【作者】 何鑫；

【导师】 张化光；

【作者基本信息】 东北大学 ， 控制理论与控制工程， 2011， 硕士

【摘要】 随着管道运输业的不断发展,输油管道运输的安全运行成为了管道安全监测的一项极为重要任务。由于不可避免的腐蚀老化和人为破坏因素,管道泄漏频繁发生,因此,针对管道泄漏检测与定位技术的研究具有重要的理论意义和实际应用价值。当前,在实际管道泄漏检测领域大多使用单一方法,应用最广泛的是负压波检测法。该方法具有定位原理简单,计算量小,灵敏度高,可迅速检查并报警等特点。但是,该方法定位精度不高,对噪声同样敏感,并且极易对例如泵站的正常工作等工况产生误报。针对负压波检测法的上述不足,本文提出了一种基于负压波和声波协同检测的管道泄漏定位系统及定位方法,通过声波法和负压波法两种检测方法的协同工作,提高了整个泄漏检测系统的定位精度；利用线性拟合角度变化规律分析压力下降点的有效性,可以在一定程度上克服噪声干扰,提高泄漏报警准确性,并且此法还具有辅助定位的功能；针对压力下降产生的原因不明的问题,采用模糊最小-最大神经网络对压力下降进行分类决策。通过利用以上理论,丰富了原有负压波检测法功能,克服其上述缺点,取得了良好的效果。本文进行了以下几个方面的工作：首先,在掌握了负压波法基本原理的基础上,提出了一种基于负压波和声波协同检测的管道泄漏定位系统及其定位方法。详细的介绍了系统的组成结构,定位原理和工作流程,为后续的研究提供目标和对象。然后,针对系统可能出现的频繁误报或漏报的问题,提出了利用线性拟合角度变化方法进行压力下降分析。详细的讨论了压力下降和上升时的角度变化过程,抓住了角度在[90,270]范围内,拐点位置附近的拟合角度大小满足“小-大-小”的变化规律,并利用该规律,合理的选择参数,实现对压力下降的有效分析。接着,研究三种经典的模糊最小-最大分类神经网络：常规模糊最小-最大神经网络,基于补偿神经元的模糊最小-最大神经网络和通用模糊最小-最大神经网络。在总结了三种分类网络的结构,特点的基础上,提出了一种基于数据质心的模糊最小-最大分类神经网络。通过IRIS数据集证明了以上方法有效性。最后,针对如何区分工况和泄漏这个问题,结合模糊最小-最大分类神经网络进行分类决策的特点,将上述的四种网络应用到管道泄漏检测系统中。利用线性拟合角度变化提取特征训练网络,从精度和耗时两个方面进行研究。针对本文运用的方法,利用MATLAB软件进行了大量的仿真实验,通过使用现场的实际压力数据,保证该方法的实验结果具有高度的可信性和实际的应用性。仿真结果充分证明了以上方法的可行性,有效性和处理泄漏检测与定位问题的强大能力。更多还原

【Abstract】 Oil pipeline operation in good condition is crucial for its fault diagnosis with the development of the pipeline transportation. Because of inevitable corrosion and man-made sabotage, the leakage of crude oil pipelines frequently happened. As a result, the research on fault diagnosis method of pipeline leakage takes position of theoretical and practical significance.But now, we always use the method based on only one theory in the field of pipelines’leakage detection. The negative pressure wave technology, which possesses straightforward principles, simple calculation, high sensitivity and outstanding immediacy, is widely used. However, this technology has little higher accuracy. Meanwhile, it is highly sensitive to the noise. What is more, it can not distinguish the right state between leakage and working condition rest on the pressure decline. Therefore, the thesis presents the method and the system based on the team working between the negative pressure wave and sound wave. The team working is built to raise accuracy of finding leakage point. Moreover, the angle change, which is inflicted by fitting lines, to some extent, can clear up the effects of noise and decrease error alarms. Fuzzy min-max neural network puts pressure features into use to represent the pattern classes, which include leakage pattern and working condition. The main research work is described as follows:Firstly, the method and the system based on team working between the negative pressure wave and sound wave are put forward after we master the basic principle of negative pressure wave. The thesis introduces the structures, locating algorithm and working procedure of the system so as to supply the object for the latter research.Secondly, based on the angle change inflicted by the fitting line, a new method is proposed to decrease the frequency of error alarms. The thesis introduces the detailed process of the angle change when pressure is rising and dropping. The rule of the angle change is obeyed as long as the angle change belongs to [90,270]. According to this rule, we can easily achieve success on the analysis of the pressure dropping if choosing parameters reasonably. Thirdly, three classical fuzzy min-max neural networks are presented, which are FMNN, GFMN and FMCN. Based on the analysis of the parameters of fuzzy Min-Max neural network, a new classification method rest on centroid for real data pattern is proposed. The availability of the new method is validated by simulation of IRIS dataset.Finally, four kinds of fuzzy min-max neural network introduced above are applied to the filed of pipelines’leakage detection. The angle change inflicted by fitting lines is utilized to obtain training features and then we do some research on accuracy and time consuming.According to the method proposed in the thesis, lots of simulation experiments are carried out by MATLAB, using pressure data obtained from real pipelines. They can ensure the experiment results with high applicability and creditability. In a word, the fault diagnosis method of pipeline leakage based on linear fitting and fuzzy min-max neural network is extremely viable and effective.更多还原