基于机器学习的黔北铝土矿石小体重回归模型研究OA

Small volumetric weight is an important parameter for estimating bauxite resource reserves and is of great significance for bauxite exploration.In northern Guizhou,the traditional measurement of small volumetric weight of bauxite ores involves sampling,processing and testing,which is often time-consuming and cumbersome due to the phased implementation of projects.Based on previously accumulated analytical test data of small volumetric weight of ore,Al2O3,SiO2,and Fe2O3,this paper proposes an interpretable regression model for small volumetric weight of bauxite ore based on the Random Forest(RF)algorithm.The Synthetic Minority Over-sampling Technique(SMOTE)can effectively address the issue of too few data in certain categories,enabling the model to better fit these data.The Shapley Additive Explanations(SHAP)method provides a unified measure for interpreting the RF regression model and enhances model interpretability.SHAP analysis indicates that SiO2 contributes the most to the small volumetric weight of bauxite ores.The Multiple Linear Regression(MLR)model achieves a Root Mean Square Error(RMSE)and Mean Absolute Error(MAE)of 0.1296 and 0.0974 on the test set,respectively,while the RF regression model achieves an RMSE and MAE of 0.0917 and 0.0672,respectively.The comparative results demonstrate that the RF regression model is more stable and performs better.Furthermore,the predicted average value(2.87 g/cm3)of the RF regression model on the validation set(the Xinmu-Yanxi deposit)differs from the adopted value of the deposit(2.82 g/cm3)by 0.05 g/cm3,with an error rate of only 1.8%,indicating high prediction accuracy.This suggests that the RF regression model can be used for estimating small volumetric weight of bauxite ores in bauxite exploration in northern Guizhou.