A case study on homogeneous and heterogeneous reservoir porous media reconstruction by using generative adversarial networks. (February 2019)

• Record Type: Journal Article
• Authors: Liu, Siyan; Zhong, Zhi; Takbiri-Borujeni, Ali; Kazemi, Mohammad; Fu, Qinwen; Yang, Yuhao
• Is Part Of: Energy procedia. Volume 158(2019)
• Page Start: 6164
• Page End: 6169
• Publication Date: 2019-02
• Subjects: porous media -- image reconstruction -- generative adversarial networks -- neural networks; Power resources -- Congresses; Power resources -- Periodicals; Power resources; Conference proceedings; Periodicals; 333.7905
• DOI: 10.1016/j.egypro.2019.01.493
• Languages: English
• ISSNs: 1876-6102

Abstract

Abstract: The advancing of modern X-ray computer tomography technology provides a powerful tool for us to illustrate the details inside the reservoir rock in three-dimensional space. Pore-scale rock characterization, modeling, and related fluid flow simulation can be challenging due to the high complexity of various rock samples. Conventional pore scale structure modeling methods such as various stochastic methods were developed for reservoir rock 3D microscopic structure reconstruction in order to generate representative realizations for numerical simulations and property upscaling approaches. In this work, generative adversarial networks (GANs) is used for generating the synthetic micro representations of porous rock by acquiring non-linear statistical information from the real 3D rock images in an unsupervised learning scheme. The related 3D image pre-processing, network training and adjusting as well as data post-processing procedures are addressed. The network prediction results from a homogeneous Berea sandstone and a heterogeneous Estaillades carbonate demonstrated the capability of GANs for high-resolution porous rock image representations reconstruction, generated and real images are compared via various visualizations and inspections. The study also illustrated the importance of the training image preprocessing, which indicating the data augmentation techniques can be one of the promising improvements in terms of capturing the sparsely distributed features from heterogenous 3D images and reconstructing the synthetic realizations, meanwhile, the robustness of the model during training process is enhanced when limited real data is available.