The thesis work addresses the integration of balancing services provision within the optimization of Smart Multi Energy Districts (Smart MED) operations. Traditionally, efforts of both researchers and professionals have been concentrated in obtaining the maximum energy efficiency from multi-energy districts operations while satisfying their internal energy demands. The advent of non-programmable renewables (NP-RES) and the decentralization of electricity production have challenged this view. Because of this, it is fundamental to develop computational instruments able to couple districts’ energy demands with balancing needs of the external public grid. Since early 90s, resources for electricity dispatching have been collected on electricity markets; these are usually composed by: − a day-ahead energy trading, where electricity demand and production first meet, − an intra-day electricity exchange, mainly exploited to correct power generation profiles according to unexpected events, and − one or more markets where ancillary services are purchased by system operators to maintain the exact frequency and voltage levels along power grids. This market sequence generates a series of price signals that push production (and consumption) units to modify their scheduled operations, their bidding strategies and their business models. The recent massive diffusion of NP-RES plants has shifted the attention from day-ahead trading, where prices are influenced by power plants marginal costs, to real-time markets, which are instead influenced by network conditions and system imbalances. This work focuses on balancing markets as enablers for the provision of balancing resources from Smart MED. The procurement of these resources is based on two fundamental steps: − the creation of an adequate level of balancing reserves prior to real-time; − the activation of upward and downward bids on balancing markets to effectively cope with power imbalances in the most efficient way for the system. The integration of balancing service provision within a Smart MED scheduling optimization can be achieved through different algorithmic and modelling approaches, widely discussed in the literature. This work contributed to this stream of work by introducing a multi-stage optimization framework, so-called General Advanced Intelligent Architecture (GAIA), where decision steps are linked to balancing market’s sessions timing. In the first step, namely the Long-Term Scheduling (LTS), Smart MED evaluate the opportunity of providing balancing reserve capacity through balancing market auctions held weekly. As a result of the LTS procedure, the Smart MED is awarded with a capacity retention profile, potentially different for upward and downward regulation services. In a second phase, it is necessary to define the day-ahead power exchange profile and the balancing energy bids; this step, namely the Short-Term Scheduling (STS), deals with the bidding strategy to be adopted by the Smart MED on the Italian “Day-Ahead Market” (DAM) and “Integrated Scheduling phase of the “Ancillary Services Market (IS ASM), i.e. the balancing market session held prior to real-time. While the LTS procedure adopts a deterministic optimization approach, the STS problem addresses the inherent Italian balancing markets’ uncertainty with a stochastic optimization approach based on a set of possible market’s realizations for the IS ASM, each having a given occurrence probability. Balancing market scenarios are generated through a statistical tool where a Random Forest algorithm is coupled with a Monte Carlo approach to produce expected market’s results; these market’s acceptance profiles are then clustered to obtain a set of reference daily market scenarios for the STS problem. GAIA’s framework entails a further step, namely the Re-Scheduling (RS). It intends to cope with Intra-Day Market (IDM) electricity trading and the bidding strategy for the real-time sessions of the Italian ASM, called “Balancing Market” (BM). However, this last step is not treated in this work but it is left to future developments. As balancing market designs can differ across countries in Europe, the definition of GAIA framework, of its constraints, objective functions and peculiarities needs a proper introduction about the main aspects of the context within which the work was developed. The thesis is hence organised in six chapters, where GAIA is presented and tested, applying it to a real-life case study, in Chapters 3-6. Chapter 1 introduces the concept of Smart MED as a part of the current and future energy system. Starting from an overview on the energy transition issues, the chapter focuses on European and Italian policies implemented to reach decarbonization targets fixed by the Clean Energy Package (and updated by the Green Deal more recently). The impact of these policies on power systems is evaluated looking at both liberalised (generation and consumption) and regulated (networks) sectors. Finally, some possible solutions to cope with the main issues related to power system decarbonization are discussed, analysing and emphasizing the possible contribution coming from Smart MED. Chapter 2 focuses on the role that electricity markets design has within the power system decarbonization process moving from network codes and guidelines recently approved by ENTSO-E (European Network of Transmission System Operators for Electricity) and ACER (Agency for the Cooperation