Intelligent Fault Identification in Power Systems Via Wavelet Transform and Machine Learning: A Comparative Study of Four Classifiers

Accurate fault identification remains a critical challenge in modern power system protection, particularly under varying fault impedance conditions. This paper proposes an intelligent fault identification framework that combines the Discrete Wavelet Transform (DWT) with machine learning classifiers. The IEEE 9-bus power system is modeled in MATLAB/Simulink, where twelve fault types are applied at two bus locations with fault impedances ranging from 0.0001 Ω to 10 Ω. Pre-fault operating conditions are established using the Newton–Raphson load-flow method. Fault current signals are processed using DWT with the Daubechies (db4) mother wavelet, and the maximum absolute detail coefficients of three-phase and ground currents are extracted to form a four-dimensional feature vector from 2400 simulation samples. Four classifiers are evaluated: Logistic Regression, Support Vector Machine (SVM), Random Forest, and Artificial Neural Network (ANN). Performance is assessed using classification accuracy, confusion matrices, ROC analysis, and computational efficiency. Results show that non-linear classifiers outperform linear approaches, achieving accuracies above 95%. The ANN attains the highest accuracy (98.3%) and the greatest robustness to increasing fault impedance, while Random Forest provides a favorable balance between accuracy and computational cost. The proposed framework offers an effective solution for intelligent fault diagnosis in power systems.