Pipeline multi-point leakage identification based on temporal convolutional network

Abstract

Accurate pipeline leakage localization remains a major challenge due to signal attenuation, noise interference, and modeling complexity. To address this, we propose a Temporal Convolutional Network (TCN)-based method that leverages negative pressure wave signals and supervised learning to directly map sensor inputs to leakage positions, avoiding the need for complex physical modeling. The method was validated through simulation and experimental studies, showing high accuracy in single-leakage scenarios. Comparative analysis indicates that TCN outperforms LSTM and Transformer models in localization tasks. For multi-leakage conditions, the model successfully identified multiple leakage points, and the impact of different sensor configurations on prediction performance was further analyzed. The results indicate that the proposed method simplifies the modeling process, enhances prediction accuracy, and shows strong potential for real-world pipeline monitoring and early-warning systems.