Small leak identification method for natural gas pipelines based on improved MVMD and DC-CNN

Abstract

Pipeline leakage is an inevitable problem. However, accurately identifying small leaks is a difficult challenge. To solve this problem, this paper proposes an effective recognition method based on iterative self-updating multivariate variational mode decomposition (I-MVMD) combined with the wavelet energy transform and DC-CNN. The method is divided into three parts. First, the I-MVMD algorithm with self-updating parameters is introduced, and the internal parameters are constantly adjusted in an iterative way to decompose the pipeline leakage signal efficiently. Then, the adaptive continuous wavelet transform is used to enhance the clustering process of only the high-energy modal signals, and the characteristics of the low-energy signals are preserved, which enhances the extraction ability of key features. Finally, it is input into the two-channel DC-CNN network designed in this paper. The high-energy channel combines an attention mechanism and max pooling to improve the perception of important features. The low-energy channel uses a large receptive field convolution to extract background information. The experimental results show that the recognition accuracy of the proposed method for small leaks reaches 97.1%, which is 10% higher than that of the traditional method.