Data-Driven Anomaly Detection and Mitigation for FACTS-Based Wide-Area Voltage Control System

Abstract

Wide-area voltage control system (WAVCS) ensures comprehensive voltage security and optimal management of power resources by incorporating flexible alternating current transmission system (FACTS) devices. However, due to its reliance on a wide-area communication network and coordination with FACTS-based local controllers, WAVCS is susceptible to cyberattacks. To address this issue, we propose a data-driven attack-resilient system (DARS) that integrates a machine learning-based anomaly detection system (ADS) and rules-based attack mitigation system (RAMS) to detect data integrity attacks and initiate necessary corrective actions to restore the grid operation after disturbances. The proposed ADS utilises the variational mode decomposition (VMD) technique to extract sub-signal modes from the measurement signals of WAVCS and computes statistics features to detect data integrity attacks using machine learning algorithms. Our proposed methodology is evaluated by emulating the fuzzy logic-based WAVCS, as developed by the Bonneville Power Administration (BPA), for Kundur's four machine two-area system. The WAVCS applies $V$ mag$Q$ algorithm that utilises synchrophasor measurements (voltage magnitude and reactive power) to compute an optimal set-point for FACTS devices. Experimental results show that our proposed algorithm (VMD-DT) with statistics features outperforms existing machine learning algorithms while exhibiting a smaller processing time. Also, the proposed RAMS is effective in maintaining transient voltage stability within acceptable voltage limits by triggering different modes of operations upon detection of anomalies in grid network.