of Energy Regulators). The Italian electricity market design is addressed discussing the dispatching reformation process promulged by the Italian Regulatory Authority through a recent consultation document (DCO 322/2019). The main aspects concern balancing market’s products characteristics and the role of different actors involved in power system operations. These reforms will influence the way in which resources are remunerated and interact among themselves and with the power system. Notably, the current Italian ASM does not include reserve capacity payments, requires a mandatory provision of balancing services from traditional power plants, and adopts an integrated scheduling process. The second part of Chapter 2 discusses the added value of a decentralized energy system within the power system framework depicted by EU Directives 2019/943 and 2019/944 . With reference to Smart MED, two concepts are relevant: the role of Distributed Energy Resources (DERs) and the figure of the aggregator as a Balancing Service Provider. This analysis highlights the need of a proper management of DERs for future energy system operations, hence the scientific and societal relevance of the main approaches proposed in the literature and adopted in real-life for multi-energy systems optimization. Chapter 3 presents the GAIA domain, together with the fundamental set of constraints regulating LTS and STS optimization procedures. The core of GAIA lays in the definition of the physical electricity exchange with the public grid, which results from the electricity trading carried out on short term markets (from a week-ahead to real-timeò-). Looking at a district’s energy assets first, the proposed optimization procedure models a large set of technologies: photovoltaic and solar heating plants, energy conversion facilities (internal combustion engines, heat pumps, compression chillers, absorption chillers, boilers), energy storage technologies (electricity, heat and cold), and electric vehicles. In particular, electrochemical storages are modelled through a novel approach where performances depend on both the battery’s state-of-charge and the requested charge/discharge power. Moreover, the optimization problem introduces a detailed modelling of natural gas and electricity bills: this is fundamental considering practices such as peak-shaving or load shifting, where it is important to properly value all the components of the bill (variable, power and fixed fees). Finally, the chapter specifically discusses how LTS and STS strategies were defined, presenting further constraints and the respective objective functions. In particular, the STS problem needs to deal with balancing market’s results uncertainty and addresses this issue through a stochastic approach, requiring market realization scenarios for the IS ASM as an input parameter. Chapter 4 is dedicated to the description of a statistical tool developed to generate market realization scenarios for the IS ASM. Handling pay-as-bid, centralized balancing markets forecast is a difficult exercise, since results are influenced by many variables. The chapter investigates a novel approach based on a Random Forest algorithm. Firstly, a set of input features is elaborated starting from available data concerning: power system structure, spot energy markets results in terms of volumes accepted and prices awarded, weather conditions and market participants’ bidding strategies. The market realization forecast model is then trained on data from North Italy bidding zone (looking at the case study location). The proposed model still shows some performances improvement possibility, especially because of the very high imbalance between the number of bids rejected and accepted on the Italian IS ASM. This is due to the fact that the provision of balancing resources in Italy is not voluntary, rather programmable power plants connected to the transmission network are obliged to offer their regulating capacity on the market. Hence, the Random Forest forecast is corrected through out-of-bag errors randomly sampled from the model’s training process; in this way, it is possible to fit the model on an input sample many times, obtaining different predictions, which are directly influenced by the model accuracy. Therefore, exploiting a Monte Carlo approach, it is possible to generate a set of market realization scenarios whose distribution can be used as a proxy of market results’ uncertainty. Finally, clustering techniques are employed to synthetize some reference market scenarios for the stochastic procedure. Specifically, the occurrence probability of each cluster (hence of each reference scenario) is calculated based on the cardinality of the cluster itself (i.e. considering the number of samples falling in that cluster). Chapter 5 is dedicated to the description of the case study used to test GAIA’s LTS and STS procedures: the university campus of Leonardo Da Vinci square, the headquarter of Politecnico di Milano. The Leonardo campus hosts a private MV electricity network, a district heating and a smaller district cooling network. One of the first Italian photovoltaic power plants (rated around 18 kW) was installed on one of the campus buildings in 2002. Then, in 2015, a cogeneration power plant (rated 2 MW el ) was added, together with some boilers (current installed capacity of 12 MW th ) and an absorption chiller (1.4 MW cool ). With a yearly consumption of 12 GWh