Chaos characteristics and least squares support vector machines based online pipeline small leakages detection

• Published in: Chaos, solitons and fractals solitons and fractals, 2016-10, Vol.91, p.656-669
• Main Authors: Liu, Jinhai, Su, Hanguang, Ma, Yanjuan, Wang, Gang, Wang, Yuan, Zhang, Kun
• Format: Article
• Language: English
• Subjects: Chaos characteristics; Hybrid kernel; LS-SVM; Small leakages
• ISSN: 0960-0779; 1873-2887
• DOI: 10.1016/j.chaos.2016.09.002

Small leakages are severe threats to the long distance pipeline transportation. An online small leakage detection method based on chaos characteristics and Least Squares Support Vector Machines (LS-SVMs) is proposed in this paper. For the first time, the relationship between the chaos characteristics of pipeline inner pressures and the small leakages is investigated and applied in the pipeline detection method. Firstly, chaos in the pipeline inner pressure is found. Relevant chaos characteristics are estimated by the nonlinear time series analysis package (TISEAN). Then LS-SVM with a hybrid kernel is built and named as hybrid kernel LS-SVM (HKLS-SVM). It is applied to analyze the chaos characteristics and distinguish the negative pressure waves (NPWs) caused by small leaks. A new leak location method is also expounded. Finally, data of the chaotic Logistic-Map system is used in the simulation. A comparison between HKLS-SVM and other methods, in terms of the identification accuracy and computing efficiency, is made. The simulation result shows that HKLS-SVM gets the best performance and is effective in error analysis of chaotic systems. When real pipeline data is used in the test, the ultimate identification accuracy of HKLS-SVM reaches 97.38% and the position accuracy is 99.28%, indicating that the method proposed in this paper has good performance in detecting and locating small pipeline leaks.