Signal Recovery in Power Systems by Correlated Gaussian Processes

Abstract

This article proposes the application of correlated Gaussian processes (Corr-GPs) for the recovery of missing intervals in power systems signals. Based on only local power system topology, the presented algorithm combines cross-channel information of the considered signals with a universal, nonparametric probabilistic machine learning regression to recover missing data. Starting from the theoretical background, the proposed approach is presented and contextualized in the framework of signal recovery for power systems. Then, by making use of real data collected from the Living Lab Energy Campus—a real-life laboratory established at Forschungszentrum Jülich—we demonstrate the use of the proposed approach for recovering distribution grid signals. We evaluate the performances of Corr-GP compared with those of other state-of-the-art techniques. In addition to outperformance in terms of recovery accuracy, it is explained when and how the accuracy of the reconstructed signal is independent of the missing interval length. Finally, detailed insights about key characteristics of the proposed approach that generate practical benefits for system operators are provided. A self-aware failing indication allowing system operators a direct evaluation of the recovered data and enabling further improvement of the proposed approach is presented, as well as recommendations for field implementation.