of electricity, 10 GWh of heat and 5 GWh of cold, Leonardo campus is a good reference example of a Smart MED. Moreover, an analysis of the main internal policy guidelines suggests the inclusion of future installations such as electric vehicles’ charging infrastructures, thermal and electrochemical storages, and a larger photovoltaic generation capacity. Five reference weeks are selected to fairly represent the yearly activity within the campus. Thus, the testing procedure can consider alternative energy assets configurations (current and prospected installations at different times in the future) and different boundary conditions (mainly in terms of energy demands). Finally, based on the analysis of the current design of balancing markets in other European countries a reference market design is defined. This is based on short-term (weekly) balancing capacity auctions and daily energy trading on the DAM and on the IS ASM. Therefore, the LTS procedure defines the optimal balancing capacity bidding profile, while the STS problem takes care of daily bidding strategy. Chapter 6 reports the results obtained from the LTS and STS testing. The LTS procedure is tested on three main reference assets’ configurations (namely Horizon 2021, 2024 and 2026). Additionally, a sensitivity analysis on the main design parameters for weekly balancing auctions is conducted to assess the impact of different market design on Smart MED optimal operations. First, alternative prices for balancing availability are tested on both upward and downward balancing capacity provision. Second, the contracted period is moved from load-ramping periods of working days (morning and late afternoon) to the weekend. Last, the minimum duration required for the balancing service is changed, mainly influencing the exploitation of limited energy content technologies. The second part of the chapter is dedicated to the STS procedure. STS testing focuses on two specific days: a winter and a summer day. First, the STS optimization problem is solved through a deterministic approach; then, a stochastic procedure is applied to find the optimal market scheduling (DAM program and IS ASM bids) based on a set of possible IS ASM market scenarios. Using a real market realization, randomly sampled from all possible scenarios, the two approaches are compared in terms of: capability to respect upward and downward balancing energy provision, imbalance volume realised, and value of the objective function. The thesis finally reports the main conclusions of the work. In particular, the main contributions of the work can be gathered into three main categories. From a technical point of view, the proposed model demonstrates the possibility for system operators to exploit Smart MED for procuring (balancing capacity) and providing (balancing energy) the resources that they need to run the power system safely. The proposed optimisation approach, based on a mix of physical and data-driven instruments, makes use of Information and Communication Technologies (ICT) and computational technologies to enable Smart MED for balancing services provision in an optimized manner. − From an economic standpoint, it is necessary to consider that the provision of balancing services often entails an opportunity cost. This work sheds light on the fact that, differently from conventional power plants, a Smart MED mainly produces energy for self-consumption, while selling electricity has historically been only a side-business; because of this, it is important that balancing market are designed to offer a fair and technology neutral remuneration framework, ensuring the participation of traditional and new resources, and enabling system operators to employ a variety of resources to balance the system at the least cost. − Looking at policy implications, results show the importance of short-term balancing capacity auctions to guarantee market revenues for distributed resources when they provide balancing services; also, they suggest that price cap or floor on balancing markets limit the possibility for some units to recover their costs and the capability of balancing markets to send proper imbalance price signals to all resources. The above contributions address a similar issue. Power systems are becoming more interconnected and are expected to accommodate more intermittent renewable resources. Thanks to digitalization and sector coupling, also the potential pool of flexible units gets wider and wider. Creating a fair, technology-neutral market environment could enhance the capability of many new resources to provide flexibility to the power system, while pushing towards more efficient and sustainable configurations for multi-energy districts.
