A Distributed Load Restoration Method in Reconfigurable Distribution Networks Based on Multi-Agent Reinforcement Learning

In the context of extreme natural disasters, the reliance on centralized control and comprehensive communication can hinder the recovery capabilities of distribution networks when infrastructure is compromised. Therefore, enhancing the power restoration speed and resilience through distributed load recovery methods is critical for ensuring the continuous supply of electricity. This paper constructs a novel hierarchical multi-agent framework and proposes an entropy-based maximum multi-agent evolutionary strategy (CD-MEMARL) algorithm for distributed load recovery in reconfigurable distribution network systems. Finally, the proposed method is tested on the IEEE 33-bus system, demonstrating its ability to learn better strategies in a shorter training time when facing random subsequent events, achieving effective robust load recovery while efficiently utilizing controllable distributed energy resources.