A geology-constrained hybrid stacking ensemble method using well logs for TOC prediction in continental shale reservoirs

Continental shale oil is widely distributed in many basins worldwide and constitutes an important potential incremental source of unconventional oil, making reservoir evaluation of such plays particularly significant. Total Organic Carbon (TOC) is a key parameter for evaluating the oil-bearing potential of continental shale oil reservoirs. This study proposes a hybrid stacking ensemble method based on conventional well logs. The method consists of XGBoost, RF, LS-SVR and IGRU. Lithology and reservoir-layer constraints, a continuous digital representation of lithology, and multiple overlapping well-log response features are incorporated as input features, while TP-CVOCA and PCA modules are used to extract additional time–frequency attributes and integrated petrophysical features. Heuristic optimization algorithms are used for feature selection and hyperparameter tuning, and RegMix is employed to achieve data augmentation. IGRU and TP-CVOCA modules are specifically designed to handle non-uniform time-step issues. Based on 2,374 TOC samples in the northern Songliao Basin, six comparative experiments are designed. The proposed method achieves R² values of 0.8304 for intra-well prediction and 0.7261 for cross-well prediction, with absolute improvements in R² of about 0.094 and 0.107 over the other methods in the comparative experiments, while both MSE and MAE are reduced.