Multi-Agent Reinforcement Learning Based Optimal PV-ESS Control In Grid

Abstract

The increasing utilization of renewable energy sources, such as photovoltaic (PV) power, has led to a growing interest in managing surplus PV power in order to generate additional profits. In particular, the use of energy storage systems (ESS) for handling surplus PV power has gained significant attention due to their ability to control the unstable and erratic nature of solar power systems. This paper presents an optimal ESS control scheme based on multi-agent reinforcement learning (MARL) that maximizes grid benefits. The proposed method is evaluated in a grid environment that includes a central ESS, multiple PV power prosumers, and consumers. The results of our empirical study demonstrate that the proposed method generates an additional profit of 18% to 36% compared to the current method used by Korean power providers for calculating prosumer profits. Furthermore, we discovered was found that as the proportion of prosumers in the total population increases, energy efficiency also increases proportionally.