| As a new generation of intelligent systems that promote the integration of information technology and industrialisation,cyber physical systems(CPSs)have been widely used in key areas such as chemical production,transport management,healthcare,and intelligent manufacturing,making a positive contribution to the improvement of human quality of life.However,in practice,the data collection and communication of CPSs are usually integrated and interacted through cyberspace,and this open network environment brings new security threats to CPSs.False data injection(FDI)attacks are precisely one highly stealthy and common form of spoofing attack,where malicious intruders exploit system vulnerabilities to compromise data integrity.In addition,existing attack and defence research has focused on the linear domain of CPSs.Considering that every control system has a degree of non-linearity,security research in the non-linear domain is imperative.Therefore,an in-depth study of false data injection attacks in the context of non-linear information-physical systems is of great significance and application value.Taking security control as an entry point and from the attacker’s perspective,this paper first proposes a new spoofing attack strategy based on the extended Kalman filtering(EKF)algorithm,which is to launch a spoofing attack by modifying the new message sequence intercepted during data transmission in a single sensor network.It is not only effective in reducing system performance but also in avoiding detection of false data detectors.The estimated error covariance at the remote end of the attack is derived in conjunction with the extended Kalman filter,and the degradation of system performance is quantified by the evolution of the error covariance.A linear spoofing attack is also performed on a practical problem of car tracking under single-sensor conditions,and the experimental results show that the attack strategy can successfully bypass the false data detector and successfully degrade the system performance.Due to the problems of one-sided data collection and poor fault tolerance of single sensor,this paper introduces the idea of multi-sensor fusion to reconstruct the system model in multiple dimensions and proposes a distributed dynamic weighted data fusion estimation algorithm to improve the estimation accuracy by reasonably distributing the weights of observed data among sensors.The experimental results along the linear spoofing attack on the multi-sensor vehicle tracking system show that the estimation results using multi-sensor information fusion are better than the local estimation results when the system is operating normally;while when the system is attacked by FDI,the fusion results obviously deviate from the actual position,which destroys the reliability and stability of the system,indicating that even after fusion,the abnormal data still have the ability to destroy the system.Considering the potential threat of FDI attacks,this paper investigates a detection scheme and compensation mechanism for FDI attacks in non-linear CPSs from a defender’s perspective.The EKF algorithm can filter out the surrounding noise while obtaining the new information sequence,and the state of the system can be determined based on the established recognition judgement rule and the threshold value selected by training.A comparison between the design scheme and the cardinality detection scheme shows that the design scheme outperforms the cardinality detection scheme,with a detection accuracy of over 93%.Subsequently,the abnormal data is corrected by a compensation mechanism to reduce the performance loss caused by FDI attacks to a certain extent.In summary,this paper focuses on two perspectives: attacker and defender,which can provide effective defense countermeasures for later stages by analyzing the defense vulnerabilities of CPSs and identifying one of the characteristic forms of FDI attacks to reduce system losses,adding to the security research of non-linear CPSs with certain practical significance. |
