A Self-Supervised Training Method for Power System State Estimation Based on Dynamic Data Masking

This paper introduces a self-supervised training method for power system state estimation, utilizing dynamic data masking based on node importance indicators. The approach dynamically adjusts the focus of power flow calculations by determining the importance of each node in the system, considering factors such as load, generation capacity, and equipment count. These importance indicators are used to generate dynamic masks, allowing for adaptive selection of critical nodes during power system operation. The method employs selfsupervised learning to optimize the power flow estimation model by minimizing the prediction error between simulated and real power flow data. A case study on the IEEE 118-bus system demonstrates the effectiveness of the proposed method in improving the accuracy and efficiency of power system state estimation, providing substantial computational savings without sacrificing precision. The results show that the method significantly enhances power flow analysis, making it a promising approach for large-scale power systems where labeled data is scarce.