【作者】 张海峰；

【导师】 刘宏昭； 原大宁；

【作者基本信息】 西安理工大学 ， 机械设计及理论， 2007， 硕士

【摘要】 有杆抽油是人工举升采油的主要方式，掌握其系统效率变化趋势和动态特征参数是系统评价和优化的重要前提。本文基于有杆抽油系统效率监测的数据，提出了利用经验模式分解和神经网络等方法对系统效率变化情况及特征参数进行分析计算的方法。论文的主要工作如下：分析了有杆抽油系统的结构特点和效率监测的基本方法，详细描述了单井测试的主要内容(功图数据和电参数)以及两种不同测试仪器的数据存储格式。结合建立的不同工况油井的参数数据库开发了相应的数据分析软件，获得被测试油井的各项动态参数。针对一段时间内的系统效率值呈现不规则的波动情况，利用经验模式分解方法提取油井在被分析时间段内的变化趋势，并与小波方法分析的结果进行了对比，表明经验模式分解方法有效实用。针对经验模式分解方法存在的端点效应问题，论文进一步引入支持向量回归方法对原始数据进行双边延拓。数值仿真实例和对实际油井系统效率测试数据的分析均表明，支持向量回归方法可以有效地抑制端点效应，使提取出的数据趋势项更加精确。在分析现有的简化计算方法和数值模拟方法获取悬点示功图特征参数存在不足的基础上，提出了将简化计算方法和神经网络相结合获得悬点示功图特征参数的方法。该方法考虑了振动、摩擦、气体和供液能力等复杂因素的影响，避免了数值计算方法中某些参数难以确定的问题。分析比较了BP神经网络和RBF神经网络对不同参数选择的影响。计算结果与实测数据对比表明，本文方法是有效和可行的。更多还原

【Abstract】 Sucker rod pumping is a leading means of artificial lift. It is the important precondition of system evaluating and optimization to master its system efficiency trend and dynamic characteristic parameters. In this paper, the movement of system efficiency and its dynamic parameters are calculated by use of empirical mode decomposition and artificial neural network base on the monitoring data of the system. The main work is as follows:The structure trait and the normal technique of efficiency monitoring of sucker rod pumping units are explained, and the chief content of signal well test containing hanging-point data and electric parameters, as well as their file format, are discussed. The corresponding software with database is developed to calculate its dynamic parameters.In allusion to the problem that the system efficiency appears in an irregular fluctuation, the EMD method was introduced to analyze the monitoring signal of sucker rod pumping units, and the trend of system efficiency of oil well was extracted. Compared with the wavelet method, it can extract the trend more effectively.In order to solve the problem of end effects of EMD, the bilateral extension of system efficiency measurement data is carried out by means of the SVR method. The result of simulate and the real system efficiency monitoring data shows that the end effects of EMD can be overcome effectively and the system efficiency trend prediction accuracy of oil well can be increased by using the SVR method.After discussing the scantiness of the simplified calculate model and numerical value simulation, the method combined simplified calculate with artificial neural network is introduced to get the characteristic parameter of hanging-point card which considers the factors such as oscillation, friction, gas and capability of oil, avoiding to confirming some parameter in the model. The transformation of the BP neural network and RBF neural network is discussed when its parameters are changed. It is show that the neural network can predict the characteristic parameter of hanging-point card of the oil well in varies condition effectively through training, and it is effective and feasible.更多还原