Il lavoro di tesi si occupa della possibilità di integrare l'ottimizzazione delle operazioni degli Smart Multi Energy Districts (Smart MEDs) con la fornitura di risorse di dispacciamento alla rete pubblica sotto forma di regolazione della frequenza e servizi di bilanciamento. Questo si ottiene introducendo un quadro di ottimizzazione a più stadi, la cosiddetta General Advanced Intelligent Architecture (GAIA), dove le fasi decisionali sono legate alla tempistica delle sessioni del mercato di bilanciamento. Lo Smart MED dovrebbe innanzitutto valutare il costo-opportunità di fornire risorse di capacità di bilanciamento all'interno delle aste del mercato di bilanciamento tenute ogni settimana (Long-Term Scheduling). I risultati delle aste di mercato impongono un livello di mantenimento della capacità al problema del day-ahead (Short-Term Scheduling). Il suo obiettivo è duplice: primo, calcolare il programma di scambio ottimale con la rete pubblica, definendo così le offerte di acquisto o vendita da presentare nelle sessioni del Day-Ahead Market (DAM); secondo, impostare il profilo di offerta ottimale per la fase di programmazione integrata dell'Ancillary Services Market (IS ASM). A differenza del problema LTS, quello STS è affrontato con un approccio stocastico. Gli scenari di realizzazione del mercato per il problema stocastico sono generati attraverso un nuovo strumento statistico, applicato ai mercati di bilanciamento italiani, dove un algoritmo Random Forest è accoppiato con un approccio Monte Carlo per produrre i risultati attesi del mercato, che sono poi raggruppati per ottenere un set di scenari di mercato di riferimento per il problema STS. La struttura di GAIA comporta un ulteriore passo di ottimizzazione, vale a dire il Re-Scheduling (RS). Questo intende far fronte all'offerta ottimale del mercato intra-giornaliero (IDM) e del mercato di bilanciamento (BM). Tuttavia, quest'ultima fase non è elaborata all'interno del lavoro, ma è lasciata agli sviluppi futuri. Il documento è organizzato come segue. Il capitolo 1 introduce il concetto di Smart MED, includendolo nel sistema energetico attuale e futuro. Partendo da una panoramica generale della transizione energetica, il capitolo si concentra sulle politiche europee e italiane attuate per raggiungere gli obiettivi di decarbonizzazione fissati dal Clean Energy Package (e aggiornati dal Green Deal più recentemente). L'impatto di queste politiche sui sistemi elettrici viene valutato guardando sia ai settori liberalizzati (generazione e consumo) che a quelli regolati (reti). Infine, vengono discusse alcune possibili soluzioni per far fronte alle principali questioni relative alla decarbonizzazione del sistema elettrico, analizzando ed enfatizzando il contributo proveniente dagli Smart MED. Il capitolo 2 si concentra sul ruolo che l'evoluzione dei mercati elettrici ha all'interno del processo di decarbonizzazione del sistema elettrico. La prospettiva europea viene descritta muovendo dai codici di rete e dalle linee guida recentemente approvate, in particolare per quanto riguarda la progettazione dei mercati di bilanciamento. Viene poi introdotto il disegno dei mercati spot italiani dell'elettricità. Inerentemente, l'Autorità di regolazione italiana ha recentemente (2019) definito in un documento di consultazione (DCO 322/2019) le principali intenzioni in merito alla riforma dei mercati elettrici: questa viene quindi discussa trattando sia le caratteristiche dei prodotti del mercato che i ruoli dei diversi attori del sistema elettrico coinvolti. Il resto del capitolo è dedicato a un'analisi approfondita di due concetti: Risorse Energetiche Distribuite (DERs) e aggregatori. Viene presentato il valore aggiunto di un sistema energetico decentralizzato, insieme ai vantaggi e agli svantaggi della struttura del sistema elettrico prevista dalle direttive UE. Questa discussione porta all'importanza di una corretta gestione delle DERs, evidenziando i diversi approcci disponibili in letteratura e nei casi reali, permettendo così di introdurre l'approccio modellistico seguito nel lavoro di tesi nei capitoli successivi. Il capitolo 3 presenta il dominio di GAIA, insieme all'insieme fondamentale di vincoli che regolano la sua procedura di ottimizzazione. Il nucleo di GAIA risiede nella definizione dello scambio fisico di elettricità con la rete pubblica, che risulta dallo scambio commerciale di energia realizzato nei mercati spot. Per quanto riguarda gli asset energetici del distretto, la procedura di ottimizzazione può ospitare: impianti fotovoltaici e solari termici, impianti di conversione energetica (motori a combustione interna, pompe di calore, refrigeratori a compressione, refrigeratori ad assorbimento, caldaie), tecnologie di stoccaggio dell'energia (elettricità, calore e freddo) e veicoli elettrici. In particolare, gli accumulatori elettrochimici sono modellati attraverso un nuovo approccio in cui le prestazioni dipendono sia dallo stato di carica della batteria che dalla potenza di carica/scarica richiesta. Inoltre, il problema dell'ottimizzazione introduce una modellazione dettagliata delle bollette del gas naturale e dell'elettricità: questo è fondamentale considerando pratiche come il peak-shaving o il load shifting, dove è importante valutare correttamente tutte le componenti della bolletta (variabile, potenza e quota fissa).La fine del capitolo discute specificamente come sono stati affrontati i problemi di schedulazione a lungo termine e a breve termine, presentando ulteriori vincoli e le rispettive funzioni obiettivo. In particolare, l'ottimizzazione STS si basa su un approccio stocastico, quindi necessita di alcuni scenari di realizzazione del mercato per impostare correttamente la procedura stocastica. Quindi, il capitolo 4 è dedicato alla descrizione dello strumento statistico sviluppato per generare scenari di realizzazione del mercato. Gestire le previsioni dei mercati di bilanciamento centralizzati pay-as-bid è un esercizio difficile, poiché i loro risultati sono influenzati da molte variabili. Il capitolo studia un nuovo approccio basato su un algoritmo Random Forest. In primo luogo, viene elaborato un set di caratteristiche di input a