基于多智能体强化学习的社区共享储能-光伏系统用于电动汽车负荷管理

Abstract The rapid growth of electric vehicles (EVs) is an unavoidable trend within the global energy transition. However, the substantial integration of EVs poses significant challenges to the stability and reliability of power systems . This study proposes mitigating EV load through community-shared energy storage and photovoltaic (ES-PV) systems. Within the framework of multi-agent reinforcement learning (MARL), multiple decision-making agents collaborate to manage various variables and systems in community, including energy storage charging and discharging strategies, intelligent EV charging strategies, and ES-PV system electricity pricing strategies. The coordination and optimization achieved through MARL enable these strategies to address the interdependencies and dynamic changes of the variables, thereby enhancing overall performance. Case studies in real-world scenarios demonstrate that ES-PV systems can support up to 38.68 % of EV load, increase photovoltaic self-consumption rates by 66.41 %, and significantly reduce community reliance on the distribution grid. In terms of economic performance, implementing the ES-PV system reduced community electricity expenses by up to 7.73 %, resulting in a net profit of €51,924.65 for the ES-PV system in summer. This indicates a win-win solution for both community residents and ES-PV system operators. Therefore, this framework can support a more efficient and resilient community energy utilization paradigm, accommodating the increasing prevalence of EVs and the rapid development of smart communities.