Research on Mineral Calculation Based on Element Yield and XGBoost Optimization Algorithm: A Case Study of Shale Oil Reservoir in Weixinan Depression

Elemental logging is an essential technique for calculating the mineral content of reservoirs. However, due to the complex and diverse types of formation minerals and the lack of manual interpretation experience, it is challenging to calculate the mineral content accurately. Constructing a mineral content calculation method based on machine learning is of great significance for enhancing the accuracy and efficiency of interpretation. This research is based on the element yield data. Among the five methods evaluated, the XGBoost algorithm demonstrated superior performance and was further optimised through low‐confidence data cleaning and the incorporation of rock physics constraints. The calculated values of the optimised XGBoost model are highly consistent with the experimental values. Except for trace minerals like calcite and dolomite, all other minerals exhibit correlation coefficients above 70%, with quartz and clay reaching 87.09% and 81.96%, respectively. The average absolute error between the experimental value and the calculated value is within 20%. The research results provide meaningful guidance for selecting and optimising machine learning methods for mineral content calculation, which will further enhance the evaluation accuracy of reservoir “sweet spots”.