Safeguarding Gas Pipeline Sustainability: Deep Learning for Precision Identification of Gas Leakage Characteristics

The growing demand for natural gas and the corresponding expansion of pipeline networks have intensified the need for precise leak detection, particularly due to the increased vulnerability of infrastructure to natural disasters such as earthquakes, floods, torrential rains, and landslides. This research leverages deep learning to develop two hybrid architectures, the Transformer–LSTM Parallel Network (TLPN) and the Transformer–LSTM Cascaded Network (TLCN), which are rigorously benchmarked against Transformer and Long Short-Term Memory (LSTM) baselines. Performance evaluations demonstrate TLPN achieves exceptional metrics, including 91.10% accuracy, an 86.35% F₁ score, and a 95.20% AUC value. Similarly, TLCN delivers robust results, achieving 90.95% accuracy, an 85.76% F₁ score, and 93.90% of the Area Under the ROC Curve (AUC). These outcomes confirm the superiority of attention mechanisms and highlight the enhanced capability realized by integrating LSTM with Transformer for time-series classification. The findings of this research significantly enhance the safety, reliability, sustainability, and risk mitigation capabilities of buried infrastructure. By enabling rapid leak detection and response, as well as preventing resource waste, these deep learning-based models offer substantial potential for building more sustainable and reliable urban energy systems.