Reinforcement Learning Enabled Microgrid Network Reconfiguration Under Disruptive Events

Network reconfiguration is an established technique for improving the performance of distribution network and microgrids. The ever-changing generation, load patterns, and increasing frequency of outages due to extreme events necessitate a more reliable, scalable, and faster reconfiguration algorithm to safeguard the grid assets and ensure smooth operations. A hybrid of value and policy based reinforcement learning (RL) algorithm is proposed to perform the network reconfiguration while considering other network constraints. The OpenAI Gym is used to build the RL-based model with an OpenDSS based environment (in the backend as a power-flow solving tool). A microgrid developed based on the IEEE 34-bus distribution test system is used in case studies with 9 switches (sectionalising and tie switches) under different loading and line contingency conditions. The proposed RL-based algorithm performs satisfactorily in terms of loss and load satisfaction compared with the baseline binary particle swarm optimization algorithm.