2025 Basnet Power System Fault Detection

Electrical faults in power transmission systems can severely hinder grid stability, equipment safety, and operational reliability. Traditional protection schemes, especially distance relays, rely on apparent impedance calculations that vary with faults, leading to misclassification and potential relay maloperation. This paper presents an intelligent fault detection and classification framework using supervised machine learning techniques to overcome these challenges. The framework's robustness was validated under varying training sizes and Gaussian noise levels, showing consistent accuracy across diverse conditions. It learns the complex nonlinear mapping between three-phase voltage/current patterns and fault types, without fixed impedance paths. By utilizing line voltages and currents, a rich set of features representing six fault categories is extracted. Models such as ANN, SVM, Random Forest, XGBoost, LSTM, and Physics-Informed Neural Networks (PINN) were assessed on SMOTE-balanced datasets. Among these, the PINN achieved the highest fault detection accuracy of 99.86% and maintained multiclass classification accuracy of 99.79% on clean datasets. The PINN also demonstrated high accuracy under varying noise and training data and millisecond-level inference, bridging the gap between traditional protection and intelligent, scalable grid analytics.