Security Constrained Unit Commitment Optimization Based on Graph Reinforcement Learning

With the construction of the new generation power system, the centralized and massive distributed new energy with high penetration rate grid-connected, AC and DC hybrid power grids have been formed, and the scale of the future "physically distributed, logically unified" optimization model for the large power grid unit commitment has shown explosive growth compared with that of the traditional hierarchical management of the regional power grids, and the existing hybrid integer planning algorithms based on the branch boundaries suffer from the problems of "variable scale limitation and low computational efficiency", which are difficult to meet the requirements of the large grid unit commitment of multi-timescale and high-efficiency optimization computation. In this paper, we propose a method of graph reinforcement learning optimization for security constrained unit commitment, which combines the advantages of graph neural networks and reinforcement learning to deal with decision-making problems in graph-structured data.