石油学报 ›› 2010, Vol. 31 ›› Issue (4): 659-663.DOI: 10.7623/syxb201004027

• 石油工程 • 上一篇    下一篇

基于独立分量分析和支持向量机的管道泄漏识别方法

王明达  张来斌  梁  伟  陈志刚   

  1. 中国石油大学油气安全工程技术研究中心  北京  102249
  • 收稿日期:2009-10-03 修回日期:2010-01-18 出版日期:2010-07-25 发布日期:2010-09-25
  • 作者简介:王明达,男,1984年2月生,2006年毕业于中国石油大学(华东),现为中国石油大学(北京)在读博士研究生,主要研究方向为油气管道与动力机组在线故障监测与安全评价。
  • 基金资助:

    国家高技术研究发展计划(863)项目(2008AA06Z209)和北京市教育委员会共建项目“天然气主管线泄漏诊断系统研究”联合资助。

Pipeline leakage detection method based on independent component analysis and support vector machine

WANG Mingda  ZHANG Laibin  LIANG Wei  CHEN Zhigang   

  • Received:2009-10-03 Revised:2010-01-18 Online:2010-07-25 Published:2010-09-25

摘要:

将独立分量分析(ICA)方法应用到管道泄漏检测压力信号的降噪中。分析了压力波信号源的独立性,探讨了运用ICA方法的可行性。利用现场采集的压力信号对该方法进行降噪效果检验。结果表明,该方法较好地保留了泄漏拐点信息且降噪效果明显。将支持向量机(SVM)方法应用到泄漏检测的识别过程中,首先采取微分进化算法优化了SVM模型参数,然后利用现场泄漏实验数据,验证了基于SVM方法的识别模型具有很好的识别准确率和泛化能力。在实验的基础上,对噪声源合并和泄漏信号源不可分问题作了初步讨论。

关键词: 管道泄漏检测, 信号降噪, 独立分量分析, 支持向量机, 模式识别

Abstract:

The independent component analysis (ICA) was introduced to the noise reduction process in pipeline leakage detection. The application feasibility of the ICA was researched by analyzing the signal source,then the noise reduction effect using the ICA was checked by the field pressure signal. The check results indicated that this method could maintain the preferable information at the leak inflection point and has the obvious noise reduction effectiveness. The support vector machine (SVM) was introduced to the recognition of pipeline leakage. The SVM model parameters were optimized by the use of the differential evolution algorithm. The favorable recognition precision ratio and the generalization ability of the SVM model were verified by the field leakage experiments. The noise source merge and indivisibility of the leakage signal source were discussed based on the experiments.

Key words: pipeline leakage detection, signal noise reduction, independent component analysis, support vector machine, model recognition