Improving electrical fault detection: A two-stage prediction hybrid model aided by feature engineering

Abstract This study introduces an innovative method for electrical fault detection and classification prediction, aiming to meet the need for rapid identification and accurate classification of common faults in power systems. We developed a two-stage prediction hybrid model integrating CatBoost and LightGBM, specifically designed to improve the prediction accuracy of electrical fault detection and classification. Through multiple experimental validations based on simulated circuit data, our model demonstrated superior performance compared to traditional models (LR, SVM, RF, XGBoost, LightGBM, CatBoost), particularly in terms of higher accuracy, precision, and F1-score across all categories. In addition, this study also presents a targeted feature engineering approach, effectively extracting key features from fault signals. The research findings provide reliable technical support for fault prevention and maintenance in power systems, which is of significant importance for ensuring the safe and stable operation of the system, and further reveal the tremendous potential of integrating advanced machine learning technologies to enhance the stability of the power grid.