The Implementation of a Power Grid Fault Contingency Drill and Verification System Based on Deep Reinforcement Learning

With the increasing share of renewable energy generation, grid stability is declining, and control faces significant uncertainty, posing challenges to the safe, reliable, and economic operation of the power grid. Traditional fault contingency planning relies on manual experience, resulting in low efficiency and poor timeliness. In response to the need for faster fault recovery in large power grids, this paper proposes a deep reinforcement learning-based dual-layer solution model. The model integrates power flow analysis and fault set generation techniques to optimize various fault scenarios. The upper layer utilizes deep reinforcement learning algorithms to determine discrete actions, such as activating backup lines and generators, while the lower layer optimizes continuous variables, such as generator output and power flow, ensuring system reliability and minimizing load loss. The proposed method is applied to a realworld grid case, demonstrating its effectiveness across different fault scenarios. Results show that the model generates efficient and reliable fault recovery strategies, ensuring stable system operation even under complex fault conditions. The method significantly enhances the speed and quality of fault recovery decision-making, providing strong support for real-time grid scheduling and management.