Statistical and neural network analysis of the relationship between the stochastic nature of pore connectivity and flow properties of heterogeneous rocks
Statistical and neural network analysis of the relationship between the stochastic nature of pore connectivity and flow properties of heterogeneous rocks
复制标题
DOI：
10.1016/j.jngse.2022.104719
复制
发表时间：
2022-09
影响因子：
--
通讯作者：
O. Ishola;Aaron Alexander;J. Vilcáez
中科院分区：
文献类型：
--
作者：
O. Ishola;Aaron Alexander;J. Vilcáez
我们使用了一个随机的三维孔隙尺度模拟方法来统计阐明随机孔隙连通性对渗透率和水力弯曲度的高度非均质多孔介质，如碳酸盐岩的影响。我们的工作流程的新颖性在于生成具有相同有效孔隙率、孔径分布、孔隙数量但不同随机孔隙连通性的多个3D孔隙微结构，其中唯一的孔隙微结构特征变化是孔隙连通性。该工作流程允许明确研究孔隙连通性在渗透率和水力弯曲度中的作用，而不受其他孔隙微观结构因素或噪声的干扰。上述特征的3D孔隙微结构的渗透率和水力弯曲度使用星星CCM+从直接孔隙尺度模拟获得。我们的方法抑制了进行数百次实验测量的必要性，并允许训练神经网络模型来预测渗透率和水力弯曲度。我们发现，一个近似的两倍增加的异质性（孔径标准差），结果在两个数量级的渗透率降低，并增加异质性的结果在系统转移的渗透率从正态分布对数正态分布。虽然孔隙的随机连通性对渗透率有显著影响，但它对水力弯曲度的影响很小。此外，渗透率从水力弯曲的可预测性随着非均质性的增加而降低。用PTSD数据沿着孔隙表面积参数训练的前馈神经网络（NN）模型在渗透率预测中获得的高决定系数表明NN算法可以捕获随机孔隙连通性对渗透率的影响。由于PTSD数据和表面参数可以从压汞毛细管压力（MICP）的测量，我们的研究结果有很大的影响，预测渗透率和水力弯曲度在高度非均质多孔介质。
We used a stochastic 3D pore-scale simulation approach to statistically elucidate the effect of stochastic pore connectivity on permeability and hydraulic tortuosity of highly heterogeneous porous media such as carbonate rocks. The novel nature of our workflow lies in the generation of multiple 3D pore microstructures of the same effective porosity, pore size distribution, number of pores, but different stochastic pore connectivity where the only pore microstructural feature changing is pore connectivity. This workflow allows the explicit study of the role pore connectivity plays in permeability and hydraulic tortuosity without the interference of other pore microstructural factors or noise. Permeability and hydraulic tortuosity of the 3D pore microstructures of the aforementioned characteristics was obtained from direct pore-scale simulations using STAR CCM+. Our approach suppresses the necessity of conducting hundreds of experimental measurements and allows the training of neural network models to predict permeability and hydraulic tortuosity. We show that an approximate twofold increase in heterogeneity (pore size standard deviation), results in a two orders of magnitude reduction in permeability, and that an increase in heterogeneity results in a systematic shift of permeability from normal distribution to lognormal distribution. While the stochastic connectivity of pores has a significant impact on permeability, it has only minimal effect on hydraulic tortuosity. Furthermore, the predictability of permeability from hydraulic tortuosity decreases with an increasing heterogeneity. The high coefficient of determination obtained in permeability prediction with a feedforward neural network (NN) model trained with of PTSD data along with pore surface area parameters indicates that NN algorithms can capture the effect of stochastic pore connectivity on permeability. Since PTSD data and surface parameters can be obtained from mercury injection capillary pressure (MICP) measurements, our findings have large implication toward the prediction of permeability and hydraulic tortuosity in highly heterogeneous porous media.