InterPore2026

Integrated Characterization Methods for Shale Reservoir Heterogeneity and Oil Content: Lithofacies, Pore Network, and AI-Based Oil Content Evaluation

21 May 2026, 15:35

1h 30m

Poster Presentation (MS15) Machine Learning in Porous Media Poster

Speaker

Zaiquan Yang

Description

Exploration of lacustrine shale oil has emerged as a crucial frontier in global energy security, particularly within the Junggar Basin of China. The Permian Fengcheng Formation in the Mahu Sag, a world-class alkaline lacustrine shale oil reservoir, serves as a significant geological analog to the Eocene Green River Formation in the United States. However, the development of this resource is severely hindered by the lithofacies heterogeneity and complex pore structures characteristic of shale reservoirs. Traditional evaluation methods rely heavily on discrete core analyses (XRD and Thin-section ), which fail to capture the continuous vertical variations of lithofacies. Furthermore, the coupling mechanism between micropore heterogeneity and oil occurrence states (free vs. adsorbed) specifically how mineralogical composition and pore network geometry synergistically control oil mobility remains poorly understood. To address these challenges, this study establishes an innovative integrated characterization approach merging optimized ensemble regression models with multifractal theory. A novel Logistic-Bayes-IGWO-Bagging ensemble learning model was developed to predict lithofacies using standard logging data. Specifically, the architecture utilizes the Bagging algorithm to ensemble Back-Propagation (BP) neural networks, significantly reducing prediction variance. Crucially, the model employs Bayesian optimization to automatically tune network hyperparameters (e.g., hidden layers) and leverages an Improved Gray Wolf Optimizer (IGWO) to optimize weights and biases, preventing the model from converging on local optima. Additionally, oil-bearing capacity formulas for free and adsorbed oil within different pore sizes across various lithofacies were established to differentiate oil states. Finally, key parameters derived from multifractal dimensions were integrated with logging parameters to mathematically derive the heterogeneity and connectivity of macro-pore and micro-pore domains at the logging scale.The primary conclusions are as follows:(1) The study constructed a Bayes-IGWO optimized Bagging-BP ensemble learning model. By integrating elemental logging with XRD data, continuous lithofacies identification was achieved across the entire well section. The model achieved an R2 0.8287–0.8767 for mineral composition prediction on the test set, with the Root Mean Square Error (RMSE) maintained between 0.067 and 0.090, significantly enhancing vertical resolution. The optimized hidden layer nodes effectively captured gradational mineralogical features, improving lithofacies identification accuracy by approximately 30% compared to traditional discrete XRD sampling.(2) Mineral composition and pore structure synergistically regulate storage capacity, with carbonate content showing a positive correlation with oil content. In lithofacies where carbonate exceeds 40% (Calcareous feldspathic lithofacies), the peak free oil content reaches 2.61 mg/g, and adsorbed oil reaches 9.8 mg/g. Conversely, low-calcium lithofacies (Feldspathic lithofacies) exhibit free oil content of only 0.26-0.52 mg/g. TOC analysis indicates that high-calcium lithofacies have an average organic carbon content of 1.59%, where dissolution-induced porosity enhancement significantly expands hydrocarbon storage space.(3) Multifractal dimension analysis reveals that pore heterogeneity significantly impacts oil distribution. The generalized fractal spectrum parameter D0-D10 (the meso-micro pore) is negatively correlated with free oil (R2=0.86); free oil content increases by 35% when D0-D10 < 1.2. Lithofacies with a Hurst index (pore-throat connectivity) > 1.7 (Calcareous feldspathic lithofacies) show free oil concentrations of 0.01-0.025 cm3. While adsorbed oil is primarily concentrated in 10-100 nm pores, macro-pores (>1μm) dominate free oil migration.