A Unified System Theoretic Framework for Cyber Attack-Resilient Power Grid

Abstract:

We developed a systematic analytical and computational framework for the vulnerability analysis and mitigation of data integrity attacks on Phasor Measurement Units (PMUs) employed for wide area monitoring and control of power system. The analytical framework is based on the stability theory of stochastic dynamical system and it allows one to systematically determine the PMUs most critical to power network security. For the mitigation of the attack, a convex optimization-based formulation is proposed for the design of stabilizing feedback controller robust to data integrity attacks on PMUs. Furthermore, unified framework based on stochastic theory of dynamical systems and control theory is proposed for modeling, analysis, and mitigation of correlated attacks on network power system. The developed analytical framework is used to determine an impact of correlated attack on stability margin of power system. For problem involving anomaly detection, we describe an online model-based approach to detect cyberattacks in power system state estimation. The proposed approach leverages information that is independent of traditional SCADA measurements, such as load forecasts, generation schedule information, and existing synchrophasor data to detect measurement anomalies in State Estimators through statistical characterization. We also describe a game-theoretic framework for Cyber-Physical Risk modeling and mitigation that captures all three aspects of risk, namely, threats, vulnerabilities and impacts appropriately in the game formulation. We show a simple and intuitive case study to illustrate the importance of game theory in capturing attacker behavior, which is otherwise neglected in traditional risk assessment approaches. An attack scenario where load buses are attacked to cause voltage instability in the system to reduce its performance. Fault Induced Delayed Voltage Recovery (FIDVR) is a phenomenon that can reduce the performance of the component and may induce system wide instability. A Lyapunov exponent-based method that detect and quantifies this FIDVR phenomenon in real-time from timeseries data is developed.