Coordinated Optimization Dispatching Method for Microgrids and Distribution Networks Based on Multi-agent Game and Reinforcement Learning

With the promotion of the "coal to electricity" project, the rural weak distribution networks may face the problems such as line congestion due to the increase in heating loads, the long payback period of power grid investment as well as low utilization rate of power grid assets during the non-heating season in the traditional line expansion scheme. It is an effective means to solve these problems that establish a microgrid that has the ability of friendly interaction and demand response based on the local renewable energy resource. For that reason, a coordinated optimization method for microgrids and distribution networks based on a multi-agent game and reinforcement learning was proposed for a practical microgrid project in Hebei Province. Each microgrid reports the optimal bidding strategy based on the intra-day rolling dispatching plan, considering the impact of the auxiliary service strategy on the future operation economy in the finite time domain; Taking minimizing the operation cost as an object, the distribution network determines the clearing scheme of the auxiliary service market based on the power demand of the auxiliary service and the quotation of each microgrid. Finally, The win or learn fast - policy Hill-Clipping(abbr. WoLF-PHC) algorithm was used to realize the fast solution of multi-agent game problems, and the effectiveness of the proposed method was verified based on the actual case of the micro-electric network group demonstration project.