基于机器学习的华南诸广山花岗岩体铀矿潜力评价  

Uranium Potential Evaluation of Zhuguangshan Granitic Pluton in South China Based on Machine Learning

作　　者：黄鑫怀 李增华 邓腾 刘志锋[1] 陈冠群 曾皓轩 郭世超 Huang Xinhuai;Li Zenghua;Deng Teng;Liu Zhifeng;Chen Guanqun;Zeng Haoxuan;Guo Shichao(Jiangxi Engineering Laboratory on Radioactive Geoscience and Big Data Technology,East China University of Technology,Nanchang 330013,China;School of Earth Sciences,East China University of Technology,Nanchang 330013,China;State Key Laboratory for Nuclear Resources and Environment,East China University of Technology,Nanchang 330013,China)

机构地区：[1]东华理工大学江西省放射性地学大数据技术工程实验室,江西南昌330013 [2]东华理工大学地球科学学院,江西南昌330013 [3]东华理工大学核资源与环境国家重点实验室,江西南昌330013

出　　处：《地球科学》2023年第12期4427-4440,共14页Earth Science

基　　金：东华理工大学江西省放射性地学大数据技术工程实验室开放基金(No.JELRGBDT202006)资助。

摘　　要：地学大数据和机器学习的结合,为矿床勘查提供了新的发展方向.华南广泛发育花岗岩体,是花岗岩型铀矿的重要产区,因此如何判断特定花岗岩体是否具有产铀矿的潜力,对于指导华南花岗岩型铀矿勘查具有重要意义.系统收集了前人已发表的华南花岗岩地球化学元素含量数据(不包括待评价的诸广山地区的九峰岩体、红山岩体和茶山岩体),共获得1711条数据.然后按照7∶3的比例划分为训练集和测试集,进而分别建立了随机森林(random forest,RF)算法和K近邻(K-nearest neighbor,KNN)算法分类模型,并对两种分类模型的精确度、召回率、ROC(receiver operating characteristic curve)曲线进行评价,选出泛化能力较好的模型,最后利用泛化能力较好的模型对诸广山地区九峰岩体、红山岩体和茶山岩体进行成矿潜力评价.结果表明,随机森林分类模型对测试集的分类精确度、预测结果可靠度均高于K近邻分类模型,随机森林分类模型对测试集上的数据分类精确度达到了93%,利用上述创建的随机森林分类模型对九峰、红山和茶山岩体进行预测.预测结果表明,红山岩体和茶山岩体含矿的概率较高,而九峰岩体含矿概率较低.该研究为进一步缩小地质找矿勘查范围提供了可靠的依据,并且该模型可以作为地质找矿工作者的辅助工具.The combination of geological data and machine learning provides a new direction for mineral exploration.The granitic pluton is widely developed in South China,which is an important producing area for granite-type uranium deposits.Therefore,whether the granitic pluton has the potential to produce uranium deposits is of great significance for guiding the exploration of granite-type uranium deposits in South China.In this paper,the geochemical data of granites in South China are systematically collected(excluding the Jiufeng,Hongshan and Chashan granite plutons to be evaluated in Zhuguangshan area)from previous published papers,and a total of 1711 data pieces are obtained.They are further divided into training set and test set according to the ratio of 7∶3.Then,the random forest(RF)algorithm and K-nearest neighbor(KNN)algorithm classification models were established respectively,and the accuracy,recall rate and ROC(receiver operating characteristic curve)curve of the two classification models were evaluated,and the models with good generalization ability were selected.Finally,the metallogenic potential of the Jiufeng pluton,Hongshan pluton and Chashan pluton in the Zhuguangshan area were evaluated using the models with good generalization ability.The results show that the classification accuracy and reliability of prediction results of random forest classification model are higher than those of K-nearest neighbor classification model,and the classification accuracy of the random forest classification model on the test set reached 93%.The random forest classification model created above was used to evaluate the metallogenetic potential of the Jiufeng,Hongshan and Chashan plutons.The prediction results show that the probability of metallogenetic potentiality in the Hongshan and Chashan plutons is high,whereas the probability in the Jiufeng pluton is low.This study provides a reliable basis for further geological prospecting,and the model can be used as an auxiliary tool for geological prospecting.

关 键 词：机器学习 随机森林算法 K近邻算法 花岗岩型铀矿 成矿潜力 岩石学 

分 类 号：P628[天文地球—地质矿产勘探] P595[天文地球—地质学]