False Data Injection Attack Analysis And Mitigation In Power Systems:Cyber-physical Approach

With the development of "Smart Grid" and "Energy Internet",automatic level of power systems have improved greatly.For example,to improve the capacities of situation awareness,control instantaneity and flexibility of power systems,much more sensors are deployed and networks are widely connected in current power systems.However,the tight coupling,between information systems and physical systems,introduces serious information security threaten in power systems.In this thesis,we focus on false data injection attacks in power system state estimation and related applications.Starting from analyzing the properties of false data injection attacks,we discuss the specific attack objectives in real-time electricity markets,topology information limitations in constructing undetectable attacks,and the sparsity of attack vectors and matrices.Utilizing the properties above,we design mitigation strategies to prevent,detect,identify,and response to false data injection attacks,which can not only reduce the cost in protecting critical meters,but also increase the probabilities of attack detection and identifiaction and the resillience of power systems.Firstly,to solve the problem that the number of protected meters is too large and the protection cost is high in the current meter-protection strategies,we propose a "incentivereduction" strategy in protecting meters,and design a low-cost false data injection attack prevention strategy.Utilizing Stackelberg game infrastructure,we analyze the interactions amongst defender,attacker and system.A heuristic method is designed to search the optimal protection set of meters,which can provent financially motivated false data injection attacks with the minimal number of protected meters.Secondly,as the relationship between attack detection/identification probabilities and the changes of parameters in power systems is not analyzed,the attack detection/identification probabilities are not optimal.In this thesis,we build false data injection attack model when measurement matrix is changed,and analyze attack detection and identification conditions in reactance perturbation.Using maximum-rank matrix-completion method,the proposed reactance perturbation strategy can greatly enhance the detection and identification probabilities without significantly increase the operational cost of power systems.Thirdly,since the current state recovery methods are designed based on the fixlocation attack assumption,the performance of those methods is not good in switching location false data injection attacks.In this thesis,we model the structural sparsity of attack matrix,and analyze the condition in recovering the system states,which proves the advantages of the proposed dynamic state recovery method over the current fix-location and static state recovery methods.The proposed dynamic state recovery method can recover system states from the compromised measurements and enhance the resilience of power systems in switching location false data injection attacks.