基于音波数据驱动的输气管道泄漏检测

发布时间：2018-03-06 23:00

本文选题：泄漏检测　切入点：音波　出处：《中国石油大学(华东)》2014年硕士论文　论文类型：学位论文

【摘要】：随着国内外经济的快速发展,天然气的需求量越来越大,因而陆上输气管道的铺设越来越多,管道的运输任务也越来越重。然而,由于管道腐蚀、自然破坏、人为破坏等诸多原因,管道泄漏时有发生。管道泄漏不但会造成环境污染、带来经济损失,严重时更会导致人员伤亡等事故。因此为保障管道正常安全运行,管道泄漏检测就变得尤为重要。本文用的音波法泄漏检测是近十几年发展成熟的一种方法,具有灵敏度高、定位准确等优点。本文依托实验室内的高压输气管道装置设计了基于音波数据驱动的泄漏检测方案,完成了实验室音波数据的采集;采用高质量时频分辨效果的广义S变换对泄漏音波信号进行时频分析与去噪处理;分析并提取了泄漏及干扰信号信号包括时域及小波域上的部分特征,为泄漏判断提供了依据;对泄漏及干扰信号常用统计特征进行了基于遗传算法的特征选取,得到了最有利于工况分类的特征组合,最后建立了基于支持向量机(SVM)的多类分类器,完成了泄漏及干扰信号的分类识别,降低了泄漏检测的误报率。通过进行基于音波数据驱动的泄漏检测试验研究,本文得出的主要结论有:①泄漏音波信号具有明显区别于管道正常运行时信号的时域及频域的特征,可以设定某个统计特征的阈值对泄漏进行判断;②针对泄漏信号高频多噪音、低频泄漏特征明显的特点,用高质量时频分辨效果的广义S变换进行时频分析及时频去噪能够取得很好的结果;③并不是所有的统计特征都适合用于信号分类,基于遗传算法的特征选取能够选取到较为合适的适用于特征分类的最佳特征子集;④对泄漏及干扰信号进行基于支持向量机的分类识别,能够提高泄漏信号的识别率,从而降低整个泄漏检测的误报率。
[Abstract]:With the rapid development of economy at home and abroad, the demand for natural gas is increasing, so more and more onshore gas pipelines are laid, and the transportation tasks of pipelines become heavier and heavier. However, due to pipeline corrosion, natural damage occurs. Pipeline leaks occur from time to time because of man-made damage. Pipeline leaks not only cause environmental pollution, bring economic losses, but also lead to accidents such as casualties when serious. Therefore, in order to ensure the normal and safe operation of pipelines, The sound wave method used in this paper is a mature method developed in recent ten years and has high sensitivity. Based on the high pressure gas pipeline device in the laboratory, the leakage detection scheme based on sonic data driving is designed in this paper, and the sound wave data is collected in the laboratory. The generalized S transform with high quality time-frequency resolution is used to analyze and Denoise the leaky sonic signal, and some characteristics of the leakage and interference signal, including time domain and wavelet domain, are analyzed and extracted, which provides the basis for leak judgment. The commonly used statistical features of leakage and interference signals are selected based on genetic algorithm, and the most favorable feature combination is obtained. Finally, a multi-class classifier based on support vector machine (SVM) is established. The classification and identification of leakage and interference signals are completed, and the false alarm rate of leak detection is reduced. The main conclusion of this paper is that the 1: 1 leaky acoustic signal has the characteristic of time domain and frequency domain which is obviously different from the signal when the pipeline is in normal operation. The threshold of some statistical characteristic can be set to judge the leakage. The characteristic of low frequency leakage is obvious. Using the generalized S transform of high quality time-frequency resolution to carry out time-frequency analysis can get good results. Not all the statistical features are suitable for signal classification. The feature selection based on genetic algorithm can select the best feature subset which is suitable for feature classification to classify leakage and interference signals based on support vector machine (SVM), which can improve the recognition rate of leakage signals. In order to reduce the false alarm rate of the whole leak detection.
【学位授予单位】：中国石油大学(华东)
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TE973.6

【相似文献】