世界核地质科学2025，Vol.42Issue(5)：1106-1122,17.DOI:10.3969/j.issn.1672-0636.2025.05.017
利用可解释机器学习揭示复杂砂岩型铀矿的地球化学控制因素
Unraveling the geochemical controls of a complex sandstone-type uranium deposit using explainable machine learning:A case study from Tamusu deposit,Bayingebi basin
摘要
Abstract
The genetic mechanisms of sandstone-type uranium deposits are complex and diverse.Traditional geochemical analysis methods face challenges such as difficulties in capturing high-dimensional geochemical relationships and weak statistical foundations when dealing with high-dimensional,nonlinear data,making it hard to fully reveal the key ore-forming controlling factors.The Tamusu uranium deposit,located in the Bayingebi basin,presents an ideal case for data-driven research due to its unique characteristics,such as being hosted in"hard"sandstone,a high degree of diagenesis.This study,for the first time,applies the Random Forest machine learning algorithm to the major and trace element geochemical dataset of the Tamusu uranium deposit.We constructed a regression model to predict the enrichment of uranium(U)and employed advanced explainability techniques,such as Permutation Importance and Shapley Additive Explanations(SHAP)(to quantify the contribution of each feature to the model's prediction),aiming to identify the key elemental assemblages controlling uranium mineralization and to quantitatively analyze their geochemical significance.The optimized Random Forest model achieved a predictive performance of RMSE=0.988 8,R2=0.644 8 on the test set,demonstrating that the nonlinear relationship between geochemical data and uranium enrichment is learnable.Feature importance analysis consistently identified lead(Pb)and rhenium(Re)as the two most robust and significant indicator elements for predicting uranium mineralization.SHAP analysis further revealed that Pb,as the final product of radioactive decay,serves as the most direct chemical signature of long-term uranium enrichment.Meanwhile,high concentrations of Re and molybdenum(Mo)strongly indicate a reducing environment favorable for uranium precipitation.Additionally,elements such as cadmium(Cd)and lithium(Li)also showed a positive contribution to mineralization,associated with co-precipitated sulfides and intense water-rock interactions involving high-salinity fluids,respectively.The"Pb-Re-Mo"core geochemical fingerprint identified in this study supports a multistage,composite genetic model for the Tamusu uranium deposit,which was influenced by the superposition of multiple geological processes,including redox reactions,deep involvement of high-salinity basinal brines,and possible hydrothermal activities.The analytical framework established in this study provide a methodological reference for similar investigations.关键词
塔木素铀矿床/可解释性/机器学习/随机森林/地球化学Key words
Tamusu deposit/explainability/Machine Learning/Random Forest/geochemistry分类
天文与地球科学引用本文复制引用
史清平,刘武生..利用可解释机器学习揭示复杂砂岩型铀矿的地球化学控制因素[J].世界核地质科学,2025,42(5):1106-1122,17.基金项目
核技术研发项目(编号:地H2401)资助 Supported by Nuclear Technology R&D Program(No.地H2401) （编号:地H2401）