partire dai dati disponibili riguardanti: la struttura del sistema elettrico, i risultati dei mercati dell'energia, le condizioni meteorologiche e le strategie di offerta dei partecipanti al mercato. Il modello di previsione della realizzazione del mercato viene poi addestrato sui dati della zona di offerta del Nord Italia (che rappresenta il caso di studio). Il modello sviluppato mostra ancora alcune possibilità di miglioramento delle prestazioni, soprattutto a causa del forte squilibrio tra il numero di offerte rifiutate e accettate sul mercato italiano dei servizi ausiliari. Questo è dovuto al fatto che la fornitura di risorse di dispacciamento in Italia non è volontaria, piuttosto le centrali programmabili connesse alla rete di trasmissione sono obbligate ad offrire la loro capacità di regolazione sul mercato. Per far fronte alle prestazioni del modello e sfruttarlo per generare scenari di realizzazione del mercato per il problema stocastico STS, viene applicato un nuovo approccio. La previsione di Random Forest viene corretta attraverso errori out-of-bag campionati casualmente dal processo di addestramento del modello; in questo modo, è possibile adattare il modello su un campione di input molte volte, ottenendo diverse previsioni, che sono direttamente influenzate dall'accuratezza del modello. Quindi, sfruttando un approccio Monte Carlo, è possibile generare un insieme di scenari di realizzazione del mercato, la cui distribuzione può essere utilizzata come proxy dell'incertezza dei risultati di mercato. Infine, per sintetizzare alcuni scenari di mercato di riferimento per la procedura stocastica, vengono sfruttate tecniche di clustering, e la probabilità di accadimento di ogni cluster (quindi di ogni scenario di riferimento) viene calcolata in base alla cardinalità del cluster stesso. Il capitolo 5 è dedicato alla descrizione del caso studio utilizzato per testare le procedure LTS e STS di GAIA. Si tratta del campus universitario di Leonardo, sede del Politecnico di Milano. Il campus Leonardo ospita una rete elettrica privata a media tensione, una rete di teleriscaldamento e una più piccola rete di teleraffrescamento. Uno dei primi impianti fotovoltaici italiani (circa 18 kW) è stato installato su uno degli edifici del campus nel 2002. Poi, nel 2015, è stata installata una centrale di cogenerazione (2 MWel), insieme ad alcune caldaie (capacità installata attuale di 12 MWth) e un refrigeratore ad assorbimento (1,4 MWcool). Con un consumo annuale di 12 GWh di elettricità, 10 GWh di calore e 5 GWh di freddo, il campus Leonardo è uno dei principali Smart MED italiani nel settore terziario. Il suo grado di complessità è destinato ad aumentare nei prossimi anni grazie all'installazione di: punti di ricarica dei veicoli elettrici, accumulatori termici ed elettrochimici e una rilevante capacità di generazione fotovoltaica. Sulla base di queste informazioni, sono state selezionate cinque settimane di riferimento per rappresentare equamente l'attività annuale all'interno del campus. Quindi, la procedura di test è stata condotta considerando diversi asset energetici (installazioni attuali e future) e diverse condizioni limite (principalmente in termini di richieste di energia). Un'altra importante ipotesi riguarda la struttura del mercato di bilanciamento. Considerando le direttive e i regolamenti UE, e guardando all'attuale design dei mercati europei, è stato definito un quadro di riferimento per il mercato di bilanciamento. Si basa su aste di capacità di bilanciamento a breve termine (settimanali) e scambi di energia di bilanciamento giornalieri. La procedura LTS definirà il profilo ottimale delle offerte di capacità di bilanciamento, mentre il problema STS si occuperà delle offerte giornaliere presentate sulla piattaforma di bilanciamento dell'energia. Infine, il capitolo 6 riporta i risultati ottenuti dai test LTS e STS. La procedura LTS è testata su tre principali configurazioni di asset di riferimento (ovvero Horizon 2021, 2024 e 2026). Inoltre, è inclusa un'analisi di sensibilità sui principali parametri che caratterizzano le aste settimanali di bilanciamento. In primo luogo, vengono testati diversi prezzi per la disponibilità di bilanciamento sia al rialzo che al ribasso. In secondo luogo, il periodo di disponibilità del bilanciamento viene spostato dai periodi di load-ramping dei giorni lavorativi (mattina e tardo pomeriggio) al fine settimana. Infine, viene modificata la durata minima richiesta per il servizio di bilanciamento, influenzando lo sfruttamento delle tecnologie a contenuto energetico limitato. La seconda parte del capitolo è dedicata alla procedura STS. Il test STS si concentra su due giorni specifici: un giorno invernale (16 dicembre) e uno estivo (24 giugno). In primo luogo, il problema di ottimizzazione STS viene risolto attraverso un approccio deterministico; poi, una procedura stocastica viene applicata per trovare la programmazione ottimale del mercato (programma DAM e offerte ASM) sulla base di un insieme di possibili scenari di mercato di bilanciamento. Considerando una realizzazione reale del mercato, campionata casualmente da tutti i possibili scenari, i due approcci sono confrontati in termini di: capacità di rispettare le chiamate di bilanciamento verso l'alto e verso il basso, volume di squilibrio realizzato e valore della funzione obiettivo. Quindi, l'ultimo capitolo è dedicato a una discussione delle principali evidenze del lavoro di tesi.
