Bayesian-Optimized Weighted Ensemble Learning for Fluid-Type Identification in Carbonate Reservoirs

Carbonate reservoirs feature strong heterogeneity, significant anisotropy, and complex pore structures, making conventional logging insufficient to meet the fluid identification requirements for complex lithologic reservoirs. To address this issue, this study proposes a Bayesian-Optimized Weighted Ensemble (BO-WE) model, which combines three base models: Support Vector Machine (SVM), random forests (RF), and Light Gradient Boosting Machine (LGBM). Bayesian optimization (BO) is used to tune the hyperparameters of the base model and determine the optimal ensemble weights. The proposed method is applied to the study area and compared with the Hard Voting (HVT) and Stacking (STK) ensemble models. The results show that BO achieves a reasonable weight distribution of the base model in the ensemble model, and the BO-WE model predicts the independent test set through four indicators: accuracy, precision, recall, and F1-score. The four indicators are all greater than 91.65%, and the fluid-type is accurately predicted. This model provides an effective method for fluid identification in carbonate reservoirs during oil and gas exploration and development.