Reinforcement Learning for Service Restoration Algorithms in Distribution Networks

Abstract

Modern Distribution Networks (DNs) are highly susceptible to faults, which affects their dependability and reliability. The operation complexity is crucial when DNs include critical infrastructure such as distributed energy resources, storage systems, charging stations and decentralized supply. FLISR (Fault Location, Isolation and Service Restoration) relies on advanced methodologies which aim to improve the quality of service with automated algorithms. This paper proposes a novel Service Restoration approach to automatically assist DNs resupply the out-of-service unfaulted customers after an event. The approach integrates Reinforcement Learning techniques in a co-simulation environment with OpenDSS. The results and contribution of this study could improve power supply quality and reliability of DNs throughout advanced Service Restoration (SR) methodologies. The idea is validated in real-time simulation to offer a performance assessment after training with co-simulated data.