Research on Distribution Network Fault Self-Healing and Dynamic Scheduling Based on Deep Reinforcement Learning

The rapid development of new power systems has introduced significant challenges in distribution network fault self-healing and dynamic scheduling, including complex topological structures and high penetration of distributed energy resources. this study proposes a collaborative control framework based on deep reinforcement learning (drl), integrating a markov decision process (mdp) model that accounts for multi-timescale coupling. a hybrid algorithm combining graph convolutional networks (gcn) and double deep q-networks (ddqn) is designed. simulation results on the ieee 33-node system demonstrate that the proposed method reduces fault recovery time by 18.7%, improves power supply reliability to 99.992%, and lowers dynamic scheduling costs by 23.4%.