Transformer-based deep learning models for predicting permeability of
porous media
Abstract
The direct acquisition of the permeability of porous media by digital
images helps to enhance our understanding of and facilitate research
into the problem of subsurface flow. A complex pore space makes the
numerical simulation methods used to calculate the permeability quite
time-consuming. Deep learning models represented by three-dimensional
convolutional neural networks (3D CNNs), as a promising approach to
improving efficiency, have made significant advances concerning
predicting the permeability of porous media. However, 3D CNNs only have
the ability to represent the local information of 3D images, and they
cannot consider the spatial correlation between 2D slices, a significant
factor in the reconstruction of porous media. This study combines a 2D
CNN and a self-attention mechanism to propose a novel CNN-Transformer
hybrid neural network that can make full use of the 2D slice sequences
of porous media to accurately predict their permeability. In addition,
we added physical information to the slice sequences and built a
PhyCNN-Transformer model to reflect the impact of physical properties on
permeability prediction. In terms of dataset preparation, we used the
publicly available DeePore porous media dataset with the labeled
permeability calculated by pore network modelling (PNM). We compared the
two transformer-based models with a 3D CNN in terms of parameter number,
training efficiency, prediction performance, and generalization, and the
results showed significant improvement. Combined with the transfer
learning method, we demonstrate the superior generalization ability of
the transformer-based models to unfamiliar samples with small sample
sizes.