Power Equilibrating of Multi-microgrids in a Market Environment: A Multi-agent Shadow-Price Deep Reinforcement Learning Based approach

With the enrichment of demand-side resources such as distributed renewable energy in modern distribution networks, the disorderly power consumption and generation behaviors of users gradually causes security problems such as equipment overload and backward flow. Therefore, it is necessary to design reasonable guiding measures to guide user groups to achieve real-time supply-demand matching, promote the local consumption of distributed renewable energy, and achieve power balance in distribution systems. However, the privacy requirements of different operators of distribution network and microgrids make it difficult for them to engage in a complete information game. The future opportunity costs of equipment with time-coupled operating characteristics, such as energy storage, should also be taken into account. For these reasons, this paper proposes a multi-agent shadow-price deep reinforcement learning approach, which forms a data isolation between different subjects and gives interaction signals of price in the market environment, so as to realize collaborative decision-making between the above operators. Case studies on a 10-microgird distribution network test system validates of the effectiveness of the proposed multi-agent shadow-price deep reinforcement learning method.