InterPore2026

Effects of physically meaningful pore structure parameters on shale anisotropy thermal conductivity and machine learning-based prediction

22 May 2026, 15:30

1h 30m

Poster Presentation (MS09) Pore-Scale Physics and Modeling Poster

Speaker

Mr Chi Xiong (Department of Energy and Power Engineering, Tsinghua University)

Description

The development of medium to low maturity shale oil resources plays a critical role in alleviating China’s energy supply constraints. In-situ thermal conversion is one of the most promising recovery technologies, in which the design and optimization of heating schemes strongly depend on the thermal conductivity of shale. However, shale exhibits complex pore structures characterized by pronounced heterogeneity and anisotropy, which may further evolve during heating, making the accurate determination of effective thermal conductivity (ETC) highly challenging. Establishing quantitative relationships between pore structure characteristics and ETC of shale is important for understanding and regulating heat transfer during in-situ conversion processes. In this study, three physically meaningful pore structure parameters are introduced to characterize shale pore morphology: pore shape anisotropy (SA), pore distribution heterogeneity (H), and pore distribution anisotropy (DA). Together with porosity (ε), these parameters constitute a parametric description framework for shale pore structures. Based on the quartet structure generation set (QSGS) combined with the lattice Boltzmann method (LBM), the effects of pore structure parameters on the ETC of shale parallel and perpendicular to bedding were systematically investigated over a porosity range of 0.05–0.2, corresponding to the porosity evolution range of shale. The results indicate that all three pore structure parameters correlate significantly with the anisotropic effective thermal conductivity (AETC) of shale. With increasing SA and DA, the thermal conductivity parallel to bedding (kx) increases, whereas the conductivity perpendicular to bedding (ky) decreases, leading to an enhanced thermal anisotropy ratio (TA = kx / ky). As H increases, the fluctuation ranges of kx and ky become broader, and the maximum TA is further amplified. At ε = 0.2, increasing SA from 1 to 2 causes kx to increase from 1.26 to 1.61 W·(m·K)-1, while ky decreases from 1.26 to 0.86 W·(m·K)-1, resulting in an increase of TA from 1 to approximately 1.9. Increasing DA from 0.4 to 1.6 leads to an increase of kx from 1.05 to 1.72 W·(m·K)-1 and a decrease of ky from 1.86 to 1.11 W·(m·K)-1. Moreover, increasing H from 0.2 to 1.6 expands the fluctuation ranges of kx and ky from 1.10–1.42 to 0.80–1.86 W·(m·K)-1, with the maximum TA increasing from 1.2 to 1.8. The results further reveal a pronounced coupling effect between porosity and pore structure parameters on shale AETC. Finally, three machine learning models are developed using ε, SA, DA, and H as input features to predict shale AETC. All models achieve high predictive accuracy (R2 > 0.93), with the random forest model performing best (R2 > 0.95). SHAP-based interpretability analysis indicates that when ε is lower than 0.1, the AETC of shale is primarily governed by porosity, while the combined influence of pore structure parameters accounts for approximately 50% of the effect of ε. In contrast, within ε = 0.15–0.2, adjusting SA, DA, and H can achieve comparable or even greater modulation of shale thermal conductivity than changing porosity alone. These findings provide theoretical support for the design and optimization of heating strategies in shale in-situ conversion processes.