Research on leakage detection and location method of natural gas pipeline under strong noise condition

Under strong-noise conditions, accurate leak localization in natural gas pipelines remains a considerable challenge. Conventional methods commonly suffer from insufficient feature extraction capability, weak noise resistance, and unstable localization performance. To address these issues, this study focuses on precise localization after leakage occurrence. Instead of conducting a separate binary leakage/no-leakage detection task, the research objective is defined as leak interval classification and continuous leak location regression.In this study, an experimental platform integrating leakage signal acquisition, storage, and processing was established, and a pipeline leakage signal dataset under strong-noise conditions was constructed. For the leak interval classification task, a Multi-Channel Hybrid Self-Attention Convolutional Neural Network (MCSA) was proposed. The model consists of a feature extraction module, an adaptive feature selection module, a multi-path mechanism, and a Softmax classifier. By integrating multi-channel feature extraction with a channel-spatial attention mechanism, the model adaptively enhances informative leakage features while suppressing noise interference. For the continuous leak location regression task, a convolutional neural network incorporating a Multi-level Cascaded Residual Mixed Attention mechanism (MCR-MA) was further proposed to enhance the representation of complex leakage features and improve model robustness and generalization capability.The experimental results show that, under strong-noise conditions, the MCSA model achieves an average classification accuracy of 99.2%, demonstrating strong leak interval identification capability and good noise robustness. The MCR-MA model achieves an average localization accuracy of 98.75% in the continuous leak location regression task, with an average bias of 0.0051 and an R 2 value close to 1, indicating a high degree of agreement between the predicted and actual leak locations.The proposed MCSA and MCR-MA models can achieve stable and accurate natural gas pipeline leak localization in complex strong-noise environments. The proposed method effectively improves leakage signal feature extraction capability and noise robustness, providing a technically promising solution for intelligent monitoring and safety warning of natural gas transmission pipelines.