A Permeability Prediction Method for Tight Sandstone Reservoirs Based on XGBoost Algorithm

Abstract

Permeability is a crucial parameter that characterizes the fluid migration capacity of reservoirs and is essential for hydrocarbon migration during accumulation and subsequent exploration and development. For tight sandstone reservoirs, the diverse pore types, complex pore-throat structures, and strong heterogeneity make it difficult to establish a stable correlation between porosity and permeability, thereby hindering accurate permeability prediction with conventional methods. To address this issue, this paper proposes a permeability prediction method based on an XGBoost (eXtreme Gradient Boosting) algorithm optimized with Optuna (an automatic hyperparameter optimization framework). The model’s accuracy, inter-well generalization ability, and interpretability are comprehensively evaluated by combining well-group validation, blind well independent verification, and SHapley Additive exPlanations (SHAP) global-local interpretation. Six characteristic parameters—core porosity (CORE-POR), natural gamma ray (GR), acoustic transit time (AC), compensated density (DEN), compensated neutron log (CNL), and photoelectric absorption cross-section index (PE)—are selected as input variables to construct the permeability prediction model. The predictive performance of the model was evaluated using the coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The SHAP interpretation method was used to conduct global and local interpretability analysis of the model. The results show that Optuna-XGBoost outperforms multiple regression models and traditional machine-learning models on both the validation set and blind wells, indicating that this method is well-suited to permeability prediction for tight sandstone reservoirs in this study area.