False Data Injection Attack Detection In Electric Cyber-Physical Systems

As the level of intelligence of Cyber Physical System(CPS)continues to increase,its security issues become more and more important.False Data Injection Attack(FDIA)is highly accessible,concealed and destructive in information physical systems.In addition,with the increasingly complex power topology and increasing power measurement data,the detection performance of False Data Injection Attack has been improved by introducing artificial intelligence algorithms to ensure the safe operation of power systems.(1)To improve the anti-interference of false data injection attack detection,this thesis proposes a stacked auto encoder(SAE)-based false data injection attack detection model.The detection model consists of an SAE predictor and a classifier.The false data is first predicted,and then the normal data and abnormal data are classified by the classifier to achieve the detection of false data injection attacks.The gray wolf optimization algorithm is added to the SAE predictor part to optimize the hyperparameters of the SAE predictor network to enhance the network convergence speed and improve the efficiency of the model data prediction.(2)In order to further improve the detection accuracy of false data injection attacks as well as to consider the temporal characteristics of temporal sequences,this thesis proposes an improved Convolutional Neural Network-Long Short Term Memory(CNN-LSTM)model for false data injection attack detection.The Long Short-Term Memory(LSTM)is inserted between the pooling layer and the fully connected layer of the convolutional neural network to extract temporal and spatial information features for data prediction of false data injection attacks,and a Softmax classifier is used to classify normal data and abnormal data to achieve the detection of false data injection attacks.Merging the forgetting gate with the input gate in the LSTM unit part reduces the learning parameters and simplifies the information storage structure,and at the same time,improves the generalization ability of the detection model,reduces overfitting and accelerates the training speed.(3)In order to improve the detection range of false data injection attacks and achieve false data correction,this thesis proposes a combination of Light Gradient Boosting Machine(Light GBM)and Unscented Kalman Filter(UKF)based false data injection attack detection and correction model.First,the load of the power system is predicted using the Light Gradient Boosting Machine.Then the load prediction results are used as state variables and state estimation is performed using the traceless Kalman filter.Finally,random variables are constructed based on the state estimation for false data injection attack detection.In the data correction section,Youden index is used to determine the optimal alert threshold to correct the spurious data.