A Deep Expected Q Network Based Regional Power System Resilient Model Under Highly Uncertain Extreme Scenarios

This paper introduces a resilience model for regional power grids based on Deep Expected Q Networks. The study focuses on the charging and discharging decisions using mobile energy storage (MES) as the primary load support component. Initially, we establish a resilience model for the power grid to withstand load reductions, highlighting the impact of spatiotemporal transfer characteristics of MES on the entire system. Next, we establish a Markov decision process for the MES system as an agent, integrating optimization model constraints into the mathematical derivation of machine learning preprocessing. Finally, we propose a resilient dispatching process based on the Deep Q Network algorithm. Simulation results demonstrate that the proposed algorithm and model enhance system resilience, ensure computational efficiency, and effectively address uncertainty.