Rapid calculation method of pore section hydraulic conductivity based on convolution neural network

Abstract

Structural differences in pore space are the direct factors affecting fluid movement in porous media, while the shape of the pore cross-section determines the process and state of fluid movement, which is closely related to the conductivity in fluid media. Considering the traditional engineering calculations such as pore network models and finite element methods use shape approximation to describe pore cross-sections, which lose part of the shape information. To address the above problems, we proposes a computational method based on convolution neural network to accurately describe the pore cross-section shape by extracting the cross-section shape features and correct the computational misalignment problem of the traditional method. In order to ensure the universality of the method to different rock types, we extract (3779) 2D pore cross-sections from the 3D X-ray images of Bethemier and Limestone samples as the sample set for the training of the convolution neural network model. Finally, the accuracy of the model prediction results and the efficiency comparison with the mainstream methods are given, proving that the method proposed in this paper outperforms other methods in terms of accuracy and efficiency. This work is of significance for oil and gas field exploitation.