Real-time Detection In Power Grid Against False Data Under The Environment Of Information Physical Fusion

With the developing of the cyber-physical system towards the direction of intelligence and scale,while bringing convenience to the society,the stability and economy of the system are facing more and more severe challenges.As a kind of abnormal data different from noise and singular value,the false data in the system can directly affect the accuracy of measurement data or system state estimation,lead to intelligent decision-making errors,and then lead to equipment abnormal actions,resulting in large-scale adverse effects.Therefore,based on the background of smart grid,in this paper,the real-time detection algorithm against false data is studied.Aiming at the shortcomings of the existing algorithm,such as large detection delay and low accuracy,the main work of this paper is as follows:(1)For false data detection on the bus side of power grid,for the shortcoming that the detection algorithm based on single index such as KL divergence(KLD)or generalized log likelihood ratio(GLLR)can not give attention to both accuracy and rapidity,and considering the influence of noise and outliers on the accuracy,the sliding time window is introduced into the measurement sequence,and a multi index hybrid detection algorithm based on parallel unscented Kalman filter(UKF)is proposed under the alternating current(AC)power flow model.The detection performance of the algorithm is tested and analyzed,and compared with the detection algorithm based on a single index.The numerical simulation shows that the algorithm has obvious advantages in the AC power flow model;in the direct current(DC)power flow model,the detection delay per 100 test sequences is also reduced by 13 sampling periods while ensuring the detection accuracy.(2)For the false load data detection on the user side of the power grid,for the problem of low accuracy and poor real-time performance of the single model detection algorithm based on data prediction using the long-short term memory(LSTM)network and data reconstruction using the stacking automatic encoder(SAE),the influence of different combination prediction models of automatic encoder and LSTM network on detector performance is studied.And a hybrid prediction network detection algorithm based on theKLD of prediction residual sequence is proposed.The detection performance of the proposed algorithm for different types of false data is tested and analyzed,and compared with the detection algorithm based on stacked LSTM network prediction model.In the simulation experiment on NYISO load data set,several evaluation indexes show that the proposed SAE-LSTM and SAE-LSTM_H algorithm has better detection performance.