Increasing Detection Rate of False Data Injection Attacks Using Measurement Predictors

A novel machine learning-based approach for the detection of False Data Injection Attacks (FDIAs) is presented. Existing machine learning-based FDIA detection methods exhibit a higher percentage of false negatives that may compromise the power grid's reliability. The proposed approach practically eliminates false negatives through the combination of a measurement classifier and measurement predictors. The detection rate increases drastically by detecting attacked measurement vectors in parallel using these two orthogonal mechanisms. Experimental results obtained using the IEEE 14-bus system support the effectiveness of the presented approach.