Towards practical and cybersecurity-aware Dynamic State Estimation in power systems

Abstract

Dynamic State Estimation (DSE), empowered by the growing deployment of Phasor Measurement Units (PMUs) within Wide-Area Measurement Systems (WAMS), has become essential for power system monitoring, control, and real-time contingency analysis. However, the effectiveness of DSE critically depends on accurate system data, which might be impacted by both cyber-attacks and real-time network variations. This thesis addresses these challenges through three interconnected research contributions. First, recognizing the vulnerability of WAMS and PMUs to False Data Injection Attacks (FDIAs), a cybersecurity-aware DSE is developed to detect and mitigate such cyber threats. This method augments the Extended Kalman Filter (EKF) used in DSE with a Machine-Learning-based Penalized Weighted Least Squares (MLPWLS) approach, optimizing measurement weights based on their vulnerability. By excluding compromised measurements, the proposed solution significantly enhances the resilience and accuracy of DSE under various cyber-attack scenarios, as validated on IEEE 14-bus and 39-bus test systems. Second, addressing the practical limitations of traditional DSE methods, this research introduces an adaptive Y-bus matrix estimation approach. A Recursive Least Squares (RLS) estimator with a Gauss-Newton Variable Forgetting Factor (GN-VFF) is developed, eliminating the need for continuous Y-bus input. By leveraging inverse power flow equations, this estimator dynamically adjusts to system changes, thus enhancing the practicality, responsiveness, and overall accuracy of DSE implementations. Finally, building upon these advancements, this dissertation presents a comprehensive real-time framework for complete Y-bus estimation using limited PMU data. This innovative two-stage architecture initially estimates a reduced Y-bus matrix through a recursive algorithm integrating a Vectorized Variable Forgetting Factor (VVFF) within the RLS framework. Subsequently, the complete Y-bus matrix is reconstructed via an overdetermined least-squares optimization based on Kron-reduction. Integrated within a simultaneous dynamic and algebraic state estimation (SDASE) framework, this approach demonstrates significant accuracy, convergence, and scalability improvements, bridging static modeling assumptions and dynamic operational realities. Together, these contributions provide a practical, secure, and dynamically adaptive framework for state estimation in a multi-machine power systems, significantly advancing real-time grid intelligence and reliability under both cyber threats and evolving operating conditions.