False Data Injection Attack And Detection For Power Cyber-Physical Power System

The traditional power grid has developed into a power cyber-physical system with deep integration of the power grid and the information network,which has greatly promoted the intelligent process of the power grid.However,the access to the information network increases the risk of network attacks on the power system,so the research on network attacks on the power system is of great practical significance.As a new type of network attack,false data injection attacks exploit the loopholes in bad data detection and illegally tamper with the received data in state estimation,which poses a serious threat to the safe and stable operation of power systems.Starting from the false data construction method,this paper mainly completes the following research work:Without mastering the topology information of the power grid,this paper proposes a new method to construct false data attack vectors based on the improved principal component analysis method.Firstly,on the premise of ensuring that the data characteristics are not lost,the high-dimensional measurement data set is reduced to a low-dimensional space by the improved principal component analysis method,and then the Jacobian matrix is approximated by using the low-dimensional principal component matrix,and combined with the principle of false data injection attack constructs an attack vector with strong concealment.The effectiveness and practicability of the construction method are verified through simulation experiments in different systems.The traditional bad data detector cannot identify the well-designed false data.To solve this problem,this paper proposes a detection method based on wavelet transform and extreme learning machine.Using wavelet transform to decompose the time series of state variables,the wavelet component will capture the data anomalies caused by false data.In order to quantify the abnormal situation of the time series of state variables,three feature data are constructed to distinguish whether the data is normal or not.The feature data is used as the input layer of the extreme learning machine,and the corresponding output layer is set as the classification label for judging whether the data is abnormal,and the parameters of the extreme learning machine are trained through a large number of feature data samples.For the suspicious data identified by the classification detection model,the correlation coefficient analysis method is used to find out whether the correlation coefficient between the suspicious data and the normal data exceeds the judgment threshold,and further identify whether the suspicious data is caused by false data injection attacks.The simulation results show that the detection success rate of the detection scheme is at a high level under different scales of power systems and different attack intensities,which confirms its effectiveness a`nd universality.After the detector identifies that the power grid is attacked by false data injection,the data center will remove the measurement data and state quantities at the corresponding moment,so the data is still incomplete.In order to ensure the integrity of the data,this paper proposes a defense method to predict the state sequence data using historical data sets by analyzing the temporal correlation of state variables.First,the state sequence is decomposed into multiple modal components with strong regularity by the variational modal decomposition algorithm,and each component is used as the input layer and output layer of the extreme learning machine,and the weight parameters and bias are obtained by training a large number of data samples.parameter.At the same time,the parameters are optimized in combination with the particle swarm optimization algorithm to find the best prediction model corresponding to each subsequence,and the predicted state sequence can be used to fill the incomplete data set to achieve the defense effect.The simulation results confirm that the absolute error of the forecast data only fluctuates within an acceptable range,which can completely replace the real data of the state sequence,which verifies the accuracy of the forecast method.