United we stand : how aggregates of distributed energy resources can shape the future energy system
Bovera, Filippo
2020/2021
Abstract
The thesis work addresses the integration of balancing services provision within the optimization of Smart Multi Energy Districts (Smart MED) operations. Traditionally, efforts of both researchers and professionals have been concentrated in obtaining the maximum energy efficiency from multi-energy districts operations while satisfying their internal energy demands. The advent of non-programmable renewables (NP-RES) and the decentralization of electricity production have challenged this view. Because of this, it is fundamental to develop computational instruments able to couple districts’ energy demands with balancing needs of the external public grid. Since early 90s, resources for electricity dispatching have been collected on electricity markets; these are usually composed by: − a day-ahead energy trading, where electricity demand and production first meet, − an intra-day electricity exchange, mainly exploited to correct power generation profiles according to unexpected events, and − one or more markets where ancillary services are purchased by system operators to maintain the exact frequency and voltage levels along power grids. This market sequence generates a series of price signals that push production (and consumption) units to modify their scheduled operations, their bidding strategies and their business models. The recent massive diffusion of NP-RES plants has shifted the attention from day-ahead trading, where prices are influenced by power plants marginal costs, to real-time markets, which are instead influenced by network conditions and system imbalances. This work focuses on balancing markets as enablers for the provision of balancing resources from Smart MED. The procurement of these resources is based on two fundamental steps: − the creation of an adequate level of balancing reserves prior to real-time; − the activation of upward and downward bids on balancing markets to effectively cope with power imbalances in the most efficient way for the system. The integration of balancing service provision within a Smart MED scheduling optimization can be achieved through different algorithmic and modelling approaches, widely discussed in the literature. This work contributed to this stream of work by introducing a multi-stage optimization framework, so-called General Advanced Intelligent Architecture (GAIA), where decision steps are linked to balancing market’s sessions timing. In the first step, namely the Long-Term Scheduling (LTS), Smart MED evaluate the opportunity of providing balancing reserve capacity through balancing market auctions held weekly. As a result of the LTS procedure, the Smart MED is awarded with a capacity retention profile, potentially different for upward and downward regulation services. In a second phase, it is necessary to define the day-ahead power exchange profile and the balancing energy bids; this step, namely the Short-Term Scheduling (STS), deals with the bidding strategy to be adopted by the Smart MED on the Italian “Day-Ahead Market” (DAM) and “Integrated Scheduling phase of the “Ancillary Services Market (IS ASM), i.e. the balancing market session held prior to real-time. While the LTS procedure adopts a deterministic optimization approach, the STS problem addresses the inherent Italian balancing markets’ uncertainty with a stochastic optimization approach based on a set of possible market’s realizations for the IS ASM, each having a given occurrence probability. Balancing market scenarios are generated through a statistical tool where a Random Forest algorithm is coupled with a Monte Carlo approach to produce expected market’s results; these market’s acceptance profiles are then clustered to obtain a set of reference daily market scenarios for the STS problem. GAIA’s framework entails a further step, namely the Re-Scheduling (RS). It intends to cope with Intra-Day Market (IDM) electricity trading and the bidding strategy for the real-time sessions of the Italian ASM, called “Balancing Market” (BM). However, this last step is not treated in this work but it is left to future developments. As balancing market designs can differ across countries in Europe, the definition of GAIA framework, of its constraints, objective functions and peculiarities needs a proper introduction about the main aspects of the context within which the work was developed. The thesis is hence organised in six chapters, where GAIA is presented and tested, applying it to a real-life case study, in Chapters 3-6. Chapter 1 introduces the concept of Smart MED as a part of the current and future energy system. Starting from an overview on the energy transition issues, the chapter focuses on European and Italian policies implemented to reach decarbonization targets fixed by the Clean Energy Package (and updated by the Green Deal more recently). The impact of these policies on power systems is evaluated looking at both liberalised (generation and consumption) and regulated (networks) sectors. Finally, some possible solutions to cope with the main issues related to power system decarbonization are discussed, analysing and emphasizing the possible contribution coming from Smart MED. Chapter 2 focuses on