Physics-Informed Evolutionary Strategy Based Control for Mitigating Delayed Voltage Recovery

In this work we propose a novel data-driven, real-time power system voltage control method based on the physics-informed guided meta evolutionary strategy (ES). The main objective is to quickly provide an adaptive mitigate fault-induced delayed recovery (FIDVR) problem. Reinforcement learning methods have been developed for same or similar challenging problems, but they suffer from training inefficiency and lack of robustness “corner unseen” scenarios. On other hand, extensive physical knowledge has in systems little leveraged learning-based approaches. To address these challenges, introduce trainable action mask technique flexibly embedding into RL models rule out unnecessary unfavorable actions, achieve notable improvements sample efficiency, performance robustness. Furthermore, our leverages past experience derive surrogate gradient guide accelerate exploration process training. Case studies IEEE 300-bus comparisons with state-of-the-art benchmark demonstrate effectiveness advantages method.