Correlation Reconstruction-Based Anomaly Detection in Transient Responses of Distributed Energy Resources

We propose to develop an innovative correlation reconstruction error-based, real-time anomaly detection (CREAD) scheme by leveraging correlations among the time-series physical measurements in a dynamic system. The rationale behind this approach is that the spatio-temporal correlations among measurements of various signals from a physical system are determined by the physics that governs the evolution of the system states and thus are inherent feature of the system. Therefore, any false data injection into a subset of or all the physical measurements will cause correlations or change of correlations to deviate from the ones determined by the governing physics of the dynamic system. Based on this rationale, we develop a machine learning (ML)-based approach to detect possible false data injections on some or all measurements via reconstruction of the correlation between time series. Taking power grid as an example, we demonstrate that the proposed approach can achieve high anomaly detection performance for the attacked transient responses of inverter-based resources (IBRs) for disturbances in the power grid.