Identification Of False Data Injection Attack Behaviors In Wide Area Measurement System Based On Spatiotemporal Correlation

Information and communication technologies drive the development of intelligent power systems but also bring new security risks.False data injection attack is an important category of network attacks on power system,which can stealthily evade the traditional mechanism for bad data detection,alter state estimation results and cause harm to power systems.So,it’s a security risk that cannot be ignored.However,most of the existing detection methods for false data injection attacks are based on independent analysis of the attack events and lack spatiotemporal correlation feature mining of the attacked data.Therefore,in order to better detect false data injection attacks and strengthen defensive of grid,thesis focus on false data injection attacks in wide-area measurement systems.Main work and contributions are as follows.Firstly,the mechanism of the false data injection attack is analyzed in depth for the study of the characteristics of false data injection attacks.A non-linear false data injection attack model under incomplete information conditions is constructed,considering the difficulty for the attacker to obtain all configuration information of the power grid under practical circumstances and the state estimation based on the AC current model.Then three different attack regions are defined in the 39-node system,and through simulation,the attack is successfully launched on the target bus.The success probability of this model under different attack strengths and the impact of different attack scales on the power system are investigated.In terms of spatiotemporal correlation analysis of the attack data,the temporal correlation of the node data is predicted by using the Cubature Kalman Filter considering the temporal autocorrelation of the system node data,and the spatial correlation of the node data is predicted by using the Gaussian Process Regression considering the spatial cross-correlation of the system node data.Simulations show that the prediction results of the two methods are very close to the real values,and it is found that the correlation coefficients between the node data change significantly when an attack is launched.A novel hybrid neural network detection model based on spatiotemporal correlation analysis is proposed for detecting false data injection attacks.The model leverages the advantages of two neural network architectures,Long-Short Term Memory and Convolutional Neural Networks,to effectively extract spatiotemporal features.The model is trained by quantile regression to predict normal bounds of state estimation and to identify attacks according to the detection mechanism.The final results show that the proposed model can detect different types of attacks with high accuracy in both 39-node and 118-node systems.