A State Estimation Framework for Mitigating Non-Gaussian Noise and Integrating Correlation

Accurate state estimation is critical for the reliable and efficient operation of modern power systems. However, conventional techniques often assume that measurement errors are independent and normally distributed, an assumption that is frequently violated in real-world environments. To overcome these limitations, this study proposes a Gaussian mixture filter-based power system state estimation framework incorporating a novel measurement denoising model based on a temporal convolutional mixture density network. Specifically, the denoising model produces Gaussian mixture outputs, which not only effectively remove non-Gaussian noise by accounting for complex correlations within the data, but also estimate full covariance matrices that characterize the distribution of measurement errors. Furthermore, the Gaussian mixture filter-based estimator processes each component of the Gaussian mixture individually and then aggregates the results to preserve the characteristics of the measurement errors. Numerical simulations demonstrate that the proposed approach significantly improves estimation performance by effectively suppressing non-Gaussian noise and incorporating measurement correlations.