Detecting False Data Injection Attacks using Canonical Variate Analysis in Power Grid

With the knowledge of the measurement configuration and the topology structure of a power system, attackers can launch false data injection attacks (FDIAs) without detection by existing bad data detection methods in state estimation. The attacks can also introduce errors to estimated state variables, which are critical to grid reliability and operation stability. Existing protection methods cannot handle dynamic and variable network configurations. In this paper, to effectively defend against FDIAs, we propose a canonical variate analysis based detection method which monitors the variation of statistical detection indicators $\boldsymbol{T}^2$ and $\boldsymbol{Q}$ about projected canonical variables before and after attacks. Unlike most statistic models that only consider cross-correlation of discrete measurements constrained by Kirchhoff's Law at each independent sampling time, we also consider the auto-correlation of measurements caused by time series characteristics of varying loads. Experiment results on IEEE-14 bus system demonstrate the effectiveness and accuracy of our proposed method based on both synthetically generated data and real-world electricity data from the New York independent system operator.