Non-cooperative Game-Based Power Trading Strategy for Multi-Microgrid based on Multi-Agent Deep Reinforcement Learning

With the increasing proportion of renewable energy sources with volatility and intermittency in the power grid, the balance between power supply and demand is facing severe challenges. The multi-microgrid (MG) system plays a significant role in maintaining power balance within the grid. Therefore, based on an improved Nash-Q multi-agent deep reinforcement learning algorithm, a multi-agent game-based power trading strategy for multi-MG is proposed. To achieve power sharing among multi-MG and reduce the operating costs of individual MG, this paper first considers the operational characteristics of different MGs and designs a peer-to-peer (P2P) power sharing-based power trading model for multi-MG. Due to the coupling of power trading strategies among MGs, the operating costs of each MG form a non-cooperative game. Then the improved Nash-Q algorithm is adopted to solve this game-theoretic problem. Finally, case studies demonstrate the proposed model can effectively enhance the MG’s operational economic efficiency and achieve power balance.