A Robust Data Driven FDI attack Detection Framework for Inverter based Microgrids

False Data Injection (FDI) attacks pose a major cybersecurity risk to power electronic systems, particularly inverter-based microgrids. These attacks compromise signals like current, voltage, and phase, causing incorrect power flow calculations, inverter instability, and system outages. This study presents a data-driven approach to detect FDI attacks using an attack classifier trained on data generated from a Model Predictive Control (MPC)-based current control scheme for a three-phase, two-level inverter. The dataset includes normal and attack-induced current deviations, enabling robust model training. The study addresses multiple FDI scenarios, including amplitude, offset, and phase manipulations. A multi-class classification approach is adopted, utilizing machine learning classifiers such as Support Vector Classifier (SVC), Decision Tree, Naive Bayes, Ensemble Boosting, and Neural Networks (NN). Classifier performance is evaluated using accuracy, precision, recall, and F1 score. Among the classifiers, the Neural Network outperforms the rest, achieving an overall accuracy of 99.6%, demonstrating superior detection for both covert and stealthy attacks. The use of an MPC-based control framework to generate training data adds novelty to the approach. The proposed methodology is implemented using MATLAB Simulink, highlighting the potential of neural networks in enhancing the cybersecurity of inverter-based microgrids.