基于生成对抗网络的页岩三维数字岩芯构建

Reconstruction of 3D Shale Digital Rock Based on Generative Adversarial Network

杨永飞 刘夫贵 姚军 宋华军 王民

YANGYongfei LIUFugui YAOJun SONGHuajun WANGMin

深层油气重点实验室·中国石油大学(华东), 山东 青岛 266580 中国石油大学(华东)石油工程学院, 山东 青岛 266580 中国石油大学(华东)海洋与空间信息学院, 山东 青岛 266580 中国石油大学(华东)地球科学与技术学院, 山东 青岛 266580

Key Laboratory of Deep Oil & Gas, China University of Petroleum(East China), Qingdao, Shandong 266580, China School of Petroleum Engineering, China University of Petroleum(East China), Qingdao, Shandong 266580, China School of Oceanography and Space Informatics, China University of Petroleum(East China), Qingdao, Shandong 266580, China School of Geosciences, China University of Petroleum(East China), Qingdao, Shandong 266580, China

页岩油气藏孔隙结构复杂，岩芯获取困难，准确表征页岩储层孔隙结构是研究页岩储层内流体渗流规律的关键。基于真实页岩岩芯的三维聚焦离子束扫描图像，对原始生成对抗网络模型的结构重新设计，同时，为了保证重建结果可以充分反映页岩岩芯的孔隙结构信息，增大了训练样本的尺寸，以此训练生成模型，进而生成页岩三维数字岩芯，对比分析了重建数字岩芯和原始岩芯的孔隙度，并基于重建数字岩芯提取了孔隙网络模型，分析了页岩孔隙结构性质。结果显示，重建岩芯的孔隙度、孔隙空间结构、连通性以及孔隙喉道的配位关系与原始岩芯具有很高的一致性，由此验证了生成模型可以实现三维页岩数字岩芯的构建。最后，构建了多个页岩数字岩芯，计算了多个孔隙结构参数的均值及变化区间，证明了生成的数字岩芯具有稳定的孔隙空间特征，训练好的生成模型具有良好的稳定性。

The pore structure of shale oil reservoir is complex, and the shale cores are hard to acquire. Accurately characterizing the pore structure of shale reservoir is the key to the study on the fluid seepage law in shale reservoir. Based on the three-dimensional focused ion beam scanning (3D FIB SEM) images of real shale cores, the structure of the original generative adversarial network model is redesigned. At the same time, to ensure that the reconstruction results can fully reflect the pore structure information of the shale core, the size of the training sample is increased, and the model is trained to generate three-dimensional shale digital rock. The porosity of the reconstructed digital rock and the original core are compared, and the pore network model is extracted from the reconstructed digital rock, then the pore structure properties are analyzed. The porosity, pore and throat sizes, connectivity, and coordination relationship of the reconstructed digital rock are highly in agreement with the original cores, which verifies that the generative model can generate high-quality three-dimensional shale digital rock. Finally, several digital rocks are generated, and the mean value and variation range of various pore structure parameters are calculated. It is proved that the generated digital rocks have stable pore space characteristics, and the trained generative model has good stability.

10.11885/j.issn.1674-5086.2021.01.15.02