【作者】 郭巧占；

【导师】 夏克文；

【作者基本信息】 河北工业大学 ， 微电子学与固体电子学， 2006， 硕士

【摘要】 地层孔隙度是反映储集层油气储量的重要参数,准确度量这一参数对评价地层有着十分重要的意义。在声波测井中,为了精确确定地层声波孔隙度、进而为油气储量的评价与预测提供解释依据,研究优于传统理论方法的神经计算方法来确定地层声波孔隙度具有重要的应用价值。本文所作的主要工作如下:(1)详细分析了确定地层声波孔隙度的传统方法并指出其使用的局限性;阐述了声波测井时差与地层孔隙度之间是一个非线性的对应关系,由声波时差来确定岩层孔隙度属于非线性系统建模问题。由于神经计算技术非常适合于非线性系统建模,因此采用神经计算技术来确定地层孔隙度是一种切实可行的方法。(2)讨论了基于BP算法的神经网络、基于LM算法的神经网络、RBF神经网络和支持向量机等四种具体模型,进行了仿真实验,结果表明RBF神经网络和支持向量机具有很强的预测功能。(3)针对典型油井井段的测井资料进行了实际应用,主要包括样本集(含声波时差和岩芯孔隙度等参数)的选取与预处理,BP神经网络隐含层节点数的选取、RBF神经网络和支持向量机的模型参数的设计,以及待识地层声波孔隙度的预测等过程。应用结果表明,基于神经计算的地层孔隙度的求取方法远优于传统方法,其中RBF神经网络和支持向量机具有很强的预测能力,尤其是支持向量机的预测精度最高。更多还原

【Abstract】 Formation porosity is an important parameter in oil-gas storage of reservoir, and it is very significant to formation evaluation with exact measuring data of porosity. In order to determine formation acoustic porosity accurately and provide logging interpretation with the evaluation and prediction of oil-gas storage, the neural computation method superior to that of traditional is studied, and it has been of important application value to determine acoustic porosity in acoustic logging. The main works of this dissertation are as follows:(1) Traditional methods to determine acoustic porosity are analyzed in detail, and the limitation of traditional methods is pointed out. The nonlinear relation between slowness acquired from acoustic logging and formation porosity is described, and it is a modeling problem of nonlinear-system to determine formation porosity with acoustic slowness. For the neural computation technique is very adapted to nonlinear-system modeling, so it is feasible to determine formation porosity by used of neural computation.(2) Four idiographic models, such as the neural network based on BP algorithm, neural network based on LM algorithm, RBF neural network and support vector machines (SVM), are discussed. The simulation result shows the predictive ability of RBF neural network and SVM is strong.(3) The practical application is carried out by used of the logging data in typical oil well, the main steps includes, selection and pre-processing of the sample set( contains acoustic porosity, core porosity, etc), selection of hidden-layer nodes in BP neural network, design of model parameters in RBF neural network and SVM, prediction of acoustic porosity for un-recognized formation, and so on. The application result shows the neural computation method to determine formation porosity is superior to traditional methods, and the predictive ability of RBF neural network and SVM is strong, especially that the prediction accuracy of SVM is highest.更多还原