Quantitative Analysis of Reservoir Mineral Component Contents in Elemental Logging Based on Hyperparameter Optimization

As the basic units composing rocks, minerals and their quantitative analysis are crucial for geological research and resource exploration. Traditional laboratory-based quantitative analysis of rock mineral components requires offline processing after sampling, with high demands for sample preparation, and fails to continuously reflect stratigraphic changes. In contrast, elemental logging can continuously measure formation elemental contents while drilling and enable rapid on-site real-time analysis. However, accurately converting elemental contents into mineral component contents remains challenging. With the development of artificial intelligence, integrating elemental logging data with machine learning has become an important direction, but deficiencies still exist in aspects such as parameter selection, which affect model accuracy. This study proposes a light gradient boosting machine (LightGBM) model optimized by the hyperparameter optimization framework Optuna for mineral quantitative analysis. The model was trained using 5,196 logging data points from 18 wells in the study area, and RMS error, coefficient of determination, mean absolute error, and mean relative error were used as evaluation metrics. Two blind wells in the study area were selected for prediction verification. The results show that compared with traditional algorithms for converting elemental contents to mineral components, such as truncated singular value decomposition, the least-squares method, and linear programming, the LightGBM-Optuna model exhibits significant advantages in prediction accuracy (with errors reduced by more than 50%). In addition, its performance was superior to that of the basic LightGBM model without Optuna integration (with errors reduced by more than 15%). To verify the reliability and internal logic of the model output, this study simultaneously introduced the SHapley Additive exPlanations method for interpretability analysis. The results indicate that the model’s prediction patterns are highly consistent with objective geological cognition. In summary, the LightGBM-Optuna model proposed in this study provides a more accurate and interpretable new method for the field of mineral quantitative analysis, with important practical and theoretical value.