Energy management systems for demand response in power distribution systems including prosumers and market participants

Decarbonisation and climate change mitigation have become priorities for many countries and governments. The growth of renewable generation sources in the electricity system can help achieve these goals, as well as the spread of electric vehicles (EVs) as prosumers in the power system. EVs served by Charging Stations (CSs) represent an additional load to the power system to be satisfied, and an additional storage system in the case of vehicle-to-grid (V2G) technology. However, this transition necessitates the development of new policies and energy management systems to coordinate various actors in the energy market, manage intermittent production and loads, and implement demand response (DR) actions. This thesis focuses on applying optimization methods and control techniques to power systems. The use of optimization techniques promotes the efficient management of the available energy resources, minimizing the operational costs and reducing the environmental impact. The development of new control strategies allows the design of new policies where emerging technologies, such as EVs, can play a key role in the system. This thesis introduces novel control strategies that are adapted to the case study. Besides, different shortcomings based on the work developed have been addressed. In such systems, EVs, traditional generators, and prosumers, among other entities, collaborate coordinately. These active elements are grouped into aggregators to attend to the necessities of the system. The main innovation of this point is the development of a multi-level optimization problem for the balancing market in which electric vehicles participate actively. The developed optimization problems are intended to be inserted in Energy Management Systems (EMSs), to identify the optimal scheduling of production and storage systems. In addition, the aim is to minimize costs, power losses, and CO2 emissions while satisfying energy demands. The developed optimization problems face the problem of defining demand response strategies in the balancing market. To this aim, it is necessary to consider different levels of decisions and different solution techniques. In particular, at the higher level, it is frequently necessary to coordinate different agents by aggregators in the energy market to provide demand response. Once the aggregator has decided the contribution of each agent, the local system (microgrids, buildings, charging parks) must manage its components, minimising costs and emissions, taking into account uncertainties and the presence of different controllers. All these issues have been formalized as optimization problems and presented in this thesis. The aggregators must coordinate with stochastic variables, such as renewable energies or demand, to supply the power system's needs and lead the system to a stable and optimal working point. For that reason, particular attention is also devoted to the analysis of possible uncertainties and the use of machine learning for the prediction of renewables and demands. A stochastic optimization problem has been derived for polygeneration microgrids. As regards uncertainties and stochastic optimization, a new approach for generating scenarios has been developed during the research period of this thesis, which makes use of an evolutionary algorithm to generate scenarios and calculates the probability associated with each scenario based on weather conditions. In order to explore different control strategies where the case of study requires higher computational processing, distributed architectures have been studied. This thesis presents research developed in collaboration with CISCO, in which a distributed architecture is proposed for optimal scheduling. Distributed approaches promote robust and parallel computing. CISCO provided smart device controllers to be installed in the field and test the algorithm in real environments. Furthermore, cybersecurity vulnerabilities referred to system integrity found in the development of the work have been addressed. The infrastructure adopted makes it an ideal target for this kind of vulnerability. For that reason, a novel algorithm based on physics-informed machine learning has been adopted into an autoencoder architecture to detect possible attacks, which addresses another line of research.