Optimal False Data Injection Attack Design In Cyber-Physical Systems

Cyber-physical systems connect information systems and physical systems together through various heterogeneous networks.Cyber-physical systems realize the real-time and accurate perception and control of the physical environment,and realize resource sharing and information exchange between individuals or between men and objects.The deep integration of physical systems and information systems brings a lot of potential security threats.There exist security problems due to the openness of interconnected networks and the weakness of the information systems.Traditional methods of information security do not consider the physical process and can not predict the consequences of physical systems when cyber attacks occur.They can not provide defense strategies for different types of attacks from the perspective of control system design.Therefore,this doctoral dissertation studies the security of CPSs from the perspective of control theory,including the data injection attack design and the resilient control.Although many scholars have made extensive and in-depth studies on these problems and made great progress,there are still many deficiencies.For instance,the modelling of the false data injection attack mainly focuses on the case where the attack location is not changed and rarely consider dynamic attacks that frequently change attack locations.A few scholars have considered the location switching and only near optimal strategies can be given through search algorithms.The analytic expression of the optimal switching strategy is hard to be given.In view of these shortcomings of existing methods,this doctoral dissertation investigates the data injection attack design to understand the intention of the attacker,explore new attack scenarios and discuss the corresponding countermeasures.The specific research contents in this dissertation are as follows:1.For the optimal actuator attack problem,the false data injection is viewed as an additive external input of the healthy system.The objective of the attacker is to damage the plant as much as possible while maintaining a low attacking energy.Sufficient conditions for the existence of an optimal solution,a steady-state solution and a special solution are investigated.The stability of the attacked system is analyzed.2.For the optimal switching actuator attack problem,the attacker is capable of switching the attack signal between multiple actuator sets.An analytic expression of the optimal solution is given,including a closed-form optimal switching condition determining the best actuator set and a switching dependent feedback attack signal.The optimal switching data injection attack problem is transformed into a traditional optimal control problem by embedded transformation.The optimal solution of the embedded problem is proven to be of bang-bang type.3.The optimal sensor switching attack problem which maximizes the energy of the compromised control signal is studied.An optimal switching condition is derived when the attack matrix is a vector.The switching attack is modeled as a switching uncertainty of the healthy system.The resilient control law is proposed to stabilize the attacked system under arbitrary switching policies.4.The norm bounded switching attack problem is proposed.Two optimal switching conditions are given for two norm bounded switching attacks.If the amplitude of the attack signal is bounded,the optimal attack signal is of bang-bang type.If the energy of the attack signal is bounded,the optimal state feedback signal is corresponding to the upper bound of the attacking energy.5.The optimal partial feedback switching attack problem is studied.The attacker is assumed to be capable of getting access to partial state variables to construct the attack vector.A diagonal partial feedback matrix is used to obtain the analytic expression of the optimal parameters and determine the compromised state set simultaneously.In the end,this thesis is concluded and some future research directions are discussed.