Consensus Filtering And Control In Networks Under False Data Injection Attacks

Multi-agent systems are composed of a large number of nodes,which have the abilities of communication,computing,sensing,etc,and have broad prospects of the applications in both military and civilian fields.Wireless sensor networks can be seen as one type of multi-agent systems.The requirements for the network security are claimed due to the wide range of applications of wireless sensor networks and multi-agent systems.This paper takes wireless sensor networks and multi-agent systems under cyber-attacks as the research objects,and considers false data injection(FDI)attacks in communication channels,sensors,and actuators,respectively.This thesis focuses on secure estimation and control problem of multi-agent systems based on the methods as graph theory,stochastic control theory,Kalman filter algorithm,attack detection algorithm,consensus and extended state observer.The specific research content is as follows:A consensus Kalman filter algorithm based on the leader-follower structure and the weighted average strategy of the sensor networks is adopted.By introducing a virtual estimation error and a confidence function,the weights can be fully allocated and adaptively designed according to the proportion of the sensor's confidence.It is proved that for a time-invariant network,the mean square estimation errors of all sensors are bounded if and only if the process nodes in the extended topology are globally reachable.A consensus Kalman filter algorithm with distributed attack detection is proposed to reduce the impact of attacks in wireless sensor networks when false data injection attacks are injected into the system.Firstly,consider that when FDI attacks are randomly injected into the communication channel with a certain probability,an attack detector is designed.The data received from its neighbors,which is judged to be attacked,is isolated in the consensus Kalman filter.It is proved that under this consensus Kalman filter algorithm based on adaptive weighting protocol and attack detection,the estimation error of the sensor networks will be bounded in probability.Secondly,consider that when FDI attacks are randomly injected into the sensors,an attack detector is designed.If it is judged that the measurement data of sensor is attacked,the measurement value is not used.An upper bound of the allowable attack probabilities is provided such that as long as the attack probabilities are less than this bound,the estimation errors of all sensors are bounded in mean square sense.A combined control structure with Kalman filter and extended state observer is proposed for a multi-agent system with false data injection attacks.The false data injection attack is regarded as an unknown disturbance in the multi-agent system and the extended state observer is introduced to add the feed-forward compensation component into the controller to estimate the unmeasured state and network attack.The Kalman filter is used as a pre-filtering stage to deal with measurement noise to prepare the necessary signals for the observer and controller.An attack detector based on residuals and thresholds is designed.If it is judged that the agent is under attack,the observer-based multi-agent consensus control strategy is used.If it is judged that the agent is not attacked,the consensus based on Kalman filter is used.Finally,a numerical simulation is carried out to prove the performance of the proposed algorithm.