Pipeline Leak Detection Method based on DTWSVM

Abstract

In order to accurately identify pipeline leaks, this paper proposes an improved complementary empirical mode decomposition (CEEMD) denoising method and a pipeline leak detection method based on Deep Twin Support Vector Machine (DTWSVM). The signal is first decomposed into intrinsic modal functions (IMF) by CEEMD, and then the IMFs with more leakage information are selected for signal reconstruction through mutual information value and multi-scale permutation entropy (MPE). The obtained signal contains less noise and clear inflection points. DTWSVM is a network model combining deep neural network and TWSVM. Several original Twin Support Vector Machine (TWSVM) data in the hidden layer are mapped to the n-dimensional space, and the input and output layers are used to judge the pipeline working conditions. The experimental results show that DTWSVM can accurately judge pipeline leakage.