Multiagent Reinforcement Learning With Learning Automata for Microgrid Energy Management and Decision Optimization

As the increasing willingness of electric users to actively participate in power scheduling and to pursue self-interest, the management and optimization of residential microgrid confront higher requirements to balance the tradeoff between overall operational objectives and individual rights; and to resolve the influence of various uncertainties. Therefore, a multiagent reinforcement learning (MARL) approach is proposed in this paper for auction-based microgrid market. Distributed model-free reinforcement learning is used for each supplier and user to make reasonable market strategies; on the other hand, equilibrium-based game theory is combined in the learning process to ensure utility balance and supply-demand balance of the whole microgrid. Besides, to guarantee the efficiency of MARL, a learning automata (LA) is introduced to improve the strategy selection procedure which plays an essential role in algorithm optimization. A case study about microgrid market operation is conducted to verify the performance of the proposed approach.