Experimental investigation on leakage localization in water pipeline based on IGWO-VMD and hybrid neural network

Abstract

Water pipelines are highly susceptible to leakage, which can lead to safety hazards, economic losses, and environmental damage. To detect leakage losses, an experimental study is conducted to collect a dataset of 969 pressure signals, including leakage, stability, and interference signals. The results show that as the volume flow decreases and the distance between the leakage point and the sensor increases, the signal strength significantly attenuates. This attenuation reduces the distinguishability of leakage signals—especially under noisy conditions, where noise may even obscure the leakage features—thereby increasing the difficulty of accurate localization. To mitigate noise in signals, this paper proposes an adaptive denoising algorithm (IGWO-VMD) based on variational mode decomposition (VMD) and an improved grey wolf optimizer (IGWO), which effectively suppresses noise and enhances leakage features. In addition, to achieve accurate leakage localization in the presence of interference signals—such as valve closures that closely resemble leakage signals—a hybrid neural network combining temporal convolutional networks (TCN) and bidirectional long short-term memory (BiLSTM) is proposed. This model exhibits strong feature extraction capabilities and robust performance in complex signal environments, effectively reducing the risk of misjudgment. Comparative evaluations with state-of-the-art methods across multiple performance metrics show that the model accurately identifies leakage signals and precisely localizes leakage positions, demonstrating its potential as a reliable solution for intelligent water pipeline management.