Game-theoretic methods for security and resilience in cyber-physical systems

Modern critical infrastructures are highly integrated systems composed of many complex interactions between different system modules or agents including cyber and physical components as well as human factors. Their growing complexity demands novel design techniques for scalable and efficient control and computations for providing system security and resilience. This dissertation develops new game-theoretic frameworks for addressing security and resilience problems residing at multiple layers of the cyber-physical systems including robust and resilient control, secure network routing and management of information security and smart grid energy systems. Hybrid distributed reinforcement learning algorithms are developed as practical modeling tools for defense systems with different levels of rationality and intelligence at different times. The learning algorithms enable online computations of defense strategies, such as routing decisions and configuration policies, for nonzero-sum security games with incomplete information. In addition, games-in-games frameworks are proposed for system-wide modeling of complex hierarchical systems, where games played at different levels interact through their outcomes, action spaces, and costs. This concept is applied to robust and resilient control of power systems in which a zero-sum differential game for physical robust control design is nested in and coupled with a zero-sum stochastic game for security policy design. At the networking layer of the system, multi-hop secure routing games also exhibit the games-in-games structure, and their equilibrium solutions are characterized by backward induction solving a sequence of nested games. This approach leads to a distributed secure routing protocol that enables the resilience of network routing and self-recovery mechanisms in face of adversarial attacks. Finally, in order to address emerging energy management issues of the smart grid, we establish a fundamental game-theoretic framework for analyzing system equilibrium under distributed generations, renewable energy sources and active participation of utility users. Furthermore, we develop a novel game framework and its equilibrium solution, named mirror Stackelberg equilibrium, for modeling the demand response management in the smart grid. This approach enables quantitative study of the value of demand response brought by emerging smart grid technologies as compared to the current supply-side economic dispatch model. It facilitates fundamental understanding of pricing, energy policies and infrastructural investment decision in future highly interconnected and interdependent energy systems. Examples from power systems, cognitive radio communication networks, and the smart grid are used as driving examples for illustrating new solution concepts, distributed algorithms and analytical techniques presented in this dissertation.