Pores and Flows in Shales: A Statistical, Numerical and Deep Learning Study
摘要
Shales play a critical role in the geological storage of energy waste, serving as caprocks for gas storage and being considered as potential hosts for high-level nuclear waste due to their low permeability. This paper focuses on pore topology in shales and the ensuing fluid flow behavior. We analyzed 240 scanning electron microscopy images of shales from diverse locations worldwide, using a deep learning-based segmentation method for fast and precise 2D pore characterization, achieving a segmentation accuracy of 97.82%. These results were extended to 3D pore structure reconstruction and flow simulation. Results show that pore sizes in shales follow a log-normal distribution, with the ratio of standard deviation to mean pore size converging to approximately 0.5 across scales. Flow simulations conducted on these reconstructed pore networks revealed hydraulic conductivity values ranging from 10–17 to 10–5 cm/s, highlighting the significant role of shale pore size and connectivity. The relationships between hydraulic conductivity, and mean pore size and porosity highlight the significance of mean pore size, pore type, organic matter content, and specific surface area in controlling fluid flow in fine- and coarse-grained geomaterials.
Highlights Develops an efficient deep learning technique to segment shale scanning electron microscope images. Reveals the pores in different porous media follows a log-normal distribution. Finds the ratio of the standard deviation to mean of pore diameter is constant at 0.5. Investigates the relationship between hydraulic conductivity and mean pore size in shales. Explores the critical role of organic matter in controlling fluid flow in shales.