Research On False Data Injection Attack Detection In Power Cyber-Physical System

With the widespread application of information technology in the power system,the power grid is developing into a Cyber-Physical System(CPS)that deeply integrates cyber systems and physical power grids.The complexity and openness of power CPS make it more susceptible to network attacks.False Data Injection Attack(FDIA)against Energy Management System(EMS)is a typical network attack that can bypass bad data detectors and directly interfere with state estimation results,thereby affecting the stable operation of the power grid.Based on this attack characteristic,FDIA poses great challenges to current detection methods based on state estimation,such as the accuracy of state estimation,setting of a priori thresholds and other issues.Therefore,it is crucial to study the detection methods for FDIA for the safe operation of power CPS.In summary,this article conducted research on FDIA detection,and the main work is as follows::(1)A robust adaptive UKF based attack detection method is proposed to address the significant decrease in estimation accuracy of unscented Kalman Filter(UKF)dynamic estimation prediction steps with mixed false data.Firstly,a state prediction model was established by utilizing the historical state information and measurement data of historical databases,which are not affected by FDIA;Then,considering different situations,the accuracy of state prediction values and UKF dynamic estimation values based on multiple regression modeling is different.An adaptive weight determination method is proposed,and the state prediction values of the two are weighted and fused to obtain a robust adaptive UKF method;Further obtain the state difference between the predicted results and the weighted least squares(WLS)static estimation values,and determine whether FDIA has occurred after consistency testing.Finally,simulation experiments have demonstrated the effectiveness of the proposed detection algorithm in small systems.(2)Considering that power CPS data naturally contains linear and nonlinear components,a detection scheme based on Kalman filtering(KF)and convolutional neural network(CNN)combined with long short term memory(LSTM)is proposed.Parallel analysis is conducted using corresponding prediction models(KF for linear models and CNN-LSTM for nonlinear models),and then ensemble learning is used to automatically assign weights to the results of the two base learners and output the final prediction results.Secondly,the threshold obtained from the cumulative distribution curve of the sum of squared errors between the concentrated predicted values and the actual values of the unasked data is used to judge the prediction results.Compared with CNN,LSTM,KF,and CNN-LSTM,the detection method in this chapter shows a higher detection rate and better detection results.