Canonical Variate Analysis for Detecting False Data Injection Attacks in Alternating Current State Estimation

Estimating the accurate states of voltage magnitudes and angles ensures reliable operation and control in a smart grid. The increased dependency and integration of information communication technologies in smart grids introduce new security issues, such as hidden false data injection attacks (FDIAs). These can successfully evade conventional residual-based detection mechanisms and cause bias to the estimated states. Since complicated power systems exhibit nonlinear characteristics, it is particularly critical to achieving satisfactory detection performance against FDIAs in Alternating Current (AC) state estimation models rather than widely-used simplified Direct Current (DC) models. This paper proposes a novel Canonical Variate Analysis (CVA)-based detection approach against FDIAs in an AC power system. Our proposed method utilizes the fact that the occurrence of FDIAs affects the correlation of consecutive measurements, including cross-correlation and auto-correlation. Unlike many of the previous studies in DC estimation models, which are not very realistic, we study the detection performance of the CVA-based detection method for AC estimation, in which kernel density estimation is introduced to determine detection thresholds. Furthermore, instead of adding advanced Phasor Measurement Units' measurements or forecasting-aided redundant measurements to increase the attack detection capability, the proposed approach is a data-based multivariate statistical monitoring process that directly monitors changes in multiple state variables. The performance of our proposed approach is verified on an IEEE 14 bus system, and different attack scenarios are considered in our experiments. Experiment results indicate that our approach performs better for FDIAs than an existing Kullback-Leibler-Distance detector.