Data-Driven Security Detection For Industrial Cyber-Physical Systems Under Integrity And Availability Attacks

Cyber-physical system(CPS)is an important development direction of information technology.Industrial cyber-physical systems(ICPS)are important components of CPSs,which are related to the national strategic layout and the economic development.The security issues are inevitable barriers in the development of ICPS.Security detection,also known as anomaly detection,is one of the important research topics in the security protection mechanisms.The existing research results are effective in detection methods for DoS,FDI,replay,Stuxnet-like,and other attacks against ICPS.However,existing methods generally assume that the model information of the controlled physical plant is known completely and design detectors only for specific types of attacks.There are no effective security detection methods for unknown or multi-class attack forms.The model information of the physical plant is usually not known perfectly in practice,and the purposes of attacks on ICPS are focused on compromising both integrity and availability(IA)of data compared to traditional cyber attacks.Therefore,it is of great theoretical and practical importance to study security detection methods that are capable of dealing with IA attacks.This study investigates the data-driven security detection methods for ICPS under the IA attacks.The main contributions are as follows.(1)An IA attack model with multiplicative matrices is developed and the essential properties of the attack matrices are analyzed.Two sufficient conditions for the existence of IA attack parameters satisfying the essential stealthiness are given to verify the practicality of the proposed model.The theoretical results are evaluated by using the quadruple tank process(QTP)for a simulation example.The results show that the IA model is capable of representing multi-class attacks.The stealthy attack sequences based on the model can be designed offline,reducing the costs and exposure risks of the attacks.(2)On the basis of the proposed IA model,a detection framework is proposed that identifies the security of ICPS by quantifying the differences in generalized models.Two detection variables,data-driven innovation estimation and extended Markov matrix,are given with the convergence analyses.Hardware-in-the-loop(HIL)experiments are conducted using servo systems as the controlled plants.The experimental results verify that the proposed detectors have significant performance for four types of IA attacks including stealthy FDI attacks.(3)A method of persistent excitation processing for operating data is proposed to eliminate the performance loss of ICPS caused by the external disturbances introduced by detectors.Then,online calculation of the amplitude frequency characteristic(AFC)matrix by using the operating data of the control layer solely is investigated to avoid the deployment of additional hardware.The convergence of the detection variables constructed by the AFC matrix is proved.Next,the definitions of degree of freedom and sensitivity of detectors are proposed for the unified performance assessment of multiple types of detectors.Finally,an ICPS attack and defense experimental platform and a semi-physical simulation method are proposed.An HIL experimental environment is built with the QTP to verify the detection performance of the proposed detection method for Stuxnet-like,essential FDI,and DoS IA attacks under a tracking control application scenario.(4)An oblique projection method based on a perturbation period matrix is proposed to eliminate the influence of periodic nonlinear perturbations and noise on the computation of the security detection variables.The consistency of the parameter estimation is analyzed.Then,the construction method of the detection variable is studied with the convergence analysis under a special periodic nonlinear form in which the expectations of the perturbation variables are unknown constants.Finally,the QTP platform is adopted to verify and analyze the effectiveness of the proposed detection strategies via some HIL experiments.The existing results do not focus on the problem of security detection for nonlinear ICPS subject to stealthy attacks,so the proposed strategies have obvious advantages in terms of application scenarios.Lastly,the dissertation is concluded and the further research work is prospected.