Understanding pore structure in tight shale formations is critical for both unconventional hydrocarbon production and subsurface geotechnical applications, such as CO2 sequestration and wellbore stability assessment. This study introduces experimental methods, such as X-ray diffraction, coupled with a hybrid TRANSFORMER-CNN-LSTM that integrates various well log measurements, including Gamma Ray (GR), Spontaneous Potential (SP), and Acoustic logs (AC), providing model-agnostic interpretability via Shapley Additive Explanations (SHAP) to relate features to pore classcification, using Mercury Injection Capillary Pressure (MICP) data. The model captures complex relationships among different reservoir properties, thereby enhancing the understanding of pore distribution with distinct pore systems (Type IV: 93%, Type I/VI: 91%, Type I/II: 84%), while transitional pore types (Type II, III, II/III) exhibit lower separability due to inherent geological continuity. Hybrid models, such as TRANSFORMER-CNN, TRANSFORMER-LSTM, and TRANSFORMER-CNN-LSTM, achieve higher prediction accuracy than the traditional methods, when using the Adam with Weight Decay (AdamW) optimizer. The TRANSFORMER-CNN-LSTM model achieved an accuracy of 95%, with Precision, Recall, and F1 scores of 0.92, 0.89, and 0.95, respectively. SHAP analysis identifies Gamma Ray and Thorium as dominant controls, particularly for clay-rich and Type III systems, while Acoustic and Neutron logs are related to more stable microporous structures (Type I/II). The findings indicate that hybrid models demonstrate superior performance in distinguishing the micropores and mesopores, offering insights into lithology-dependent pore systems that are critical to engineering geological evaluations. Results demonstrate applicability for characterizing heterogeneities in low-permeability formations, informing fracture mechanics, wellbore stability, and geotechnical risk assessment.
Abelly et al. (2026) studied this question.