【作者】 单彬；

【导师】 许少华；

【作者基本信息】 东北石油大学 ， 计算机软件与理论， 2012， 硕士

【摘要】 随着油田开发过程的持续进行，地下储层结构、储层物性、孔隙结构等都发生一定变化，这些变化使得油气开采变得困难。目前在油田地质建模研究中，三维静态模型从空间的角度对储层的分布和各种属性进行了定量的分析，但由于油田开发是一个随着时间不断变化的动态过程，三维模型在某种程度上不能满足油田的需求。因此，需要进行时空维数据建模技术的研究。本课题针对油藏动态开发问题，从三维空间储层注采过程模拟角度建立时空维分析模型。在空间三维静态模型的基础上，结合指标变量随时间的变化进行时空维模型研究。在建模中，考虑时间变量的连续性，需要模型解决多元过程信号输入输出关系，并有快速处理信号的能力。论文对时空维模型进行研究，通过研究时空维建模的理论方法和研究算法，建立一种能够处理多个时间片段的网络模型。论文分析了时空维数据模型特征，研究了可用于构建地质动态模型的算法，如涉及油藏知识的渗流微分方程，整体建模的径向动态回归模型，以及在机制上可直接对时空维系统进行建模的多聚合过程神经元网络。多聚合过程神经网络是把传统的神经网络扩展到时间域上，网络的输入、输出、连接权函数均可为依赖于时间变化的多元连续函数，网络具有良好的自适应、较高的容错能力和模型适应能力强等优点，可作为时空维系统建模的一种通用方法。本文针对时空维建模数据运算量大、信息变化机制复杂、训练时间长等问题，设计了分式多聚合神经网络和勒让德正交基变换多聚合神经网络两种网络模型。利用三维静态模型结合动态模型来对油藏储层进行可视化显示，通过对显示图像的分析处理可以为油田工作者提供直观的帮助，具有良好的应用前景。更多还原

【Abstract】 With the continuous development of oilfield exploitation, the underground reservoirstructure, reservoir property, and pore structure have occurred some changes, which make theoil and gas exploitation difficult. At present in the field of geological modeling study, athree-dimensional static model on the spatial perspective is used to make a quantitativeanalysis of reservoir distribution and various properties, however oilfield exploitation is adynamic process with time, the3D model can not meet the needs of oil field to some extent.Therefore, spatiotemporal data modeling technology research is necessary.In this paper, an analytical model of the space-time dimension on three-dimensionalreservoir injection and process simulation was established for dynamic reservoir developmentissues. We conduct research on indicator variables change with time based on threedimensional static model. Considering the continuity of the time variable, the model need tosolve the problem of multi-process signal input-output relationship and the ability of fastsignal processing on modeling. The paper do research on space-time dimensional model, anetwork model that can solve more than one time-slice problem is established by the theory ofspace-time dimensional modeling methods and algorithms. The algorithm can be used to buildthe geological dynamic model, involving the differential equations of seepage of the reservoirof knowledge, the overall modeling of the radial dynamic regression models, as well as aneural network that can be directly used on the space-time dimension system for modelingmulti-polymerization process neural network. Multi-polymerization process neural networkextends the traditional neural network to the time domain, the input and output connectionweights of the network function can depend on the time-varying multivariate continuousfunction, the network has the advantages of good adaption, high fault tolerance adaptability,and an the model can be used as a general method of space-time dimensional modeling.In this paper, we designed two network model of fractional multi-polymerization processneural networks and legendre orthogonal basis transform polymer neural network forspace-time dimensional modeling data computation, the complex mechanism of informationchanges, long training time,3D static model and dynamic model used to the reservoir visualdisplay, the analysis processing for display of the image to provide intuitive help for the oilfield workers, which has a good application prospects.更多还原