A comparative study of empirical, statistical and virtual analysis for the estimation of pore network permeability
A comparative study of empirical, statistical and virtual analysis for the estimation of pore network permeability
复制标题
DOI：
10.1016/j.jngse.2017.07.002
复制
发表时间：
2017-09
影响因子：
--
通讯作者：
A. Ismail;Qamar Yasin;Q. Du;A. A. Bhatti-A.
中科院分区：
文献类型：
--
作者：
A. Ismail;Qamar Yasin;Q. Du;A. A. Bhatti-A.
多孔介质的渗透率是分析储层渗流特性和表征储层特征的关键参数，是油气开采优化的基础。渗透率通常是通过实验确定的，如果没有实验数据，可以使用经验关联式来估计渗透率。近年来，人工神经网络（ANN）建模在解决复杂问题（如预测非均质地层的渗透率）方面得到了广泛的应用。与每种技术相关的不确定性程度要求更谨慎地使用任何其他技术。本研究的目的是估计地层渗透率使用三种技术，“经验关系”，“多元回归分析”和“虚拟测量”，显示出潜力，在实现我们的目标。以岩心渗透率为目标数据，验证了这些技术的有效性。本研究选取了巴基斯坦Sawan气田下Goru非均质地层的6口威尔斯井。可获得这些威尔斯井的测井数据和相应的渗透率值。结果表明，在不同地质条件下，Morris和Biggs经验关系式与威尔斯井的实测渗透率具有较好的拟合效果。五个测井响应（伽马射线（GR）、体积密度（RHOB）、声波测井（DT）、深电阻率测井（LLD）和中子孔隙度（NPHI））被用作ANN中的输入以预测所有威尔斯井的渗透率。为了确保神经网络技术的特性不是孤立事件，在所有可用的威尔斯井中重复相同的练习以预测渗透率。根据威尔斯井的测井响应进行多元回归分析，以获得电缆测井对渗透率的定义。将多元回归分析与神经网络渗透率预测相结合，提出了一种基于测井曲线的渗透率预测方法。结果表明，渗透率可以估计在精确和准确的方式通过集成的统计和虚拟技术，根据所研究的岩石间隔的地质条件。
Permeability of any porous medium is a key parameter to analyze the flow behavior and characterization of reservoir for the optimization of hydrocarbon production. Permeability is usually determined experimentally, and if no laboratory data is available, empirical correlations can be used to estimate it. In recent years, artificial neural network (ANN) modeling, have gained popularity in solving complex problems, such as prediction of permeability in heterogeneous formation. Degree of uncertainty associated with each technique requires more careful use of any other technique. The present study aims to estimate formation permeability by using three techniques, “Empirical Relations”, “Multivariate Regression Analysis” and “Virtual Measurements” that show potentials in achieving our goal. Core permeability is used as target data to test the validity of these techniques. For the purposes of this study, six wells from a heterogeneous Lower Goru formation from Sawan Gas Field, Pakistan are selected. Well log data and corresponding permeability values for these wells were available. The result shows that Morris and Biggs empirical relations provide acceptable results with measured permeability in different geological conditions of the wells. Five well log responses (gamma ray (GR), bulk density (RHOB), sonic log (DT), deep resistivity log (LLD) and neutron porosity (NPHI)), are used as inputs in the ANN to predict permeability in all wells. To ensure that the characteristic of neural network technique is not an isolated incident the same exercise is repeated in all available wells to predict permeability. Multivariate regression analysis is performed on the basis of well log response in wells to access definition of permeability in terms of wireline logs. Hybrid approach is developed in this paper by the integration of multivariate regression analysis and estimated permeability from neural network, which suggest a verifiable and accurate prediction of permeability from well logs. The results indicate that permeability can be estimated in precise and accurate manner by the integration of statistical and virtual techniques depending upon the geological conditions of the studied rock interval.