Robust Recovery of PMU Signals with Outlier Characterization and Stochastic Subspace Selection

This paper proposes an improvement on the standalone robust principal component analysis (R-PCA)-based approach for recovering clean signals from corrupted synchrophasor measurements. The contributions of this paper are twofold. First, a kernel principal component analysis (K-PCA)-based metric is proposed for detecting and differentiating event-induced outliers from spurious outliers in data, which is then used as an indicator to suspend R-PCA in the event window to minimize the overall error in signal recovery. Second, a formal approach based on the recursive Bayesian framework is proposed for selecting the most appropriate subspace from a library of subspaces to be used by R-PCA. The paper combines the ideas of robust signal recovery, corruption-resilient event outlier detection, and stochastic subspace selection into a composite approach for correcting anomalies in synchrophasor data. The effectiveness of the proposed methodology is validated on simulated data from IEEE 16-machine, 5-area test system.