A Robust Machine Learning Approach for Ac Microgrid Fault Diagnosis Using FFT and XGBoost

The increasing penetration of renewable energy sources in AC microgrids necessitates intelligent and datadriven fault diagnosis techniques to maintain system stability and reliability. This paper presents a novel fault detection and classification framework utilizing Fast Fourier Transform (FFT) for feature extraction and Extreme Gradient Boosting (XGBoost) for fault identification and classification. FFT transforms time-domain voltage and current signals into the frequency domain, enabling the extraction of critical features such as power spectral density (PSD), RMS, spectral entropy, and median frequency. These features are used to train the XGBoost classifier, known for its robustness, speed, and accuracy. The proposed method is validated on a seven-bus AC microgrid model simulated using RSCAD, encompassing various fault scenarios such as LG, LL, LLG, LLL, and LLLG. A comprehensive comparison with Decision Tree (DT), Random Forest (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R F}$</tex>), and CatBoost models demonstrates the superior performance of XGBoost, achieving an overall accuracy of 99.72 %. The results highlight the method's effectiveness and robustness in accurately detecting and classifying faults under diverse conditions, making it suitable for real-time deployment in intelligent microgrid protection systems.