Abstract
The reconstruction of porous media is widely used in the study of fluid flows and engineering sciences. Some traditional reconstruction methods for porous media use the features extracted from real natural porous media and copy them to realize reconstructions. Currently, as one of the important branches of machine learning methods, the deep transfer learning (DTL) method has shown good performance in extracting features and transferring them to the predicted objects, which can be used for the reconstruction of porous media. Hence, a method for reconstructing porous media is presented by applying DTL to extract features from a training image (TI) of porous media to replace the process of scanning a TI for different patterns as in multiple-point statistical methods. The deep neural network is practically used to extract the complex features from the TI of porous media, and then, a reconstructed result can be obtained by transfer learning through copying these features. The proposed method was evaluated on shale and sandstone samples by comparing multiple-point connectivity functions, variogram curves, permeability, porosity, etc. The experimental results show that the proposed method is of high efficiency while preserving similar features with the target image, shortening reconstruction time, and reducing the burdens on CPU.
Highlights
The reconstruction of porous media plays a key role in many engineering disciplines
This paper proposes a reconstruction method of porous media based on deep transfer learning (DTL), which uses deep learning to learn the features of porous media and copies the learned features into the reconstructed results by transfer learning
Log ypi re Statistical methods represented by Multiple-point statistics (MPS) and some other methods are widely used for the reconstruction of highresolution 3D porous media
Summary
The reconstruction of porous media plays a key role in many engineering disciplines. A model of porous media can be used to quantitatively study the effects of various microscopic factors (e.g., pore structures, wettability, and aqueous films) on the macroscopic properties of oil and gas reservoirs, showing its great significance for the study of the seepage mechanisms of oil and gas [1,2,3]. Deep learning is an algorithm that extracts complex features by performing multiple nonlinear transformations on data through multiple layers and neural networks [18]. This paper proposes a reconstruction method of porous media based on DTL, which uses deep learning to learn the features of porous media and copies the learned features into the reconstructed results by transfer learning. The method designs a deep learning model to learn the features of TIs and to reconstruct porous media by transferring the features learned from TIs through a deep neural network (DNN). The reconstructed results have similar features (e.g., pore structures, connectivity, and permeability) hidden in the TIs. There are two important tasks in the proposed method. Once the parameters for the reconstruction can be determined, they are stored after training and can be quickly used for new reconstructions, displaying the effectiveness of the proposed method in reconstruction quality, speed, and memory demands
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
