Reinforcement learning-based distributed secondary optimal control for multi-microgrids

In this study, a novel reinforcement learning (RL)-based distributed secondary optimal control (DSOC) is proposed to facilitate the secondary control of droop-controlled multi-micorgrids (MMGs). The proposed scheme is implemented based on multi-agent system (MAS) and the RL algorithm. The most distinguishing features of this work are follows: 1) the definition of the local and global rewards which coordinate both frequency recovery and voltage regulation of an MMG, and 2) the RL-based DSOC to cooperate multiple agents for a MAS-based MMG. By pinning parts of agents to the predefined reward, all the other existing agents can be controlled through neighboring communication coupling among them. Hence, the frequency deviations are asymptotically eliminated while the voltages are regulated considering power losses in the micorgrid. Simulation results are presented and discussed to verify the effectiveness and adaptability of the proposed control.