Abstract One of significant ways for carbon dioxide sequestration and storage is to store carbon dioxide in underground geological structures. The development of digital reconstruction technology of shale has great potential of aiding in fine characterization of carbon dioxide sequestration for shale reservoirs, but the complicated structures of real shales make the accurate reconstruction very difficult. In recent years, generative adversarial network (GAN) has emerged as a valuable tool and proven to be successful for simulating digital rocks, owing to its robust feature learning and extraction capabilities. However, GAN faces challenges such as the requirements of a large number of training samples and a time-consuming reconstruction process. To overcome these challenges, this paper presents a novel approach using a multi-stage dual-branch discriminator GAN to stochastically reconstruct shales. This model design allows for the full extraction of structural features from a single training image and enhances the efficiency by sharing the common parameters between adjacent stages. Besides, the dual branches of the discriminator judge the internal content and scene layout of the reconstruction respectively, which ensures the reasonable quality of shale reconstructions. By comparing our method with several other reconstruction methods, it is demonstrated that our method has superior performance both in terms of feature extraction and reconstruction diversity.
Zhang et al. (Wed,) studied this question.