Continuous inversion of rock mechanics parameters from well logs based on XGBoost and analysis of dominant factors
To address the issues of high cost, discrete data, and difficulty in continuous evaluation associated with traditional core experiments, this paper establishes SVM, Ridge, LightGBM, and XGBoost models using conventional logging curves such as natural gamma ray, density, P-wave transit time, S-wave transit time, neutron porosity, and resistivity as inputs to continuously predict static Young's modulus, static Poisson's ratio, and uniaxial compressive strength. R², RMSE, and MAE are used to evaluate model accuracy, and SHAP is combined to interpret the controlling factors. The results show that XGBoost has the best overall prediction performance, with sonic transit time, density, and natural gamma ray contributing significantly. This study can provide a theoretical basis for evaluating rock mechanical parameters in well sections with few or no core samples.