Buried Pipeline Third-Party Damage Signals Classification Based on LS-SVM

Abstract

To monitor third-party damage (TPD) activities on oil transmission pipeline such as man-made drilling, hammering and excavating on metallic pipe, acoustic method is proposed based on wavelet packet energy feature extraction and least square support vector machine (LS-SVM). To effectively detect and classify pipe TPD signals with small sampling, multi-class LS-SVM classifier algorithm and a novel feature extraction method is presented. Original TPD signal is divided into third level with wavelet transform, then approximation signal which covers main information of TPD signal is extracted to be decomposed into third level with wavelet packet decomposition. Wavelet packet energy is selected as feature to LS-SVMs. Feature extraction method reduces computation cost of on-line implement. When detection spacing is 600m, four TPD signals: normal, drilling, hammering and excavating conditions, classification success rate is more than 85%. The monitoring system can effectively detect and classify pipe acoustic TPD signal