Accurate, fast, and interpretable fault identification on electrical transmission lines is essential for maintaining power system stability and reducing outage durations. In this study, we propose a hybrid 1D convolutional neural network–Decision Tree (1D-CNN–DT) for transmission line fault detection and classification, in which the 1D-CNN acts solely as a feature extractor. During this process, the Decision Tree performs the final, interpretable classification. By preserving decision transparency and achieving high diagnostic accuracy, the proposed architecture differs from conventional end-to-end deep learning models. In MATLAB/Simulink, four distinct transmission line scenarios were simulated to evaluate the framework under realistic operating conditions. These scenarios included short lines, long distributed lines, source-end faults, and load-end faults. We developed a large, balanced dataset of three-phase voltage and current measurements per unit, covering standard operation and ten types of faults. According to the proposed model, fault detection accuracies were 99.89%, 99.94%, 99.94%, and 99.97%, and fault classification accuracies were 99.93%, 99.58%, 99.44%, and 99.86% across the four transmission line configurations. In addition to its high accuracy, the hybrid framework demonstrated significantly lower computational complexity and shorter training and inference times than conventional ANN- and LSTM-based approaches, without requiring manual signal transformations. The SHapley Additive Explanations (SHAP) are integrated to enhance trust and practical usability, providing both global and instance-level interpretability that reveals how voltages and currents contribute to individual faults. According to the results, a hybrid architecture that combines deep learning and explainable AI offers reliable, efficient, and transparent real-time transmission line monitoring and protection.
Biswas et al. (Fri,) studied this question.