Real-time Detecting False Data Injection Attacks Based on Spatial and Temporal Correlations

The false data injection (FDI) attack is a threat to the cyber-physical power system, which may bypass bad data detections and induce substantial damages. To mitigate such a threat, a novel detection method is proposed in this work, which utilizes both the spatial and temporal correlations among measurements to identify anomalies caused by FDI attacks. The vector auto-regression (VAR) is applied to form a set of spatial relation matrices, which reflects correlations between the latest measurement vector and previous several ones. A time sequence of such relation matrix sets is taken as data samples for training an hidden-Markov model (HMM). This HMM recovers the temporal correlations among successive measurement vectors, which helps to distinguish abnormal measurements by estimating their conditional probabilities. Then, a statistical method is introduced to locate the attack source according to numeric variations of the spatial relation matrices. Case studies on an IEEE 9 bus system are presented. Test results validate the efficacy of the proposed FDI attack detection method.