Research On The Filtering Problem Of Networked Systems Agasint Multiple Cyber Attacks

Networked control systems(NCSs)are a kind of spatial distributed systems that combine the controlled object,sensors,controllers and actuators organically by using communication network.Due to the introduction of communication networks,NCSs have the advantages of strong flexibility,high reliability,easy expansion,etc.,and are increasingly widely used in national life(such as water supply,power supply,transportation,medical treatment,etc.).However,while NCSs bring convenience to people's lives,they also bring security issues.In the open network environment,NCSs are vulnerable to cyber attacks,which poses a serious threat to industrial production.Therefore,the research on the security of NCSs is of great practical significance.At present,the security problem of NCSs has made some progress,but there are still some deficiencies.Specifically,first of all,when the system is attacked by cyber attacks,the corresponding attack compensation strategy is missing to mitigating the impact of attacks on the system performance.Second,most of the existing research results consider a single type of cyber attacks,which is not realistic,while there are few research results about multiple cyber attacks acting simultaneously on systems; Third,most of the literature uses random variables with known probabilities to model attacks,but this method of modeling attacks is ineffective when the probability of attacks is unknown; Finally,when the system is in a complex network environment,such as random cyber attacks,sensor failures,limited communication resources,etc.,which bring certain difficulties to the security analysis of the system.Therefore,the relevant research is not very sufficient; Based on the above shortcomings,this thesis focuses on the security filtering problem of networked systems under various attacks,and the specific research contents are as follows:(1)The security filtering of discrete uncertain networked systems under replay attacks is studied.First,in order to cope with limited communication resources,event triggering mechanism(ETM)is used to improve the effective utilization of resources.Secondly,a Bernoulli random variable is introduced to describe the replay attacks that occur randomly between sensors and filters,and an attack compensation term is constructed to reduce the impact of the attacks on the filtering performance.Then,considering the effects of uncertainty,random disturbance and replay attacks on the systems,the performance index of filter security is presented.Then,by using the random analysis method and linear matrix inequality technique,two sufficient conditions for the filtering error systems to satisfy the proposed security performance are given.(2)The security state estimation of discrete time-varying networked systems under deception attacks is investigated.First,a dynamic ETM with dynamically adjustable trigger threshold is introduced to reduce the network transmission burden.In addition,bounded but unknown variables are used to model deception attacks that occur randomly in the sensorto-estimator channel in the network.Secondly,considering the effects of deception attacks and gain perturbations on systems performance,an ETM-based non-fragile state estimator is designed.Finally,based on Lyapunov stability theory,two sufficient conditions for an augmented system to achieve a given performance are derived.(3)The secure set-membership filtering problem of networked systems under mixed attacks and sensor failure is studied.First,two sets of stochastic variables that obeying to Bernoulli distribution are adopted to model the random deception attacks and denial of service attacks.To further improve the effective utilization of network communication resources,an ETM is used.In addition,considering the impact of bounded but unknown noise,sensor failures,and mixed attacks on system performance,a security set-membership filter based on ETM is designed.Then,with the help of Lyapunov stability theory and random analysis technique,sufficient conditions for the real state of the systems to converge within a given set of estimated ellipsoids are given.