A Novel Hybrid CNN-LSTM Compensation Model Against DoS Attacks in Power System State Estimation

Denial of Service (DoS) attack blocks the transmission of the power system measurements by the interference, which greatly degrades the performance of power system state estimation performance. In order to reduce the impact of DoS attacks on estimated performance, it is necessary to compensate for lost measurements. In this paper, a hybrid compensation model based on deep neural networks is proposed for power system state estimation under DoS attacks. In this compensation model, the advantages of Convolutional Neural Network (CNN) and Long-Short Term Memory Neural Networks (LSTM) are respectively utilized in the automatic feature extraction and long-term prediction of measurements. The spatio features of the measurements are fully figured out by squeeze-and-excitation blocks in the CNN, which enables the characteristic response of the input channel can be adaptively adjusted. Moreover, an autoregressive is applied for dealing with the problem of the scale insensitivity of neural network models. The proposed model is validated in a Cubature Kalman Filtering (CKF) state estimation algorithm on the IEEE-30 and IEEE-118 bus test systems. Simulation results illustrate that the CKF algorithm with the proposed model withstands the DoS attacks and performs a higher estimation accuracy than that with the LSTM, CNN, and the Multilayer Perceptron (MLP).