New Power System Wide-Frequency Oscillation Source Localization Method Based on Informer-EGAT Multimodal Spatio-Temporal collaborative Perception

A novel method for localizing wide-frequency oscillation sources by combining Informer-based signal compression-reconstruction processes with Edge Graph Attention Networks (EGAT) to improve localization accuracy is proposed. At the substation level, the Informer network’s advanced temporal modeling and feature extraction capabilities are utilized to efficiently compress and encode the measurement signals from the power system. At the master station, the compressed signal features are integrated with the system’s network topology, and EGAT is employed to achieve accurate oscillation source localization. By adaptively adjusting the weights of the nodes based on edge feature information, EGAT captures inter-node correlations, which enhances the accuracy and robustness of the localization process. This approach proves effective in accurately identifying the locations of oscillation sources within complex power network topologies, even in the presence of noisy measurement data. Experimental results demonstrate that the proposed method maintains high accuracy and timeliness across various power system topologies, even under conditions of noise interference, highlighting its robustness and effectiveness.