and opportunities for uncontrolled and controlled charging strategies that EV could have in the future generating systems. 4.7.1 Static Reserve Evaluation with Electric Vehicles The Equation (3.4) presented in Chapter 3 is used to measure the level of risk in which a future generating system configuration is able to meet the forecast load. In order to include the effect of the EV load in this generation assessment, the 99 Chapter 4 Equation (3.4) is then modified according to Equation (4.13) G−(Lf+LEV )≤0 (4.13) where LEV is the total EV load calculated through the EV models. 4.7.2 Operating Reserve Capacity Evaluation with Electric Vehicles Originally, Equation (3.10) is set to assess the risk indices associated to the longterm operating reserve. It captures the risk of forecast errors linked to load and wind power generation, as well as the forced outages of the generating units. The inclusion of EV in the generating system should also be represented in the ORC evaluation. Therefore, the Equation (3.9) is then modified in order to meet this new random variable according to Equation (4.14). ∆G=Gsync −[(Lf+LEV ) + RP+RS] (4.14) where LEV is the total EV load, as previously stated. This change is illustrated through the Figure 4.18, where the expected EV load is added on the conventional system load. Figure 4.18: Operating Reserve Capacity evaluation with EV. 100 4.7. Proposed EV Models Integration on SMCS Method Figure 4.18 is a general representation of the operating reserve capacity evaluation. The synchronised generating units meet the system requirements as in Equation (4.14). The hatched block, represents the discrete effect of the scheduling procedure. Then, as stated in Equation (3.10) the available capacity sinchronised RSplus the fast tertiary reserve RTshould meet the system deviations. From the left to the right side of Figure 4.18, the system moves from a success to a failure state. Regarding the EV integration in SMCS method, the Figure 4.19 shows the flowchart of the entire procedure. Basically, the simulation process is divided into five stages: system composition, system state selection, state evaluation, EV load control and end of the simulation process. 101 Chapter 5 Controlled Electric Vehicle Charging Modelling 5.1 Introduction The uncontrollable charging strategies, addressed in the previous chapter, represents the EV as a conventional load, the EV owner being free to charge the vehicle at any time of the day. The main difference of the EV load profile is related to the periods of the day in which the EV owners will charge the battery. These strategies do not account controlled models, which could effectively manage the EV battery charging. From [106] the electric vehicles could provide system support in 81% of the time, when charging spots are available at home and at work. Then, smart charging schemes are desirable in order to mitigate the EV impact and to exploit EV as an electrical component that is able to provide ancillary services for the system. The main challenges of the integration of electric vehicles as active components of the electric system are the infrastructure, management and control sectors. 109 Chapter 5 From the technological point of view, the expected large scale deployment of equipments on the grid such as the energy boxes, the transformer controllers and the improvement of the distribution management systems (DMS) is viewed as an opportunity to create active demand side management solutions and control strategies to reduce the EV impact on the power systems, taking advantage of their ability to provide ancillary services through an aggregation entity. The smart electronic devices record consumption of electric energy in intervals of an hour or less and communicate that information at least daily back to the utility for monitoring and billing purpose. They also enable a two-way communication between the device and the aggregation entity. These aspects make these charging schemes possible, at least, from the technological infrastructure perspective. The charging process of these structures can take less than half an hour (for fast charging rates) up to 8 hours (for slow charging rates). In this sense and, assuming a higher price to charge the EV battery in fast charging stations, it is expected that EV will be connected to the grid for large periods of time, being potentially possible to exploit their storage capacity to enable a grid support service. Since the choice for controlled charging schemes will always be a decision taken by the EV owners, this thesis addresses two possible charging schemes which fit the EV owners’ needs and the system requirements. In one hand, an active demand side management strategy is implemented in order to contribute for the provision of the operating reserve capacity, which is named as controlled charging strategy. On the other hand, a V2G charging strategy is developed to mobilise enough electric power capacity to increase the operating reserve in moments where conventional generation capacity is not available and to compensate the impact of the wind variability on the operating reserve capacity. In normal operating conditions, a commercial aggregation entity is expected to manage and control the EV charging in controllable mode. The main objective is clustering the EV, and exploit business opportunities in the electricity markets to 110 5.2. Active Charging Framework provide reserve. Throughout this thesis, the definition of reserve is an amount of available electric energy capacity to meet unexpected variation on load, renewable power and generating capacity due to forced outages. In order to successfully respect the agreements, both with the clients and with the electricity market, the EV aggregator must be capable of sending set-points to the charging points related to rates of charge. In [67], an optimization approach to support the aggregation agent participating in the day-ahead and secondary reserve sessions is presented. The aggregator will be responsible for managing the EV charging; therefore, whenever the security of operation is threatened, it is able to mobilise EV to provide operating reserve in order to aid the system. Before the description of the proposed EV models, two sections are included to give technological support for the controlled charging schemes: an overview on the specifications for the EV charging interface, communication and smart metering technologies and a life cycle study which is a matter of interest in V2G studies. Each model were developed in the framework of the European projects MERGE and STABALID, respectively. This chapter is organised as follows. Section 5.2 presents an overview on the active charging framework for V2G strategy. The controlled charging model is described in Section 5.3. The reliability aspects of V2G are presented in Section 5.4. Section 5.5 presents the V2G charging model approach and in Section 5.6 presents the final remarks. 5.2 Active Charging Framework Technologically, the achievement of a plug-and-play capability for the interface linking electric vehicles and grid is desirable. This interface must be able to access identical charging points (CP) across Europe, for instance, that can be used by any equipped vehicle, just as the roaming technology for cell phones. 