Spatio-Temporal Correlation-Based False Data Injection Attack Detection Using Deep Convolutional Neural Network
Guangdou Zhang, Jian Li, Olusola Bamisile, Dongsheng Cai, Weihao Hu, Qi Huang
Abstract
There are lots of cyber-attack, especially false data injection attacks, in modern power systems. This attack can circumvent traditional residual-based detection methods, and destroy the integrity of control information, thus hindering the stability of the power system. In this paper, a novel Spatio-temporal detection mechanism is proposed to evaluate and locate false data injection attacks. In the proposed method, temporal correlation and spatial correlation are analyzed by cubature Kalman filter and Gaussian process regression, respectively, to capture the dynamic features of state vectors. Then, a deep convolutional neural network is trained to depict the functional relationship between Spatio-temporal correlation functions and the output, which is set as the detection indicator to access whether the power system under attack or not. Furthermore, the performance of the proposed mechanism is evaluated with comprehensive numerical simulation on IEEE 39-bus test system. The results of the case studies showed that the proposed method can achieve 99.84%-100% accuracy.