An Adaptive Penalized Weighted Least Squared Approach for Detecting and Mitigating Cyberattacks on Dynamic State Estimation

Leveraging measurements from phasor measurement units (PMUs) within wide-area measurement systems (WAMSs) has empowered dynamic state estimation (DSE) to assume a crucial role in power system control and real-time contingency analysis. Yet, the susceptibility of WAMSs and PMUs to exploitation by attackers poses a significant threat. This vulnerability is underscored by the potential impact of false data injection attacks (FDIAs), where malicious data infiltrate DSE, compromising its accuracy and reliability. To address this concern, this article introduces an approach to detect and mitigate cyberattacks on DSE in power systems. The proposed method enhances the extended Kalman filter (EKF), commonly used for DSE, by incorporating a machine learning-based penalized weighted least squared (MLPWLS) approach. This augmentation ensures the accurate assignment of weights to measurements through an optimization process. Higher weight indicates greater trust in measurements, while lower weight signals vulnerability to cyberattacks. Consequently, measurements with lower weights are excluded to mitigate the impacts of cyberattacks. In fact, incorporating machine learning into the classic penalized weighted least square (CPWLS) method enhances its adaptability across a spectrum of operating conditions and in the face of diverse FDIAs. The performance of the proposed approach is evaluated on the IEEE 14-Bus and IEEE 39-Bus test systems under various attack scenarios.