Multiclass False Data Injection Attacks Detection and Classification in Automatic Generation Control

The reliance of automatic generation control (AGC) on modern communication exposes it to potential cyber-attacks, particularly false data injection (FDI) attacks, posing significant risks to power systems. While existing research has addressed the impact and detection of these attacks, a gap remains in their effective classification. To bridge this gap, this paper introduces an innovative machine learning approach to accurately classify four FDI attack types (pulse, random, ramp, and scaling). Utilizing original signal data from the AGC power system, measuring frequency and tie line power deviations, the research employs error-correcting output codes with a sparse random coding design for multiclass classification. Three binary learners, namely support vector machine (SVM), k-nearest neighbors (KNN), and decision tree (DT), are employed to classify these FDI attacks. The simulation results highlight the DT binary learner's performance achieving higher precision, recall, and F-score compared to KNN and SVM. This precise multiclass FDI classification establishes a solid foundation for proposing effective mitigation strategies and enhancing cybersecurity measures.