Deep reinforcement learning-based strategy for enhancing power grid resilience under earthquake disaster scenarios

Extreme events such as earthquakes can readily cause structural damage and operational disturbances in power grids, thereby weakening the system’s supply stability and recovery capability and posing substantial challenges to overall grid resilience. Conventional resilience risk assessment methods generally rely on empirical judgments and static analysis, which are insufficient to support rapid response and dynamic decision-making under disaster impacts. To overcome these limitations, this paper proposes a deep reinforcement learning (DRL)-based strategy for minimizing resilience risk in power grids under earthquake disaster scenarios. First, based on a three-level potential seismic source zoning method combined with a seismic elliptic attenuation model, ground motion parameters corresponding to various epicentral distances are derived to generate representative multiple fault scenarios. Subsequently, a probabilistic model of line failures is employed to quantify functional losses under disaster conditions, and a resilience risk quantification model is further established. Finally, a deep reinforcement learning algorithm is introduced, enabling an intelligent agent to adaptively learn a risk-minimization strategy through interactions with earthquake scenarios. The test results indicate that the proposed method reduces the overall functional loss of the power grid under earthquake scenarios by 18.1% compared with the benchmark method, thereby significantly decreasing resilience risk and enhancing the reliability of power supply.