Attack Detection Of Cyber Physical Power System Based On Potential Feature Mining

Smart grid has become the trend of future power system owing to its efficient allocation of resources,real-time monitoring and decisionmaking ability.However,due to the deep integration of power physical system and information system,the smart grid becomes more vulnerable and under serious threat such as malicious attacks and so on.Once there occurs failures or attacks in the power system without timely detection,it may cause serious consequences,including huge economic losses or even security accidents.Therefore,based on the above problems,this paper studies the blind topology identification of power cyber physical system,false data injection attack detection,load redistribution attack detection,the construction of hardware in the loop simulation platform and the experimental verification of attack detection.The main contents and contributions of the paper include the following four aspects:(1)This paper proposes a blind topology identification model based on structure feature fusion.By analyzing the topology model,various characteristics of the network structure are deeply excavated,and the network structure characteristics are integrated into the structure identification process as prior knowledge,so as to realize the construction of the blind identification model of the power grid structure.The proposed topology identification model can make full use of various structural characteristics,so as to greatly improve the identification accuracy.It is found that the proposed method has accuracy advantages by experiments.(2)Aiming at the concealment of false data injection attack,this paper proposes an attack detection model based on the fusion of attack characteristics and system characteristics.Through the in-depth study of attack behavior,this paper excavates the structural sparsity of the attack matrix.By studying the stable operation characteristics of the system,it is extracted that the normal measurement matrix has low rank characteristic.Aiming at the complex problem of solving the attack detection model,an attack detection optimization algorithm based on the separation of structural sparse matrix is proposed,which can accurately separate the attack matrix from the normal measurement matrix.A large number of comparative experiments show that the proposed attack detection algorithm can effectively improve the detection accuracy.(3)Based on the periodicity and volatility of load data mining,this paper proposes a load redistribution attack detection method based on twostage learning model.Compared with the traditional supervised learning methods based on machine learning,this paper uses the unsupervised learning detection model considering the fact that the attack data is little or even hardly collected.Aiming at the problem that the machine learning method cannot accurately identify the attack behavior due to the periodicity and volatility of load data,this paper fully considers the data characteristics and proposes a new attack detection model to greatly improve the detection accuracy.(4)Based on the basic framework of industrial cyber physical system,this paper establishes the hardware in the loop simulation platform of power information physical system,realizes the integration of information layer and physical layer,and embeds the attack and defense toolbox on the simulation platform to realize the function of online attack and detection.Figures(18),Tables(11),References(84).