An Explainable Machine Learning Framework for Fault Detection, Classification, and Localization in Smart Distribution Networks

Abstract

Smart distribution networks are more complicated due to the combination of distributed generation, renewable energy sources, and enhanced monitoring technologies. These developments describe difficulties in determining electrical defects, identifying different types of faults, and precisely finding fault locations under various operating circumstances. This study suggests an Explainable Artificial Intelligence-Based fault Detection, Classification, and Localization (XAI-FDCL) in smart distribution networks. The suggested structure uses three-phase voltage and current measurements to extract pertinent electrical information, such as sequence components, statistical traits, and root mean square values. For detecting fault conditions and categorizing common fault types, such as single-line-to-ground, line-to-line, double-line-to-ground, and three-phase faults, Random Forest, Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost) algorithms are evaluated. To determine the fault's position within the distribution network, feeder-section classification and distance regression are used in fault localization. Furthermore, Shapley Additive explanations (SHAP) are integrated to determine the electrical characteristics that aid in defect diagnosis and to evaluate model forecasts. The framework is evaluated using accuracy, precision, recall, F1-score, mean absolute error, root mean square error, and computational latency. The suggested approach aims to model transparency, localization accuracy, and fault-diagnosis reliability in order to facilitate better decision-making and effective fault management in smart distribution networks.