煤层气产能数值模拟及预测方法研究

发布时间：2018-03-07 22:22

本文选题：煤层气产能　切入点：数值模拟　出处：《中国矿业大学》2015年博士论文　论文类型：学位论文

【摘要】：煤层气产能是衡量煤层气井优劣的重要指标,而煤层气井产能的高低直接影响着经济效益。那么如何准确预测煤层气产能,是煤层气田为了进行高效开发而急需要解决的关键问题。煤层气产出机理有别于传统油气藏,探索建立适合煤层气藏特点的数值模拟技术对煤层气开发具有重要意义,因此建立有效的煤层气产能数值模拟模型,对煤层气勘探开发有着重要的实际价值。煤岩体的复杂结构,特别是断层等结构体的影响,常规规则差分技术难以适应。有限体积法(Finite Volume Method简称FVM)具有适应复杂边界和非结构网格的优点。另外该方法积分守恒不管对于整个计算区域还是控制子区域都能得到满足,即使在粗网格条件下,也能够显示出准确的积分守恒,其计算结果比较精确;能避免有限差分(FDM)中常出现的数值振荡及数值弥散现象;积分网格划分比较灵活,故在复杂的边界条件上的积分网格划分也较为方便。FVM不仅具有有限元法(FEM)的精确性又保持了FDM的简单性。本文在综合前人大量研究成果的基础上,在煤层气产能数值模拟及预测方面主要完成了以下研究工作:(1)运用渗流力学、煤层气地质学等学科的理论知识,对煤层气的储集、运移和产出机理进行了研究,概化出了煤层气储层模拟的地质模型;以数值模拟方法为工具,建立了拟稳态条件下关于三维、气水两相、双孔隙、非平衡吸附的煤层气产能模拟模型。(2)引入计算流体力学中的FVM方法来求解煤层气运移控制方程,并且给出了相应的单元中心格式的FVM计算公式。煤层气运移的非线性方程的离散选取的是全隐式格式,并采用FLUENT软件对煤层气运移计算求解。(3)把煤层气产能数值模拟模型运用于沁水盆地实际工程中,运用该井的气、水排采资料,经过对气、水产量进行历史拟合,对影响该井煤层气产出的主要参数进行了校正,在此基础上对煤层气井产能变化趋势进行了预测,并对影响产出的主要参数进行了敏感性分析,获得了较好效果。(4)应用相空间重构理论和贝叶斯证据框架理论,提出了一种基于混沌时间序列和贝叶斯证据框架下最小二乘支持向量机(Least Square Support Vector Machine,LS-SVM)的煤层气产能预测的新方法。本模型解决了BP神经网络模型中存在的局部最优问题,比SVM模型速度更快,故更便于使用。由于SVM确定模型参数耗时较长,还较易产生过拟合现象,所以利用贝叶斯证据三层推断寻找到模型的最佳参数,实现预测输入变量的自适应选择,以减少预测模型建立过程中对经验的依赖,并提高模型的适应性。应用相空间重构理论研究了煤层产气量时间序列的混沌特性,运用贝叶斯证据框架下的LS-SVM方法对重构相空间后的时间序列进行预测,并与SVM预测方法、BP神经网络预测方法进行对比,实例验证表明,该方法具有计算速度快、拟合精度高、结构灵活、泛化能力强等优点。也为煤层气产能预测研究提供了一种新的思路。
[Abstract]:Coalbed methane production is an important index to measure the quality of coalbed gas, and coalbed gas productivity directly affects the economic benefits. So how to accurately predict coalbed methane production, is a key problem for the efficient development of CBM and needs to be solved. Different from the traditional oil and gas accumulation mechanism of coalbed methane production, explore the establishment of suitable numerical characteristics coalbed gas reservoir simulation technology has important significance for the development of coal-bed gas, so the establishment of simulation model of coalbed methane production value effectively, it has great practical value for the exploration and development of coalbed methane. Complex coal and rock mass structure, especially the effect of fault structure, the general rule of differential technology is difficult to adapt to the finite volume method. (Finite Volume Method referred to as FVM) has the advantage of boundary and unstructured grid. In addition, the method for the calculation of integral conservation regardless of region or control sub area The domain can be satisfied, even in the coarse grid condition, also can show the accuracy of flow, the result is very accurate; can avoid the finite difference (FDM) numerical oscillation and numerical dispersion phenomena often occur; integral mesh is more flexible, so the integral mesh in complex boundary conditions on the it is convenient to.FVM not only has the finite element method (FEM) accuracy while maintaining the simplicity of FDM. Based on the previous research results, the numerical simulation and prediction of coalbed methane production mainly completed the following research work: (1) using percolation mechanics theory knowledge of coalbed gas geology, the coalbed gas reservoir, migration and production mechanism of generalized geological model of coalbed gas reservoir simulation; by numerical simulation method, a quasi steady state conditions on the three-dimensional, gas and water two Phase, dual porosity, simulation model of non equilibrium adsorption capacity of coalbed gas. (2) the introduction of FVM method in computational fluid dynamics to solve the coalbed gas migration control equation, and gives the corresponding formula of unit center format FVM calculation. Coalbed methane migration and nonlinear equations of discrete selection is fully implicit format. And using FLUENT software to calculate CBM migration. (3) the numerical simulation of coalbed methane production model used in the Qinshui Basin in the actual project, the use of gas wells, water drainage data, after the amount of gas, aquatic history matching, the main parameters which influence the production of coalbed methane wells was corrected. On the basis of the prediction of coalbed gas productivity trends, and the main parameters influencing the output of sensitivity analysis, obtained better results. (4) reconstruction theory and Bias evidence framework theory of the phase space is put forward A chaotic time series and Bias evidence framework based on least squares support vector machine (Least Square Support Vector Machine, LS-SVM) a new approach to predicting the coalbed methane production. This model solves the problem of the local optimal BP neural network model, the SVM model is faster than the speed, so it is more convenient to use. The SVM model parameter determination long, also more prone to overfitting, so the use of Bias three layer evidence to find the best parameter inference model, realize the adaptive prediction of the selection of input variables, the dependence on experience to reduce the prediction model in the process, and improve the adaptability of the model. The application of the theory of phase space reconstruction of chaotic characteristics of coalbed gas production time sequence, time series based on phase space reconstruction method after LS-SVM Bias evidence under the framework of the forecast and prediction method and SVM neural BP Compared with the network prediction method, the example shows that the method has the advantages of fast computation speed, high fitting accuracy, flexible structure and strong generalization ability. It also provides a new idea for the prediction of coalbed methane productivity.

【学位授予单位】：中国矿业大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TE37

【相似文献】