Deep Reinforcement Learning Based Coalition Formation for Energy Trading in Smart Grid

Peer-to-peer energy trading is a promising approach to better integrate renewable energy resources, reduce customer costs and increase the reliability of the smart grid by employing microgrids and allowing them to share their surplus energy with each other using 5G-enabled communications. However, the varying nature of the generation and the demand of each microgrid impose a dynamicity and uncertainty on the system. In this paper, we address the problem of minimizing cost in the coalitional microgrid communities considering the dynamic nature of the system. We propose a deep reinforcement learning approach that helps to minimize the total cost through forming efficient coalitions. The results show 16% to 30% improvement in terms of cost minimization compared to an existing Q-learning-based scheme and a conventional coalitional game theory (CG)-based approach from the literature, respectively.