Non-cooperative game theoretical control for green and efficient energy communities

Energy communities are socially driven initiatives that emphasize collective participation in energy production, distribution, and consumption. Differently from smart grids, which focus on the infrastructural and technological side, energy communities are concerned with the market and economical incentive design, aiming at guaranteeing a socially, environmentally, and economically sustainable energy grid. Thus, the study of energy communities is tightly intertwined with the analysis of the behavior that arises when several agents are faced with conflicting needs and resource scarcity. Non-cooperative game theory has proved to be a solid tool for tackling the challenge of optimally controlling energy communities: this dissertation aims to contribute to the topic by addressing aspects such as transactive market design, plug-in electric vehicles (PEVs) integration, and learning-based decision-making. In particular, the first part explores the economical and operational design of energy communities from the transactive perspective. Such a term refers to relational patterns between two or more agents that are based on “transactions”, i.e., on the exchange of one or more commodities or services. Such a setup can be effectively studied under the lens of game theory since any transaction, in order to be successful, requires two or more parties to agree on the quantities to be exchanged. Energy markets are perfectly suitable to be considered as transactive environments, because of the ubiquitous need that agents have to exchange energy. In particular, the analysis focuses on the modelling of energy communities with independent and selfish agents, as well as the subsequent design of decentralized and distributed schemes for the equilibrium seeking of the arising non-cooperative game. Among the diverse actors that characterize the modern energy community, PEVs are the ones that, in the last decade, have caused a consistent push towards a more decentralized and dynamic energy market system. This is due to their inherent unpredictability with respect to their energy demand, which serves the purpose of recharging their batteries. Such a unilateral energy exchange gives rise to the V1G-based energy market. However, being equipped with storage devices, PEVs can be considered as non-static ESSs, which can be used by the grid operators for tasks such as voltage and frequency regulation, but also by prosumers and actors that can experience temporary energy surpluses. Such a bilateral exchange is captured by the V2B and – in the most general sense – V2X protocols. The second part of this dissertation frames PEVs as non-cooperative agents, which participate in the energy market with the aim of recharging their batteries in the most economically efficient way, or providing temporary storage service for a fee. The two aforementioned research directions frame the agents in the energy community as rational entities, whose prerogatives are described by an optimization problem constituted by a certain cost function to minimize and operational constraints to satisfy. Such an approach is reasonable when modelling a perfectly rational entity, e.g., control systems that solely follow its instructions. However, as many grid actors are backed by the decisions and actions made by people, the perfect “rationality assumption” becomes unrealistic: people are often driven by habits and belief systems that do not necessarily yield the “optimal” decisions. The third part collects the result of a work-in-progress which aims at defining a learning-based equilibrium – and its related seeking methods – where the agents’ behavior is not modelled by an optimization problem, where its objective expresses some cost (utility) to minimize (maximize), but through a neural network. The idea is to capture behavioral patterns on the basis of existing data, representing the agents’ response to the environment and the other players’ strategies.