Ensemble learning approach for accurate virtual borehole prediction in 3D geological modeling

The use of virtual drilling technology can effectively accelerate the establishment process of 3D geological models and improve grid structure and visual performance. However, most existing virtual drilling prediction techniques mainly rely on traditional interpolation methods, which not only increase computational overhead but also lack sufficient automation. To address this problem, this study introduces an innovative virtual borehole prediction technology combined with a machine learning stacking strategy. This technology integrates Random Forest (RF), XGBoost, CatBoost, and LightGBM algorithms as basic models and improves prediction accuracy through stacked generalization technology. This study adopted a method of adding zero-thickness layers to unify stratigraphic sequences. The 3D position information of boreholes is used as model input, and the bottom height of boreholes at stratigraphic boundaries is used as the prediction target. Through a hierarchical training method using 85[Formula: see text] of borehole data for training and 15[Formula: see text] for verification, a regression prediction model for stratigraphic boundaries is established. The model is able to predict detailed stratigraphic sequences containing information on stratigraphic type and thickness to accurately simulate virtual boreholes. Research results show that the integrated model has excellent prediction performance, achieves efficient automated prediction, and provides a new solution for virtual drilling prediction.