Detection And Recovery Of False Data Injection Attacks In Power Systems Based On Data-Driven Algorithms

Power system state estimation plays a crucial role in the stable operation of modern smart grids,while it is vulnerable to cyber attacks.Existing research mainly focuses on false data injection attacks(FDIA),it can tamper with measurement data and bypass the bad data detection mechanism,leading to incorrect state estimation results.This dissertation proposes a data-driven FDIA defense framework against power system state estimation,which includes an anomaly detection part and a data recovery part.For the anomaly detection part,the FDIA detection framework based on graph edgeconditioned convolutional network is proposed.The model can utilize the topology information,node features and edge features of power systems.Through deep graph architecture,the correlation of sample data is effectively mined to establish the mapping relationship between the estimated values of measurements and the actual states of power systems.Besides,the edgeconditioned convolution operation allows processing data sets with different graph structures.Case studies are undertaken on IEEE 14-bus system under different attack intensities and attack degrees to evaluate the detection performance.Simulation results show that the proposed model has better detection performance compared with commonly used models such as convolutional neural network,deep neural network and support vector machine.Furthermore,the satisfactory detection performance obtained with the data sets of IEEE 14-bus,30-bus and 118-bus systems verifies the effective scalability of the proposed model.For the recovery part,a data recovery framework for FDIA based on variational convolution auto-encoder is proposed to ensure continuous monitoring of state estimation.The model combines deep learning ideas with Bayesian inference.Besides,it uses convolution and deconvolution operations with excellent feature capture capabilities in the encoder network and decoder network,respectively,to effectively restore the abnormal values after FDIA to the values close to normal operation.Moreover,knowledge distillation technique is used to compress the recovery model,making it possible to deploy a lightweight model on equipment with limited resources.Case studies are undertaken on IEEE 14-bus system under different attack intensities and attack degrees to evaluate the recovery performance of the model.Simulation results show that the mean absolute error and mean absolute percentage error of the proposed model are lower than those of the comparison models.Furthermore,the satisfactory recovery performance in IEEE 30-bus and 118-bus systems validates the scalability of the proposed model.In addition,the knowledge distillation technique allows the scale of the lightweight recovery model to be about one tenth the size of the original model,with almost no increase in mean absolute error and mean absolute percentage error.