地质矿物的地球化学数据可视化分析与构造背景预测  

Visualization Analysis and Structural Background Prediction of Geochemical Data of Geological Minerals

作　　者：陈伟雄 杨华健 周泽东 张明 CHEN Wei-xiong;YANG Hua-jian;ZHOU Ze-dong;ZHANG Ming(College of Civil Engineering,Shaoguan University,Shaoguan 512005,China)

机构地区：[1]韶关学院土木工程学院

出　　处：《世界有色金属》2019年第14期132-134,共3页World Nonferrous Metals

摘　　要：本文采用目前世界上使用广泛的GEOCOR数据库对地质矿物进行数据挖掘。首先利用Python和Matlab等工具对全球矿物的地球化学数据集(41360件)进行预处理(数据清洗),然后对特征数据集进行空间分布、聚类分析和社区发现的可视化分析,最后利用二次清洗得到的建模数据,采用ELM和SVM两种机器学习方法对地质矿物构造背景进行智能预测判别,得出以下重要结论:(1)通过对地质矿物的28个属性进行K-Means聚类得到的雷达图,结合学术界的研究得出,地质矿物的聚类可视化效果比较明显。(2)通过社区发现算法来挖掘新的构造背景与原有的七个构造背景间的潜在联系,每一个构造背景为一个节点,都可以进行系统地探索。运用逆向推导的思维可以推断出地球化学元素与地质矿物原有的七种构造背景间存在的特定联系,如类别3的矿物的主量元素MgO、微量元素Cr、Ni与类别6的主量元素K2O、微量元素Th、Nb、Rb在原有大陆溢泥质矿物构造背景上可以明显区分于其它构造背景。(3)本文采用大数据思维,在建模数据有限的情况下,利用支持向量机对地质矿物的六种构造背景进行预测判别,识别准确度高达91.7%。充分说明利用支持向量机对矿物的构造背景进行预测判别是可行的。This paper uses GEOCOR database which is widely used in the world to mine geological minerals.Firstly,the global mineral geochemical data sets(41360 pieces)are pre-processed(data cleaning)using Python and Matlab tools.Then the feature data sets are visualized by spatial distribution,clustering analysis and community discovery.Finally,the modeling data obtained by secondary cleaning are used for ELM and SVM machine learning.Methods Intelligent prediction and discrimination of geological and mineral tectonic background were carried out,and the following important conclusions were drawn:(1)Through K-Means clustering radar maps of 28 attributes of geological minerals,combined with academic research,it is concluded that the visualization effect of geological minerals clustering is obvious.(2)Through the community discovery algorithm,the potential relationship between the new tectonic background and the original seven tectonic backgrounds can be excavated.Each tectonic background is a node,which can be explored systematically.The specific relationship between geochemical elements and the original seven tectonic backgrounds of geological minerals can be deduced by using reverse deduction.For example,the major elements MgO,Cr,Ni and K2O,Th,Nb and Rb of minerals in Category 3 can be clearly identified in the tectonic backgrounds of the original continental overflowing argillaceous minerals.It is clearly distinguished from other tectonic settings.(3)In this paper,we use large data thinking and support vector machine to predict and discriminate the six tectonic backgrounds of geological minerals under the condition of limited modeling data.The recognition accuracy is as high as 91.7%.It fully shows that it is feasible to use support vector machine to predict and discriminate the structural background of minerals.

关 键 词：构造背景 地质矿物 数据挖掘 数据可视化 

分 类 号：X143[环境科学与工程—环境科学]