the role that electricity markets design has within the power system decarbonization process moving from network codes and guidelines recently approved by ENTSO-E (European Network of Transmission System Operators for Electricity) and ACER (Agency for the Cooperation of Energy Regulators). The Italian electricity market design is addressed discussing the dispatching reformation process promulged by the Italian Regulatory Authority through a recent consultation document (DCO 322/2019). The main aspects concern balancing market’s products characteristics and the role of different actors involved in power system operations. These reforms will influence the way in which resources are remunerated and interact among themselves and with the power system. Notably, the current Italian ASM does not include reserve capacity payments, requires a mandatory provision of balancing services from traditional power plants, and adopts an integrated scheduling process. The second part of Chapter 2 discusses the added value of a decentralized energy system within the power system framework depicted by EU Directives 2019/943 and 2019/944 . With reference to Smart MED, two concepts are relevant: the role of Distributed Energy Resources (DERs) and the figure of the aggregator as a Balancing Service Provider. This analysis highlights the need of a proper management of DERs for future energy system operations, hence the scientific and societal relevance of the main approaches proposed in the literature and adopted in real-life for multi-energy systems optimization. Chapter 3 presents the GAIA domain, together with the fundamental set of constraints regulating LTS and STS optimization procedures. The core of GAIA lays in the definition of the physical electricity exchange with the public grid, which results from the electricity trading carried out on short term markets (from a week-ahead to real-timeò-). Looking at a district’s energy assets first, the proposed optimization procedure models a large set of technologies: photovoltaic and solar heating plants, energy conversion facilities (internal combustion engines, heat pumps, compression chillers, absorption chillers, boilers), energy storage technologies (electricity, heat and cold), and electric vehicles. In particular, electrochemical storages are modelled through a novel approach where performances depend on both the battery’s state-of-charge and the requested charge/discharge power. Moreover, the optimization problem introduces a detailed modelling of natural gas and electricity bills: this is fundamental considering practices such as peak-shaving or load shifting, where it is important to properly value all the components of the bill (variable, power and fixed fees). Finally, the chapter specifically discusses how LTS and STS strategies were defined, presenting further constraints and the respective objective functions. In particular, the STS problem needs to deal with balancing market’s results uncertainty and addresses this issue through a stochastic approach, requiring market realization scenarios for the IS ASM as an input parameter. Chapter 4 is dedicated to the description of a statistical tool developed to generate market realization scenarios for the IS ASM. Handling pay-as-bid, centralized balancing markets forecast is a difficult exercise, since results are influenced by many variables. The chapter investigates a novel approach based on a Random Forest algorithm. Firstly, a set of input features is elaborated starting from available data concerning: power system structure, spot energy markets results in terms of volumes accepted and prices awarded, weather conditions and market participants’ bidding strategies. The market realization forecast model is then trained on data from North Italy bidding zone (looking at the case study location). The proposed model still shows some performances improvement possibility, especially because of the very high imbalance between the number of bids rejected and accepted on the Italian IS ASM. This is due to the fact that the provision of balancing resources in Italy is not voluntary, rather programmable power plants connected to the transmission network are obliged to offer their regulating capacity on the market. Hence, the Random Forest forecast is corrected through out-of-bag errors randomly sampled from the model’s training process; in this way, it is possible to fit the model on an input sample many times, obtaining different predictions, which are directly influenced by the model accuracy. Therefore, exploiting a Monte Carlo approach, it is possible to generate a set of market realization scenarios whose distribution can be used as a proxy of market results’ uncertainty. Finally, clustering techniques are employed to synthetize some reference market scenarios for the stochastic procedure. Specifically, the occurrence probability of each cluster (hence of each reference scenario) is calculated based on the cardinality of the cluster itself (i.e. considering the number of samples falling in that cluster). Chapter 5 is dedicated to the description of the case study used to test GAIA’s LTS and STS procedures: the university campus of Leonardo Da Vinci square, the headquarter of Politecnico di Milano. The Leonardo campus hosts a private MV electricity network, a