【作者】 刘金海；

【导师】 张化光；

【作者基本信息】 东北大学 ， 控制理论与控制工程， 2009， 博士

【摘要】 随着世界范围内石油资源的紧缺,石油运输过程中的安全问题日益引起人们的重视。由于管道老化、打孔盗油等原因,输油管道经常发生泄漏。为了及时发现泄漏,基于负压波、声波、光纤等原理的实时泄漏检测系统被设计和应用。其中基于负压力波信号分析的输油管道泄漏故障诊断方法是目前我国输油管道泄漏检测所用的主要方法。在基于负压波的泄漏诊断理论中,输油管道压力时间序列(Oil Pipeline Press Time Series, OPPTS)的波动一直以来都被认作是随机的,导致微小泄漏无法检测,阻碍了基于动态压力信号的输油管道泄漏故障诊断系统的发展。本文主要以OPPTS作为研究对象,研究了其内部动态特性,探讨了提高基于负压波的泄漏诊断技术整体性能的方法,主要工作如下：1对OPPTS实测数据集依次利用硬件滤波、滑动均值滤波和小波滤波方法进行顺序滤波处理,为后续的数据分析和应用打下基础。在利用小波方法滤波过程中,通过对实测数据和模拟数据的滤波仿真,对比分析了离散小波、提升小波和非抽样小波处理数据的特点,并且得出利用非抽样小波处理OPPTS能够较好的保存管道内部的动态特性。2提出了改进的Darbyshire-Broomhead Lyapunov指数谱算法,通过互信息函数确定嵌入延迟,利用伪近邻法(False Nearest Neighbors, FNN)方法确定最佳嵌入维,从而确定了Lyapunov指数的个数,克服了原算法排除可疑指数过程中容易引起的指数个数不确定的缺点。3为了进一步提高基于压力波的泄漏检测方法的性能,本文首次分析出管道内的油品的压力波动具有混沌特性,为基于压力波的泄漏检测研究提供了一条新的研究途径。根据管道内具有湍流的特点,利用Haykin实测数据分析理论对管道内部流体的特性进行了理论分析,求解和分析了数据集的分形维数,计算Lyapunov指数的计算方法并且对目标数据集的Lyapunov指数进行了估计和分析,评估了数据集的平稳性和非线性,根据结果得到了输油管道压力波动具有混沌特性。4利用OPPTS的混沌特性,给出了一种基于BP和RBF神经网络OPPTS在线故障诊断方法。该方法利用OPPTS的非线性特性重构相空间,以重构向量作为神经网络模型的输入,在线训练神经网络模型,实现网络模型权值在线调整,从而实现实时对故障信号的检测。分别利用BP网络和RBF网络作为该算法的网络模型,利用OPPTS中的故障数据进行了仿真验证。5分析了视神经网络的特点和参数之间的关系,利用OPPTS的混沌特性给出了OPPTS故障诊断的视神经网络方法。该方法具有需要先验数据少,处理速度快等优点。分别比较了BP、RBF和视神经网络在处理故障诊断问题上的优缺点,并给出了每种方法的使用场合。6在对OPPTS理论分析和应用研究的基础上,设计了输油管道故障诊断专家系统。利用本文提出的故障算法作为故障定位方法,再利用模糊推理对检测到的压力故障数据进行分类,从而到达检测泄漏并且定位的目的,通过验证,系统性能具有较大提高。更多还原

【Abstract】 Oil pipelines often leak because of aging, steal etc, so it is an important problem to detect the leak points on oil transmission pipelines. Many methods based on negative press wave, sound wave and optical fiber are used to find leak of oil pipelines, of which negative press wave is the common method in china. By far fluctuation of oil pipeline press time series (OPPTS) is assumed stochastic or white noise, which hinders the improvement of fault diagnosis system that designed based on pressure fluctuation. In this dissertation, the internal dynamic of pressure fluctuation in oil pipelines is reseached, based on which several new leak-detection motheds is proposed, the main work is as follow:1 RC-filter, mean-filter and wavelet-filter are used in turn to minish the effect of noise before studying internal dynamic of OPPTS. Filting of disperse wavelet, lifting wavelet and undecimated wavelet is discussed and is simulated separately, the result is the undecimated wavelet is the most fitting method.2 An improved method of computing the Lyapunov-spectrum based on Darbyshire-Broomhead’s arithmetic is given, in which the delay is computed based on mutual-information function, the optional embedding is obtained based on false nearest neighbors (FNN). The improved method can get the number of Lyapunov exponents and can evaluate Lyapunov-spectrum of OPPTS including noise. This method can avoid the step of removing spurious exponents.3 To improve the performance of leak detection system based on OPPTS, the possibility of existing chaotic characters is validated using the method of nonlinear analysis. Experimental data sets of OPPTS are studied on which phase spaces are reconstructed, fractal dimensions and Lyapunov exponents are computed, the stationarity and nonlinearity are validated. By the analysis of the results, the rigorous chaotic characters in OPPTS are found, that is a theoretical basis for the correlative investigation based on pressure time series of oil pipelines.4 An on-line fault-detection method of BP and RBF based on chaos for OPPTS is proposed. After reconstructing phase-space by OPPTS and using vectors of reconstruction as the input of neutral network model, this proposed method can real-time modify the parameters of model by training the model on-line. Then the default and even little fault can be diagnosed by the error between real-time data and the predicted value of neural network model. Experiential datasets of OPPTS are simulated by the proposed method.5 Another on-line fault-detection method of retina neural network based on chaos for OPPTS is proposed and the model of retina neural network is given also. This mothod has the excellences of needing little data and faster speed. Then the comparation and the situation of the methods based on BP, RBF and retina neural network is analysed.6 A leak diagnosis expert system of oil pipeline is desgined, which classes the fault data of OPPTS by fuzzy reasoning and finds fault point of OPPTS by on-line fault-detection method. The expert system has the better result than the old one by simulation,.更多还原