A Gaussian Process Regression Framework for Enhancing Convergence in Power System State Estimation

This research introduces a novel hybrid state estimator that seamlessly integrates Gaussian Process Regression (GPR) (Kriging) with the traditional Weighted Least Squares (WLS) method. The core innovation lies in using GPR after each WLS iteration to generate probabilistically informed virtual measurements that leverage the spatial correlations inherent in the power network's topology. This approach acts as an intelligent spatial regulariser, providing a superior initial state for subsequent iterations. Benchmarking studies on test systems with up to 190 buses demonstrated a 30–50% reduction in iteration count and 46% faster computation time for large networks. The framework also improved estimation accuracy, particularly for voltage phase angles, and demonstrated inherent robustness to measurement noise and data scarcity, establishing a scalable and resilient paradigm for modern grid management.