Leakage aperture recognition based on ensemble local mean decomposition and sparse representation for classification of natural gas pipeline

Leakage in a natural gas pipeline is influenced by many factors, including aperture, distance from sensors, and pressure inside the pipeline. The feature extraction and recognition algorithm is complex; thus, a new leakage aperture recognition method is proposed that presents a feature extraction algorithm based on the Ensemble Local Mean Decomposition (ELMD)-K-L (Kullback-Leibler) model and Sparse Representation for Classification. This method applied ELMD to perform adaptive decomposition of the leakage signals and obtain feature information of the leakage signals with different apertures. It then selected the product function components that contained major leakage information according to the K-L divergence from which we extracted a variety of time-frequency feature parameters to obtain the comprehensive and accurate eigenvector of the leakage signal. Realization of an accurate classification of leakage aperture using sparse representation classifiers was proposed to classify small samples of the complex signals. The classifiers obtained the sparsest solution of the test signal through the over-complete dictionary and used this solution as the sparse reconstruction coefficients of the test signal to obtain the reconstructed signal of this test signal under different categories. Finally, it completed the classification by determining the residuals of the test and the reconstructed signals. The experimental results showed that the proposed algorithm can achieve higher accuracy than the traditional support vector machine and Back-Propagation classification algorithms.