A dual-channel signals-based method for pipeline leak detection by signal reconstruction and deep feature fusion

Abstract

Accurate detection of pipeline leaks is crucial for the stable operation of urban natural gas pipelines. Pipelines are typically situated in complex environments, and most leak detection methods rely solely on data from a single type of sensor. Due to the impact of environmental noise and the inability of a single type of sensor to comprehensively describe the operational state of pipelines, these methods often have high false detection rates. Accordingly, this paper proposes a dual-channel signals-based method for pipeline leak detection by signal reconstruction and deep feature fusion. The contributions are two fold: First, a Variational Mode Decomposition with Dual Signals Dynamic Time Warping (VMD-DSDTW) method is designed for signal reconstruction and noise removal of the dual-channel signals. Second, an AutoEncoder based Feature Fusion Convolutional Neural Network (AE-FFCNN) method is devised for dual-channel signals information fusion and pipeline leak detection. The proposed method was tested based on data extracted from an experimental platform, and the results demonstrated the superior performance of the proposed method by comparison with other state-of-the-art methods.