Machine Learning for Fault Detection, Classification, and Location in Transmission Lines: A Case Study of Bonga–Mizan, Ethiopia

Abstract Faults in transmission lines are among the leading causes of power system instability, equipment damage, and supply interruption. Accurate and rapid fault detection, classification, and location are vital for maintaining system reliability and minimizing outage duration. However, traditional distance relays used in the Ethiopian power grid face challenges in identifying complex fault types and precisely locating fault points, thereby reducing the speed and dependability of fault clearance. This study proposes an intelligent prediction scheme that integrates wavelet transform–multi-resolution analysis (WT-MRA) for fault detection and a hybrid convolutional neural network–long short-term memory (CNN-LSTM) model for fault classification and location. The model was developed and simulated in MATLAB using data from the 88 km Bonga–Mizan transmission line, with actual system parameters from Ethiopian Electric Power. The WT-MRA achieved detection delays as low as 4.1–13 ms, outperforming simulated and real-time CYG PRS-753 distance relays, which averaged 30.9 ms and 411 ms, respectively. The CNN-LSTM model achieved 99.8% classification accuracy and improved fault location accuracy by 93.2%, with an RMSE of 0.072 km compared to 1.059 km for the relay. The proposed hybrid approach demonstrates superior dependability and accuracy, offering a promising direction for smart grid fault diagnosis in long power transmission lines.