Machine Learning-Based Intelligent Fault Detection and Classification in Power Networks

Abstract

Power system faults, including ground disturbances and short circuits, pose a substantial hazard to the reliability of electrical power systems, which are considered critical infrastructure. It is imperative to maintain system stability and minimize service interruptions by rapidly and precisely identifying and classifying faults. This paper proposes a durable framework for real-time fault detection and multi-class classification in electrical power grids, powered by AI-driven models trained via supervised learning. To produce exhaustive datasets of three-phase voltage and current signals under a variety of fault and normal operating conditions, an IEEE 5-bus power system is modelled and simulated in MATLAB/Simulink. These measurements are processed and used as input features for a variety of classifiers, including Random Forest, Support Vector Machine, Decision Tree, and Logistic Regression. Model performance is assessed for both binary and multi-class classification tasks by analyzing the confusion matrix, accuracy, along with precision, recall, and F1-score. Experimental results demonstrate that the Random Forest classifier consistently outperforms other models. It exhibits high robustness to noise and strong capacity to capture the behavior of nonlinear systems. The increasing availability of high-resolution measurement data from modern monitoring infrastructure further enhances the applicability of machine learning–based fault diagnosis. The proposed framework enables fast, reliable, and scalable fault identification, providing an effective foundation for intelligent protection schemes in modern and future smart grid environments.