An XAI-driven CNN-Transformer model for transmission line fault detection and classification

Fault detection in power transmission lines is vital for maintaining the electrical grid stability and reliability. Conventional methods for identifying and classifying faults generally suffer from inefficiencies, inaccuracies, and poor interpretability. This study introduces an Explainable Artificial Intelligence (XAI)-driven approach to enhance fault identification and classification in transmission lines. The study uses MATLAB Simulink simulated data to evaluate and compare the performance of various Machine Learning (ML) models, including eXtreme Gradient Boosting (XGBoost), Extreme Learning Machine (ELM), Long-Short Term Memory (LSTM), Long Short-Term Memory-Gated Recurrent Unit (LSTM-GRU), and Vision Transformer (ViT) against the proposed CNN-Transformer hybrid model. Results demonstrate that the CNN-Transformer hybrid model outperforms alternative models, achieving superior fault detection accuracy across diverse resistance values. The CNN-Transformer model achieved the highest accuracy (0.97), followed by XGBoost (0.93), LSTM-GRU (0.92), LSTM (0.88), ViT (0.93), and ELM (0.87). Additionally, we implement SHAP, an explainable AI framework, to enhance transparency and interpretability in fault diagnosis. The developed methodology supports a mobile application enabling utility operators to monitor and identify faults in real time. This work illustrates ML along with XAI’s potential in power system fault detection, paving the way for intelligent and reliable grid management solutions. • Novel CNN-Transformer hybrid model achieves 97 % fault detection accuracy. • Outperforms XGBoost (93 %), LSTM-GRU (92 %), LSTM (88 %), ViT (0.93) and ELM (87 %). • Robust performance across diverse fault resistance values. • SHAP integration enhances model transparency for diagnostics. • Mobile app enables real-time fault monitoring for practical use.