Robust Power State Estimation in Internet of Things Grids via Unsupervised Clustering and Data-Driven Weighting

This paper investigates power system state estimation utilising remote terminal units equipped with wireless IoT modules in lieu of legacy wired systems. However, this wireless sensing paradigm renders measurements highly susceptible to bad data caused by industrial impulsive noise. Traditionally, using weighted least squares state estimator (WLS) in such situations leads to degradation in performance. While several robust weighted least squares variants have been suggested in the literature—most notably the M-estimator-based iterative reweighted least squares (IRWLS) algorithm—they require considerable computational load. To address this, we propose an efficient two-stage framework. In the first stage, a Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and with Gaussian mixture models (GMMs) are trained and deployed locally at the edge-sensor node to enable maximum a posteriori (MAP) detection and filtering of bad data. In the second stage, the central fusion centre collects the filtered measurements to construct the weight matrix required for the final WLS estimation. The proposed distributed algorithm, denoted as DGM-WLS, has been extensively simulated in the IEEE 30-bus test system demonstrating superior performance and resilience under high impulsive noise setting. On the local level, the DGM-WLS detector outperforms the standard Wald test while requiring the same linear computational complexity. On the global level, its estimation accuracy is 11% and 39% better than the IRWLS and WLS algorithms respectively. Moreover, it maintains a constant average convergence time of 5 iterations across all tested noise rates and is computationally 15 times faster than the IRWLS algorithm.