Leakage aperture identification of natural gas pipeline based on compressed acquisition and DSAE

Abstract

Traditional gas pipeline leak detection methods usually face challenges in that there are redundant data collected on account of the time-domain sampling theory and the valuable leakage information is hidden in the complex vibration signals. This paper put forward an intelligent aperture identification method for natural gas pipeline leakage. This method integrated the compression of monitoring data based on compressed sensing (CS) transformation and automatic feature extraction and identification based on deep neural networks, namely, denoising sparse autoencoder (DSAE). The compressed acquisition can greatly reduce the volume to be processed and effectively extract the aperture information. DSAE combined the merits of sparse auto-encoder (SAE), denoising coding and dropout to achieve robust feature extraction and aperture classification. Experimental results validate the effectiveness and good performance, and the proposed method can realize faster and more accurate identification of different leak sizes compared with the traditional methods.