An Extended Q-Routing-based Event-driven Dynamic Reconfiguration of Networked Microgrids

Large-scale integration of distributed energy resources (DER) in the power distribution grid leads to the possible operation of multiple microgrids (MG). DER management system (DERMS) is responsible for the optimal network configuration, dispatch of DERs, and providing appropriate set-points for the DERs. The network of multiple MGs needs to be reconfigured dynamically when extreme events occur in the distribution grid. Reinforcement learning methods are used in various optimal energy management, and dispatch of DERs. Q-routing is a reinforcement learning-based method to discover the optimal path between source and destination nodes. Since in networked MGs, each MG can have multiple boundaries, the Q-routing algorithm is enhanced with a reward mechanism depending on the events in the distribution grid. In this paper, extended Q-routing-based dynamic reconfiguration of a networked MG is proposed including the event-driven TCP/IP communication. Real-time results validate the effectiveness of the proposed framework for various reconfiguration test cases.