Digital Core Reconstruction Method Based on Markov Chain Monte Carlo Algorithm and Generative Adversarial Network

We propose a digital core reconstruction method that leverages both the Markov chain Monte Carlo algorithm and Generative Adversarial Network (GAN) model. Traditional 2D to 3D methods often suffer from poor connectivity, resulting in unreasonable skeleton and small pores. To address these issues, we present a 3D reconstruction model trained on real 2D core data, which uses deep convolution and GAN to learn the planar distribution characteristics of 2D images. We employ the Markov chain Monte Carlo algorithm to sample and enhance the continuity and variation between neighboring reconstructed layers, leading to a more logically structured three-dimensional pore network. We validate our model using both isotropic and anisotropic porous media, obtaining excellent agreement with actual samples in terms of porosity, two-point correlation function, Minkowski functionals, and visual representation. Compared to using only the Markov chain Monte Carlo algorithm, our model yields better results, reducing the generation of isolated skeletons and un-reasonable small pores. Furthermore, our model requires less training sets and less training time on porous media, while maintaining similar generating effects, making it a more cost-effective option.