Emergency Control of Power Grid under Topology Changes Based on Graph Reinforcement Learning

Abstract

The current power grid topology changes frequently, and power angle instability and voltage instability often occur at the same time. Emergency control methods based on physical characteristics are usually difficult to model in complex power systems, and have poor adaptability to power system structural changes. The traditional analysis methods based on the physical characteristics of the power grid can no longer meet the requirements of power grid control, and new methods need to be found to solve this problem. In this paper, an emergency control method based on graph reinforcement learning (GRL) is proposed for the emergency control problem of AC/DC hybrid grid. The graph neural network is used to extract the environmental features, and the state space, action space and reward function of reinforcement learning are designed for the emergency control task. Combined with the transient stability analysis, the emergency control strategy framework is constructed, and the effective emergency control decision after failure is realized. The effectiveness of the method in the topology change scenario is verified.