111 Chapter 5 Basically, this plug-and-play should consist of a power stage and an information and communication technology (ICT) stage. A power stage embraces the physical connections and functionalities between EV owner (user) and charging point. The ICT stage communicates between, at least, five identified parties: the user, the EV itself, the charging point, the SO and the aggregation entity. Figure 5.1 shows an illustrative representation of this relationship. Figure 5.1: Information and Communication Technology scheme. The definitions for the identified parties are: •The charging point is the equipment containing the electrical and communication connection to the EV owner, electricity grid, system operator and aggregation entity. It is composed by the electricity meter, the communication modules, the electronics and software needed to control the charging process, a main control board, and a connector for the charging cable to be plugged into. •The electric vehicle interacts with the charging point in order to give the needed information: state of charge and charge/discharge rate. The charging electronics (AC-DC converter, AC-AC converter, power electronics etc.) are 112 5.3. Proposed Controlled Charging Modelling embedded in the EV as an option of the vehicles manufacturers. •The EV owner (user) needs to provide the inputs needed by the charging point software: departure time, how many kilometres the user needs to drive or desired SOC. •The system operator is the responsible entity for the transmission/distribution system, and the one that provides electrical power for such charging point. •The aggregator is the entity that manages the EV charging and mobilises the EV in order to provide operating reserve to the system. These features define the basic framework to an aggregation entity provides the controlled charging schemes. Next sections present the controlled charging model and the V2G charging model for adequacy evaluation of generating systems. 5.3 Proposed Controlled Charging Modelling Differently from the conventional flexible loads, the EV have batteries capable of storing electric energy to be used later. Through the vehicles mobilisation, an aggregation entity may supply reserve for the system. This action needs to consider a predefined SOC that must be guaranteed to ensure the EV usability. The main objective of the controlled charging strategy is to avoid the impact that uncontrolled charging strategies may have on the electric systems and, at the same time, to provide operating reserve to the grid releasing generation capacity to meet the system variations. Figure 5.2 illustrates such situations. After the system requirements have been 113 Chapter 5 met, through the synchronised generating units, the Equation (4.14) is tested. In situation (a), it is showed that the operating reserve capacity (ORC) is sufficient to cover all system uncertainties, resulting in a success state. However, due to the variation of the wind power, the load forecast error and the forced outages of the generating units, situation (b) presents a system failure state, where the result of Equation (4.14) becomes true. Thus, assuming that the EV battery charging can be quickly interrupted, by decreasing the EV load, as showed in situation (c), the operating reserve RS+RT, is increased through the release of pre-scheduling generating units. These units were synchronised to meet the additional load that EV represents. Since, these EV stopped and the EV can effectively contribute to the operational reserve of the system and perhaps change the system state from a failure to a conditioned success state. Note that the released pre-scheduled generating units are part of the ones scaled to meet the additional EV load demand. Therefore, this action makes possible to take advantage of load shifting instead of the ordinary load curtailment. In turn, the EV, which releases reserve for the electric system, will be charged at another hour, when no generation deficit is expected. Moreover, as the ORC evaluation has the objective to measure the system flexibility when dealing with the system variability, through the already mentioned risk indices, the management of a massive EV charging may avoid or postpone the generating system reinforcement. On the other hand, the simulation process using different charging strategies may provide, through the risk indices analysis, a notion of the charging strategy required in order to maintain the system reliability when an EV fleet is integrated in the electric system without changing the desired generating system configuration. 114 5.3. Proposed Controlled Charging Modelling Figure 5.2: Operating Reserve Capacity evaluation with controlled charging strategy. The flowchart presented in Figure 4.19, was modified to include this charging strategy. Figure 5.3 shows a new flowchart for the SMCS process, which can be described by the following steps: 1. Initiate all the component state. Commonly, it is assumed that all the components are in the state UP. Define the maximum number of years to be simulated, Nmax and the convergence criteria β. Set the number of years to one Nyear = 1. 2. Set the simulation time to zero t= 0 and sum one in the number of simulated years Nyear =Nyear + 1. 3. Define the EV model approach desired. If HPP is the chosen one, go to step 4. Otherwise, go to step 5. 4. Sample the current state duration of each system’s component. If using an exponential distribution to approach the state duration, then it is calculated as follows: Ti=−1 αi ln(Ui).(5.1) 115 Chapter 5 Figure 5.3: Flowchart of the SMCS with the EV controlled charging strategy. Where Uiis a uniformly distributed random number between [0,1], istands for the component number. The MTTF and MTTR values are represented by αaccording to the current system’s state. The load transitions occur in an hourly basis with 8760 load points. Go to step 6. 5. Sample the current state duration of each system’s component through Equation (5.1). Sample the EV arrivals and calculate the departure time of 116 5.3. Proposed Controlled Charging Modelling each one, according to Equation (4.10). The load transitions occur in an hourly basis. 6. Update the simulation clock t, according to the selected state transition. If using the NHPP approach, each arrival is considered as a state transition. Otherwise, the EV load is added in each system load transition. 7. In order to obtain yearly reliability indices, evaluate the test function over the accumulated values. If a failure state occurs, then go to step 8. If a success or conditional success state occurs, go to step 10. 8. Reduce the EV load of the vehicles in controlled charging mode to be charged later after the departure hour defined before the failure occurrence. Return to step 7. If there is no EV in controlled charging mode, go to step 9. 9. Update the outcome of reliability test functions and the corresponding indices. 10. If the simulated year is not in the end, then return to step 4 or 5, according to the EV model approached chosen. Otherwise, go to step 11. 11. Estimate the expected mean values of the yearly indices as the average over the results for each simulated sequence. 12. Test the stopping criteria according to their definitions in the beginning of the simulation process. 13. If the stopping criteria is not reached, repeat the step 2 each time span and record the results of each duration sampled for all components. Otherwise, go to step 14. 14. End the process if the desired degree of confidence is achieved. If not, return to step 2. 117