基于机器学习的闪锌矿微量元素特征在铅锌矿床类型识别中的应用  

Application of machine learning to predict types of Pb-Zn deposits by using trace elemental characteristics of sphalerite

作　　者：董赛娜 王达[1] 马国桃[2] 魏守才 代克刚 张晓暄 徐大兴 DONG Saina;WANG Da;MA Guotao;WEI Shoucai;DAI Kegang;ZHANG Xiaoxuan;Xu Daxing(School of Earth Sciences and Resources,China University of Geosciences,Beijing 100083,China;Chengdu Center,China Geological Survey(Geosciences Innovation Center of Southwest China),Chengdu 610081,China;The Second Geological Brigade,Bureau of Geology and Mineral Exploration and Development of Tibet Autonomous Region,Lhasa 850000,China;The Fifth Geological Brigade,Bureau of Geology and Mineral Exploration and Development of Tibet Autonomous RegionLhasa 850000,China)

机构地区：[1]中国地质大学地球科学与资源学院,北京100083 [2]中国地质调查局成都地质调查中心矿产资源室,成都610081 [3]西藏自治区地质矿产勘查开发局第二地质大队,拉萨850000 [4]西藏自治区地质矿产勘查开发局第五地质大队,拉萨850000

出　　处：《成都理工大学学报（自然科学版）》2024年第4期614-629,共16页Journal of Chengdu University of Technology: Science & Technology Edition

基　　金：北京市科学技术协会2023—2025青年人才托举工程(BYESS2023411);地质工程与矿产资源国家重点实验室开放研究课题(GPMR202308);中国地质大学(北京)新教师基本科研能力提升项目(2-9-2020-010)。

摘　　要：闪锌矿微量元素特征是识别铅锌矿床成因类型的重要指标。拟通过机器学习方法识别出判断铅锌矿床成因类型的关键控制元素,建立基于闪锌矿微量元素特征的广义铅锌矿成因类型判别图。系统收集了密西西比河谷型、火山块状硫化物型、喷流-沉积型、矽卡岩型4种成因类型铅锌矿床中3700条闪锌矿的12种微量元素数据(Cd,Mn,Ag,Cu,Pb,Sn,Ga,In,Sb,Co,Ge和Fe),使用支持向量机和随机森林2个机器学习分类模型对其进行分类,对这些特征元素重要性进行排序。基于闪锌矿微量元素特征,利用大数据和机器学习方法构建的铅锌矿床分类模型可以有效地区分不同成因类型的铅锌矿床,支持向量机和随机森林分类模型在测试集上的准确率分别为98.5%和96.9%。同时,通过主成分分析对12种元素特征进行统计分析和降维可视化,并结合随机森林模型特征元素重要性排序,识别出能区分铅锌矿床4种成因类型的关键化学元素。结果显示,闪锌矿的12种微量元素中,Mn,Ge,In,Co,Sb和Ga这6种元素用于区分铅锌矿床成因类型效果显著。新构建了4种成因类型铅锌矿床闪锌矿ln(Mn/Cd)-ln(Ge/Cd)-ln(Co/Cd)三元二维图,以及ln(Mn)-ln(Sb)、ln(Co)-ln(Ga)和ln(Mn)-ln(In/Ge)二元二维图,可用来有效区分密西西比河谷型、火山块状硫化物型、喷流-沉积型、矽卡岩型4种类型铅锌矿床。The trace elemental characteristics of sphalerite are crucial indicators for identifying the genetic types of Pb-Zn deposits.In this study,we identify the key control elements to distinguish between different genetic types of Pb-Zn deposits by using machine learning,and establish a generalized map of their identification based on the characteristics of their trace elements of sphalerite.To this end,we collected data on 12 trace elements(Cd,Mn,Ag,Cu,Pb,Sn,Ga,In,Sb,Co,Ge,and Fe;3700 samples)in sphalerite from four genetic types of Pb-Zn deposits:Mississippi Valley,Volcanogenic Massive Sulfide,Sedimentary Exhalative,and Skarn deposits.We applied two machine learning-based models of classification-the support vector machine and random forest-to classify the elemental data,rank the importance of these characteristic elements,and identify the key controlling elements to distinguish between the genetic types of Pb-Zn deposits.The use of big data in conjunction with the machine learning-based techniques enabled the accurate identification of the different genetic types of Pb-Zn deposits based on the characteristics of the trace elements of sphalerite.The support vector machine and random forest-based models of classification achieved accuracies of 98.5%and 96.9%,respectively,on the test dataset.Moreover,we subjected the 12-dimensional elemental features to principal component analysis.This,in conjunction with statistical analysis,visualized dimension reduction,and the ranking of importance of the elements according to their features based on the random forest model,enabled us to identify six key elements(Mn,Ge,In,Co,Sb,and Ga)that can be used to distinguish among the four genetic types of Pb-Zn deposits within the 12 trace elements of sphalerite.Following this,we constructed a 2D ternary diagram of ln(Mn/Cd)-ln(Ge/Cd)-ln(Co/Cd),and 2D binary diagrams of ln(Mn)-ln(Sb),ln(Co)-ln(Ga),and ln(Mn)-ln(In/Ge)for sphalerite to distinguish among the four genetic types of Pb-Zn deposits.

关 键 词：闪锌矿 微量元素 机器学习 大数据分析 铅锌矿床 

分 类 号：P618.13[天文地球—矿床学]