district heating and a smaller district cooling network. One of the first Italian photovoltaic power plants (rated around 18 kW) was installed on one of the campus buildings in 2002. Then, in 2015, a cogeneration power plant (rated 2 MW el ) was added, together with some boilers (current installed capacity of 12 MW th ) and an absorption chiller (1.4 MW cool ). With a yearly consumption of 12 GWh of electricity, 10 GWh of heat and 5 GWh of cold, Leonardo campus is a good reference example of a Smart MED. Moreover, an analysis of the main internal policy guidelines suggests the inclusion of future installations such as electric vehicles’ charging infrastructures, thermal and electrochemical storages, and a larger photovoltaic generation capacity. Five reference weeks are selected to fairly represent the yearly activity within the campus. Thus, the testing procedure can consider alternative energy assets configurations (current and prospected installations at different times in the future) and different boundary conditions (mainly in terms of energy demands). Finally, based on the analysis of the current design of balancing markets in other European countries a reference market design is defined. This is based on short-term (weekly) balancing capacity auctions and daily energy trading on the DAM and on the IS ASM. Therefore, the LTS procedure defines the optimal balancing capacity bidding profile, while the STS problem takes care of daily bidding strategy. Chapter 6 reports the results obtained from the LTS and STS testing. The LTS procedure is tested on three main reference assets’ configurations (namely Horizon 2021, 2024 and 2026). Additionally, a sensitivity analysis on the main design parameters for weekly balancing auctions is conducted to assess the impact of different market design on Smart MED optimal operations. First, alternative prices for balancing availability are tested on both upward and downward balancing capacity provision. Second, the contracted period is moved from load-ramping periods of working days (morning and late afternoon) to the weekend. Last, the minimum duration required for the balancing service is changed, mainly influencing the exploitation of limited energy content technologies. The second part of the chapter is dedicated to the STS procedure. STS testing focuses on two specific days: a winter and a summer day. First, the STS optimization problem is solved through a deterministic approach; then, a stochastic procedure is applied to find the optimal market scheduling (DAM program and IS ASM bids) based on a set of possible IS ASM market scenarios. Using a real market realization, randomly sampled from all possible scenarios, the two approaches are compared in terms of: capability to respect upward and downward balancing energy provision, imbalance volume realised, and value of the objective function. The thesis finally reports the main conclusions of the work. In particular, the main contributions of the work can be gathered into three main categories. From a technical point of view, the proposed model demonstrates the possibility for system operators to exploit Smart MED for procuring (balancing capacity) and providing (balancing energy) the resources that they need to run the power system safely. The proposed optimisation approach, based on a mix of physical and data-driven instruments, makes use of Information and Communication Technologies (ICT) and computational technologies to enable Smart MED for balancing services provision in an optimized manner. − From an economic standpoint, it is necessary to consider that the provision of balancing services often entails an opportunity cost. This work sheds light on the fact that, differently from conventional power plants, a Smart MED mainly produces energy for self-consumption, while selling electricity has historically been only a side-business; because of this, it is important that balancing market are designed to offer a fair and technology neutral remuneration framework, ensuring the participation of traditional and new resources, and enabling system operators to employ a variety of resources to balance the system at the least cost. − Looking at policy implications, results show the importance of short-term balancing capacity auctions to guarantee market revenues for distributed resources when they provide balancing services; also, they suggest that price cap or floor on balancing markets limit the possibility for some units to recover their costs and the capability of balancing markets to send proper imbalance price signals to all resources. The above contributions address a similar issue. Power systems are becoming more interconnected and are expected to accommodate more intermittent renewable resources. Thanks to digitalization and sector coupling, also the potential pool of flexible units gets wider and wider. Creating a fair, technology-neutral market environment could enhance the capability of many new resources to provide flexibility to the power system, while pushing towards more efficient and sustainable configurations for multi-energy districts.File | Dimensione | Formato | |
---|---|---|---|
PhDThesis_Bovera_rev5.pdf
Open Access dal 09/07/2022
Dimensione
12.23 MB
Formato
Adobe PDF
|
12.23 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/176383