Gas pipeline networks are vital urban infrastructure, susceptible to leaks caused by natural disasters and adverse weather, posing significant safety risks. Detecting and localizing these leaks is crucial for mitigating hazards. However, existing methods often fail to effectively model the time-varying structural data of pipelines, limiting their detection capabilities. This study introduces a novel approach for leak detection using a spatial–temporal attention network (STAN) tailored for small-hole leakage conditions. A graph attention network (GAT) is first used to model the spatial dependencies between sensors, capturing the dynamic patterns of adjacent nodes. An LSTM model is then employed for encoding and decoding time series data, incorporating a temporal attention mechanism to capture evolving changes over time, thus improving detection accuracy. The proposed model is evaluated using Pipeline Studio software and compared with state-of-the-art models on a gas pipeline simulation dataset. Results demonstrate competitive precision (91.7%), recall (96.5%), and F1-score (0.94). Furthermore, the method effectively identifies sensor statuses and temporal dynamics, reducing leakage risks and enhancing model performance. This study highlights the potential of deep learning techniques in addressing the challenges of leak detection and emphasizes the effectiveness of spatial–temporal modeling for improved detection accuracy. [ABSTRACT FROM AUTHOR]

Copyright of Sensors (14248220) is the property of MDPI and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)