Characterization of Optimal Cyber Attacks on Control Systems

Securing control systems which manage the nation's critical physical infrastructures and key resources is becoming one of the highest priorities and can no longer be ignored. Several researches and incidents have confirmed the real vulnerability of these systems. Due to the vital role of control systems, the study and the development of advanced strategies to overcome these cyber-security breaches are very important. Control systems are mainly composed of sensors and actuators, which cyber-security is vital for the security of the whole control system.

Characterizing and guarding against worst case attack scenarios is important to secure control systems. This dissertation proposes to study worst case optimal attacks for various estimation and control algorithms. In particular, optimal attack signals on the ever popular least squares and weighted least squares algorithms are explicitly computed. Optimal strategies for bounded and finite energy sensor and actuator attack signals are characterized using functional analytic tools. Moreover, optimal random signal attacks on the Kalman filter are characterized. In particular, a worst case probability distribution is provided.

A new optimal attack proof estimator is derived. This estimator mitigates optimal attack strategies on the Kalman filter. Illustrative numerical studies are provided to support our methods for analyzing the worst case attack signals, including optimal attack' scenarios on a power grid network, and a radar tracking system, showing the validity of the results obtained and also the effectiveness with real applications.