Game And Control-theoretic Research On Hierarchical Security Defense In Multi-agent Industrial CPSs

With the advent of the Industry 4.0 era,the development of intelligent Chinese factories and manufactures have been specified in the plan of China Manufacture 2025.The mass,energy flow in the physical world,and information flow in the cyber world are required to be integrated and coordinated.The technology of cyber-physical systems(CPSs)is therefore developed and widely applied to various types of industrial infrastructures,e.g.,Urban water supply system,power system,natural gas transmission system,railway transportation,and brings great convenience for social productivity.Nonetheless,due to its multi-agent and hierarchical framework,the systems are much readily penetrated by advanced persistent threat(APT)attacker,and the accidents,e.g.,water pollution,power outage,gas leakage,traffic congestion,would be given rise to.The national security,people’s lives,and property are under huge threat.The current research on industrial CPSs security is limited in compliance strategy design based on the mindset of traditional computer network security problems and lacks a comprehensive understanding of the internal operating mechanism of physical plants together with the influence that cyber components impose on physical ones from a perspective of dynamic attack-defense interactions.Game theory,as a tool for analyzing multi-player behaviors under different and competitive targets,is suitable for capturing the characteristics of the interactions between attacker and defender in practical security scenarios.Besides,game theory is highly matured with rich research results.As a consequence,we study the complex coupling relationship between the cyber and physical layers,and between different agents.Based on game and control theory,we propose a hierarchical defense strategy as the fundamental and theoretic basis for practical security scenarios.The main contributions of this dissertation could be concluded as follows:1.Secure controller design in the physical layer based on incentive Stackelberg gameWe focus on how to maintain the operational normalcy of physical components and consider the situation that the physical controller is assumed to be a defender as well as a leader with an information advantage.We build an incentive Stackelberg game model for capturing the interactions between controller and disturbance imput generated by an industrial APT attacker.According to the affine incentive solution,we can mislead the optimal solution of disturbance such that the desired team-optimal solution could be obtained with physical plant devices being protected.2.Optimal cyber defense strategy and physical secure controller design based on the coupled game modelConsider the worst-case security scenario where attacker’s ability is assumed to be maximized,we formulate a repeated Stackelberg security game and a two-person zero-sum stochastic differential game in cyber and physical layer respectively.The structural state transition rate of physical dynamics together with cybersecurity utility is used for capturing and quantizing the cross-layer couplings.Through solving the coupled games,we derive a comprehensive security defense scheme for individual CPSs.3.Hierarchical security defense for encrypted industrial CPSs based on the modified coupled game modelIn the cybersecurity game,we further introduce imperfect information structure and formulate a modified security game(MSSG)with a quantitative model for a real-time performance index.Meanwhile,in physical security game for optimal control,we introduce the stochastic time delay determined by cyber interactions.Furthermore,disturbance attenuation level is regarded as a physical control performance index.Through solving the modified coupled games,we derive a hierarchical security defense scheme,the effectiveness of which is verified on testbench.4.Multi-agent cybersecurity incentive strategy design based on interdependent security gameWe formulate a Stackelberg interdependent security game for multi-agent industrial CPSs for quantizing the security negative externalities.The cross-layer interdependency is taken into accounts when formulating the overall cost function.Through solving for subgame perfect Nash equilibrium(SPNE),we derive the multi-agent optimal cyber incentive strategy and eliminates the strategy gap between individuals and a social planner.5.Multi-agent secure controller design based on multi-person nonzero-sum stochastic differential gameUnder the centralized information structure,the state equation and the cost function of each individual CPSs are assumed to be coupled through the states of all the other agents.Through solving for the solution of an N-person worst-case nonzero-sum stochastic differential game,we derive the state-feedback Nash equilibrium as a physical security control strategy such that each individual CPSs would maintain operational normalcy under both of physical state interdependencies and worst-case disturbance.At last,we conclude the thesis and discuss the possible extensions of current work.