地学前缘 ›› 2021, Vol. 28 ›› Issue (3): 87-96.DOI: 10.13745/j.esf.sf.2021.1.10

• 数学地质与矿产定量勘查 • 上一篇    下一篇

磁铁矿元素地球化学大数据构建及其在矿床成因分类中的应用

洪双(), 左仁广*(), 胡浩, 熊义辉, 王子烨   

  1. 中国地质大学(武汉) 地质过程与矿产资源国家重点实验室, 湖北 武汉 430074
  • 收稿日期:2021-01-11 修回日期:2021-01-20 出版日期:2021-05-20 发布日期:2021-05-23
  • 通讯作者: 左仁广
  • 作者简介:洪 双(1996—),女,硕士研究生,主要研究方向为地质找矿大数据挖掘。E-mail: 1201911206@cug.edu.cn
  • 基金资助:
    国家优秀青年科学基金项目(41522206)

Magnetite geochemical big data: Dataset construction and application in genetic classification of ore deposits

HONG Shuang(), ZUO Renguang*(), HU Hao, XIONG Yihui, WANG Ziye   

  1. State Key Laboratory of Geological Processes and Mineral Resources, China University of Geosciences(Wuhan), Wuhan 430074, China
  • Received:2021-01-11 Revised:2021-01-20 Online:2021-05-20 Published:2021-05-23
  • Contact: ZUO Renguang

摘要:

磁铁矿广泛分布在岩浆、热液及沉积等各类矿床中,其地球化学元素组成往往受温度、氧逸度等物理化学条件的影响,能反映矿床形成环境并指示矿床成因类型,是一种重要的勘查指示矿物。自20世纪60年代以来,磁铁矿的主微量元素数据被用来构建不同的判别图,试图来区分矿床的成因类型。然而,由于矿床成因类型的多样性以及同一类型矿床的磁铁矿的主微量元素地球化学组成的复杂性,以往基于少数磁铁矿的主微量元素地球化学成分构建的矿床成因类型判别图存在一定的局限性。基于此,本文收集了前人发表在国内外期刊上的主要矿床类型的磁铁矿的元素地球化学数据(7 388条),初步构建了基于电子探针(EPMA)和激光剥蚀-电感耦合等离子体质谱(LA-ICP-MS)磁铁矿元素地球化学大数据集,建立了基于随机森林算法的矿床成因分类模型,并对磁铁矿主微量元素在矿床成因分类中的重要性做出排序。研究结果表明,基于磁铁矿大数据和机器学习算法构建的判别模型,能有效区分主要矿床类型,整体分类准确度高达95%。由于LA-ICP-MS磁铁矿数据集的测试元素多,分析精度高,使得基于LA-ICP-MS磁铁矿数据集的矿床成因分类模型精度高于基于EPMA数据集,表明磁铁矿中元素种类多少和数据测试精度影响矿床成因分类精度。同时,研究发现V元素在矿床成因分类过程中起到了较为重要的作用。此外,基于大数据和机器学习建立的判别模型对新的磁铁矿数据进行测试,可给出该数据属于每种矿床类型的概率,能有效判别矿床成因类型。

关键词: 磁铁矿, 元素地球化学数据, 随机森林, 矿床成因类型

Abstract:

Magnetite is an oxide mineral commonly found in magmatic, hydrothermal and sedimentary deposits. Its geochemical elemental composition is largely dependent on temperature, oxygen fugacity and other physicochemical conditions, and can reveal the ore-forming environment and indicate the genetic type of ore deposits. The major and trace elements in magnetite have been used for genetic classification of ore deposits since the 1960s. However, due to genetic diversity of ore deposits and complexity of geochemical composition of magnetite from the same type of ore deposits, the applicability of magnetite discrimination diagrams is often limited based on limited magnetite geochemical data. In this study, we collected from various publications a large amount of magnetite geochemical data (n=7388) determined by electron probe microanalysis (EPMA) and laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) to construct, preliminarily, two magnetite geochemical big data sets, and subsequently established a new genetic classification model based on random forest algorithm, and explored the importance of trace elements in the genetic classification of ore deposits. The results show that magnetite big data mining based on a machine learning algorithm can effectively distinguish the main types of ore deposit, with an overall classification accuracy up to 95%. Because the LA-ICP-MS magnetite data set contains high quality data on many trace elements, the classification accuracy is higher based on LA-ICP-MS data than on EPMA data, indicating the classification accuracy of ore deposit is affected by the number of trace elements in magnetite and by the accuracy of data analysis. At the same time, we found element V plays an important role in the classification of ore deposits. In addition, analyzing new magnetite data using the new discrimination model can yield the probability of each ore type and effectively distinguish the genetic type of ore deposit.

Key words: magnetite, geochemical big data, random forest, genetic type of ore deposit

中图分类号: