致密砂岩储层物性参数建模方法探讨

Discussion on the Physical Parameter Modeling Method for Tight Sandstone Reservoir

数理统计方法和地质理论结合已逐渐成为研究复杂储层的方向。由于致密砂岩储层的复杂性,其孔隙度、渗透率与测井资料不只是简单的线性关系,常用的线性回归方法难以满足致密砂岩储层物性研究的精度要求。选取致密砂岩储层研究区内3口取心井511个岩心分析样品,首先运用贝叶斯判别法将储层砂岩分为三类:石英砂岩、岩屑砂岩和岩屑石英砂岩,然后采用多元逐步回归、主成分分析和支持向量机方法对不同类型砂岩分别进行建模,对比各方法得出的复相关系数,发现支持向量机回归得到的复相关系数明显高于其他两种方法 ,且支持向量机回归方法得到的预测值与原始值的平均绝对误差也是最小的。支持向量机模型效果检验结果表明,孔隙度绝对误差小于1.5%,渗透率绝对误差小于0.25×10^(-3)μm^2,说明该模型预测效果较好,适用于研究区致密砂岩储层物性参数建模。

The combination of mathematical statistics method and geological theory has gradually become the direction of the complex reservoir study.For the complexity of the tight sandstone reservoir,the correlation between logging data and porosity and permeability is not the simple linearity.The accuracy of physical property of the tight sandstone reservoir is so low with linear regression method.Five hundred and eleven core samples were selected from three cored wells in the study area of the tight sandstone reservoir.The reservoir sand stones were classified to three types using Bayesian discriminating method：silicarenite,litharenite and quartzarenite.Then multiple stepwise regression analysis,principal component analysis and support vector machine were used to build the physical parameter models for different types of sandstones,and multiple correlation coefficient obtained by these methods was compared.It found that the multiple correlation coefficient obtained by the support vector machine is obviously higher than that by other methods,and mean absolute error between predictive value and initial value obtained by the support vector machine is the minimum.Testing result of the support vector machine model shows that absolute error of the porosity is less than 1.5% and that of the permeability is less than 0.25×10^-3μm^2.So the support vector machine is the best for prediction and suitable to physical parameters modeling in the study area of the tight sandstone reservoir.