Attack Detection And Secure State Estimation Of Cyber Physical Systems Under Attacks

With the wide spread application of the information science and technology in the modern industrial and society,the Cyber Physical System is presented to define a kind of complicated systems,such as transportation networks,high speed train systems,unmanned grounded vehicles,future power systems and smart grids,that are characterized by tight integration and cooperation between cyber and physical components,like processing units monitor,control physical processes by means of sensors and actuators networks.Essentially,CPSs can be treated as the kind of networked systems with control mechanisms.Thus it is significant to explore the CPS security with the control theory.On the one hand,it is called for a unify architecture or model to describe the attack,including its knowledge and performance;on the other hand,it is also called for secure control design methods for CPS under different kinds of adversarial attacks,including the conditions and methods of attack detection and secure state estimation.It is noted that,some results are already presented;however,there still exist many drawbacks.For instance,the existing attack detection and secure state estimation can be classified into two classes.The first is based on a finite time measurement sequences,which ensures the attack detection and secure state estimation to be obtained in a finite time,but its robustness to disturbance and noises needs to be enhanced.The second is based on Luerberger observers,which ensures the attack detection and secure state estimation to be achieved asymptotically or finite but quite a long time,thus novel attack detection and secure state estimation methods whose estimation error can converge to zero more fast or in a given finite time is needed.Utilizing the linear and nonlinear theory and observer design methods,this thesis addresses the attack detection,attack reconstruction and secure state estimation problems for the CPS under attacks,including the model of CPS,the model of attacks,the detectability of attacks,the identifiability of attacks,the solvability of the secure state estimation,and some methods to solve these problems for different kinds CPS under attacks.The methodologies and the contributions are:1 The model of CPS under attack is presented and established,which includes the ability,performance and preferability of the attacks,with that,some existed attacks are also explored,such as their strageties,working modes and damage to the control systems.The fouus of the thesis is also formulated as two problems,the attack detection problem and the secure state estimation problem.2 For a class of linear time invariant CPS under s-sparse sensor attacks,the solvability of the attack detection and secure state estimation problems are presented and established with the r-resilient observability of the CPS.Then a finite time observer based method is presented.It is shown that,with the proposed finite time observer as well as an estimation error based detector,the attack detection can be ensured in a given finite time.Then with the fintie time observer,an attack detecton alarm based secure subset searching stragety is proposed,which can solve the secure state estimation problem in a given finite time.3 Then for a class of linear CPS under state attacks,a sufficient condition that the attack can be reconstructed and the secure state estimation can be obtained is presented and established,and a sliding mode observer(SMO)with an online parameter approximation is proposed.It is shown that,with the adaptive SMO as well as a robust differentiator,the secure state estimation and the attack reconstructed can be ensured,and with the reconstructed signal,attack detection can be also obtained.4 Then for a class of nonlinear uncertain interconnected CPS under both state and sensor attacks,a sufficient condition that the attack can be reconstructed and the secure state estimation can be obtained is presented and established,and distributed robust observers with neural network based approximation of the uncertainties are also presented.It is shown that,with the proposed schemes,the state and attack estimation can approach the state and attack simultaneously,and a distributed scheme is also presented which ensures that the robust attack detection can be achieved.