Event-Driven Q-Routing-Based Dynamic Optimal Reconfiguration of the Connected Microgrids in the Power Distribution System

The operation of connected microgrids enables the availability of local distributed energy resources (DER) at the network level. During grid events such as electric faults, part of the power distribution network needs to be isolated such that the healthy part of the network can operate generally through optimal reconfiguration. Most of the topology agnostic reconfiguration schemes are complicated, computationally expensive, and offer a single optimal path. To overcome these disadvantages, a reinforcement learning-based extended q-routing method is proposed in this article to achieve optimal network reconfiguration. The proposed method utilizes a model-free adaptive learning technique, thus efficiently discovering optimal paths in a dynamically changing network. To validate the proposed method in a real-time environment, a detailed dynamic distribution network model is developed, including primary and secondary control of integrated DERs and the protection functions. Furthermore, event-driven communication is designed to exchange data between the dynamic network model and the reconfiguration agent. The results obtained from the real-time agent-in-the-loop set-up showcase the effectiveness of the proposed method that achieves network reconfiguration within 1.5 s.