An Evolutionary Reinforcement Learning Method for Multi-energy Microgrid Energy Management

Multi-energy microgrids (MEMG) play a critical role in managing distributed energy resources in modern power systems containing many renewable energy sources. Deep Reinforcement Learning (DRL) has demonstrated significant potential in energy management problems in MEMG. However, DRL methods still have some critical limitations, including insufficient exploration and unstable training, in handling the uncertainties and dynamics of MEMG systems. In this paper, we propose a mutation selection evolutionary reinforcement learning algorithm (MSERL) to solve the MEMG management problem more effectively and efficiently. MSERL integrates an evolutionary algorithm (EA) with DRL for better exploration. Two mutation operators are applied with a simple selection strategy, catering to enhance the exploration and exploitation abilities of the algorithm, respectively. Experimental results on three real-world datasets demonstrate that MSERL has a fast convergence speed and can generate very good and robust management policies in complex and uncertain MEMG environments.