基于机器学习的主成分分析方法在金矿类型判别中的应用:以黄铁矿元素地球化学特征为例  

Application of principal component analysis method based on machine learning to gold deposit type discrimination:A case study of the geochemical characteristics of pyrite

作　　者：邓依 张静[1,2] 钟日晨[3] 何西恒 王潇逸 DENG Yi;ZHANG Jing;ZHONG RiChen;HE XiHeng;WANG XiaoYi(Frontiers Science Center for Deep-time Digital Earth,China University of Geosciences(Beijing),Beijing 100083,China;State Key Laboratory of Geological Processes and Mineral Resources,China University of Geosciences,Beijing 100083,China;School of Civil and Resource Engineering,University of Science and Technology Beijing,Beijing 100083,China)

机构地区：[1]中国地质大学(北京)深时数字地球前沿科学中心,北京100083 [2]中国地质大学,地质过程与矿产资源国家重点实验室,北京100083 [3]北京科技大学土木与资源工程学院,北京100083

出　　处：《岩石学报》2024年第6期1801-1816,共16页Acta Petrologica Sinica

基　　金：国家重点研发计划(2023YFC2906902);深时数字地球前沿科学中心项目(2652023001);高等学校学科创新引智计划(BP0719021)联合资助.

摘　　要：黄铁矿是金矿床中一种重要的载金矿物,其所含微量元素特征可以反映成矿热液性质、矿床类型等关键信息。机器学习方法能高效处理海量的黄铁矿微量元素数据并进行相关研究,前人利用随机森林、决策树、神经网络等不同算法,训练了相关的分类器,但是仍存在训练金矿类型偏少、使用元素种类偏少、矿床类型判别效果不佳等问题。因此本文采用机器学习中的主成分分析方法,拟训练并建立一个能直观反映大部分金矿类型不同特征的判别图解,进而评估使用黄铁矿化学元素作为不同矿床类型判别器的稳健性。数据集中共收集了来自卡林型、浅成低温热液型、造山型、斑岩型、IOCG、SEDEX和VMS七种矿床类型、近百个矿床的6939套黄铁矿LA-ICP-MS微量元素数据。统计结果显示,卡林型和IOCG型金矿床中黄铁矿的Au、As、Cu、Se等微量元素富集程度最高,造山型和浅成低温型的较富集,VMS和SEDEX型矿床中个别元素显著富集,而斑岩型金矿床中黄铁矿的各微量元素含量普遍较低。通过数据预处理,选择十种微量元素,绘制了不同成因类型金矿床黄铁矿成分的二维判别图解,并选择4个矿床实例进行了验证判别。两张判别图解联用能较好的区分出卡林型、斑岩型、IOCG、造山型和浅成低温热液型,但SEDEX和VMS型仍存在部分重叠,需要结合其他地质证据来综合判断。通过与传统判别图解和其他机器学习方法的判别效果对比,本研究构建的主成分分析判别图解具备简明直观、覆盖类型更广、突出强调矿床类型、对卡林型和斑岩型的判别效果最佳等优势,说明其在解决实际矿床问题方面具有有效性和准确度,为广大研究者提供了借鉴参考。Pyrite is an important auriferous mineral in gold deposits,and its trace element composition can reflect key information such as the property of mineralizing fluids and deposit types.Machine learning methods can efficiently process massive amounts of trace element data in pyrite and conduct related research.Previous studies have trained relevant classifiers using various algorithms such as random forest,decision tree,and neural network.However,there are still problems such as insufficient types of gold deposit for training,limited use of element types,and poor discrimination of deposit types.Therefore,this article adopted the principal component analysis method in machine learning to train and established a discrimination diagram that can intuitively reflect the distinguishing features of most gold deposit types,and then evaluated the robustness of using pyrite chemical elements as discriminators for different deposit types.The dataset comprises 6939 sets of pyrite LA-ICP-MS trace element data from seven deposit types:Carlin,epithermal,orogenic,porphyry,iron oxide-copper-gold(IOCG),sedimentary exhalative(SEDEX)and volcanogenic massive sulfide(VMS),encompassing nearly one hundred deposits.Statistical analysis revealed that pyrite in Carlin and IOCG type deposits exhibits the highest enrichment of trace elements such as Au,As,Cu,and Se,followed by orogenic and epithermal types.Individual elements are significantly enriched in VMS and SEDEX type deposits,while pyrite in porphyry type gold deposits generally has lower trace element contents.Through data preprocessing,ten trace elements were selected,and two-dimensional discrimination diagrams for the composition of pyrite in different genetic types of gold deposits were plotted,and validation discrimination was performed on four deposit examples.The combination of two discriminant diagrams can effectively distinguish the Carlin,porphyry,IOCG,orogenic and epithermal type deposits.However,there is still some overlap between the SEDEX and VMS types,and other geological

关 键 词：黄铁矿 LA-ICP-MS 金矿床 主成分分析 机器学习 

分 类 号：P578.292[天文地球—矿物学] P618.51[天文地球—地质学]