Short Transmission Line Fault Identification and Classification Using Supervised Machine Learning

This research presents a new approach to detecting and classifying different types of shunt faults in short transmission lines, which is based on wavelet transforms and machine learning. The approach relies on calculating root-mean-square (RMS) values of wavelet-transform coefficients from current signals on all three phases and ground (Phase A, Phase B, Phase C, and Ground) across a variable window length of the RLC branch. Faults are detected and classified using the Daubechies (db4) wavelet, which allows for comprehensive coefficient analysis with specified threshold values at different levels. A Radial Basis Function Neural Network (RBFNN) is the developed by calibrating the input and output data for maximum performance. The proposed wavelet-assisted supervised machine learning for fault identification and classification is tested with simulated power system data. For various simulated faults in short transmission lines, the ratio of energy change, signal energy, and maximum detail coefficients are analyzed. The proposed approach shows excellent performance in detection and classification of faults as compared with other models including KNN, NN, and ELR.