Stochastic Planning for Resilient Infrastructure of Distribution System Under Extreme Weather Events

Recent power disruptions due to extreme weather underscore the critical need to boost power system resilience. This article introduces a resilient infrastructure (RI) planning model using deep reinforcement learning (DRL). The model enhances stochastic planning by leveraging the capabilities of DRL to motivate resilience enhancement. It perceives the operation and line fault states of the distribution network during severe disasters as observation sets. The model establishes a resilience enhancement return function for a self-learning agent to evaluate RI actions. Stochastic scenarios encompassing line faults and system operations during extreme weather are generated via Monte Carlo simulation (MCS) techniques. The DRL, structured around the dueling deep Q network (DDQN), is utilized to capture the complex relationship between RI actions and resilience improvement. The IEEE 33-node distribution system serves as the testing ground for the proposed methodology. The convergence profile of the loss function demonstrates the dueling network's proficiency in approximating the intricate relationship between RI action and resilience enhancement. A comprehensive analysis of the resilient infrastructure's stochastic planning scheme is presented. The findings validate that our proposed approach effectively mitigates load losses resulting from unpredictable weather events.