Advancement in power system resilience through deep reinforcement learning: A comprehensive review

Abstract

The power system resilience is crucial for ensuring a stable and constant power supply in the midst of disruptions and challenges caused by sources and frequent catastrophic events. Reinforcement learning (RL) and its advanced form, deep reinforcement learning (DRL), have emerged as effective methods for enhancing power system resilience in recent years. It enables intelligent decision-making in dynamic and unpredictable environments, solving difficulties and challenges that the power systems face. This paper provides a comprehensive overview of RL's applications in power system resilience, such as resilience metric development, grid control and operation, fault detection and diagnosis, islanded microgrid management, the integration of resilient Distributed Energy Resources, and adaptive control for cyber-physical security. We look at the fundamental algorithm of DRL, such as policy-based, value-based, and actor-critic techniques, as well as their practical applications in dynamic response, recovery, energy management, and cyber security. Despite its potential, the integration of DRL involves problems and constraints, which are also highlighted. The main aim of this work is to provide a comprehensive understanding of RL and DRL's role in strengthening dependability and sustainability for enhancement of power system resilience while offering insights into future research directions and the potential of these technologies in addressing uncertainties and optimizing decision-making in complex environments.