Power System Operation State Identification Based on Particle Swarm Optimization and Convolutional Neural Network

With the proposal of the "dual-carbon" goal, there has become an important trend of the high penetration access of the renewable energy and the wide applications of the power electronic equipment. However, their intermittences and uncertainties are posing severe challenges to the real-time operation state identification of the power system. A novel method of the real-time operation state identification for a electric power system based on the particle swarm optimization and the convolutional neural network (PSO-CNN) is proposed．Firstly, considering the transient problems under both the safety domain and the stability domain of the power system, this method is suitable for the identification of a power system operation status in multiple scenarios before, during and after a transient stable fault. Secondly, in order to ensure the comprehensiveness of the output modes of the new energy unit in the sample data, the Latin supercube sampling is used to select the refined simulation data. Regarding of the problem of extreme imbalance in the sample classification in the actual data and simulation data of the power system, the PSO algorithm is used to adjust the loss function weights for different status classifications to improve the state identification effect. Finally, taking the IEEE39 system and an actual provincial power grid system as examples, the effectiveness and robustness of the proposed state identification method is verified. © 2024 Power System Technology Press. All rights reserved.