Salient Feature and Dual Attention Based Method for Fault-Cause Identification in Transmission Line

The stable operation of power transmission lines is fundamental to national economic growth and energy security, acting as the primary arteries of the electrical grid. Accurate and rapid identification of fault causes is not only crucial for swift service restoration but also essential for implementing targeted protective measures to enhance overall grid reliability. However, fault-cause data collected from actual transmission line incidents are often contaminated by environmental noise and on-site interference. These background disturbances degrade signal quality, limiting the effectiveness of existing deep learning models and impeding accurate fault-cause identification. To address this, a novel fault-cause identification framework based on salient feature selection and dual attention is proposed. The framework incorporates a salient feature selection (SFS) module that dynamically highlights discriminative regions in the input while suppressing redundant information, and a dual attention (DA) module that enhances both spatial localization and channel-wise feature interactions. Together, these modules strengthen the model’s ability to extract robust fault representations under complex conditions. Experiments on field data from the State Grid Electric Power Research Institute demonstrate that the proposed method achieves precision rates of 90.13%, 86.74%, and 77.12% across three diagnostic tasks. In addition, visualization using saliency maps and confusion matrices confirms the model’s interpretability and reliability. This method provides a practical and scalable solution for fault-cause identification in real-world power transmission systems, with the potential to enhance grid stability, improve the overall efficiency of fault management and preventive maintenance in power networks.