基于多点地质统计学的数字岩心建模

Digital core modeling based on multiple-point statistics

常用的数字岩心建模方法有X射线CT扫描法、模拟退火法和过程法。X射线CT扫描法虽然能够建立准确的三维数字岩心模型,但是其价格贵、周期长;模拟退火法和过程法能够降低建模成本提高建模效率,不过模拟退火法不能有效地表征孔隙结构的长距离连通性,过程法不适用于模拟成岩作用复杂的岩石。为了解决以上问题,提出基于多点地质统计学重构数字岩心模型的方法。以密西根盆地Waverly组的Berea砂岩为例,从X射线CT扫描法构建的Berea砂岩模型中提取两个1503体素的代表体积元,令一个为训练图像,另一个为真实模型;把真实模型的两张切片作为条件数据,使用多点地质统计学法重构出三维数字岩心模型。通过以变差函数、孔隙结构参数和单相流渗流参数为评价指标,对重构模型的准确性及该方法的可行性进行研究。结果表明:重构模型与真实模型的孔喉几何形态和拓扑结构具有较高的一致性,证明了重构模型的准确性和多点地质统计学法建立数字岩心模型的可行性。

There are three commonly used digital core modeling methods： X-ray computed tomography scanning,simulated annealing and process-based method. Although the X-ray computed tomography scanning method can build an accurate 3 D digital core model,it is expensive and time-consuming. The simulated annealing and process-based method can reduce the cost and improve the modeling efficiency,but the simulated annealing method cannot effectively characterize the long-range connectivity of the pore structure,and the process-based method is not suitable for simulating the rocks that have undergone complex diageneses. In order to solve these problems,a method of reconstructing digital core model based on multiple-point statistics was proposed. Taking as an example the Berea sandstone of the Waverly group in the Michigan Basin,two representative volume elements with the size of 1503 voxels were extracted from different locations of the Berea sandstone. One was set for the training image,the other for the real model. Two orthogonal two-dimensional slices were chosen from the real model as the conditioning data. The 3 D digital rock models were reconstructed using multiple-point statistics. The variogram,pore structure parameters and percolation parameters of single phase flow experiment were applied as the evaluation indexes to validate the accuracy of the reconstructed models.The result shows that the reconstructed models agree well with the real model in the pore-throat geometry and topology features,which verifies the accuracy of the reconstructed models and feasibility of reconstructing digital core model using multiple-point statistics.