摘要
本研究对指示样品中的元素组合进行描述。结果表明:在伊朗西北地区Glojeh多金属矿中,Ag,As,Pb,Te,Mo,Au和Zn发生强烈矿化,而S,W和Cu矿化程度较低。本文采用后向消元法通过初始化,降阶和模型验证对Au浓度的主效应(X)和二次项(X^2)以及Ag,Cu,Pb和Zn的一阶交互作用(X_i×X_j)进行定量预测。后向消元法是基于二次多项式模型完成的,通过去除不重要的指示变量进行消元而得到简化二次多项式模型。在二次多项式优化过程中,R^2(pred)增加而R^2减小,R^2,R^2(adj)和R^2(pred)都具有明显的收敛趋势。基于288个沟槽和679个钻孔岩石样品的预测结果表明:简化二次多项式模型包含阈值变量(Cu,Ag×Cu,Pb×Zn和Ag^2–Pb^2)和主指示变量(Pb,Ag×Cu,Cu×Zn,Pb×Zn和Ag^2)。由于Au矿化具有强烈的遗传效应,Pb,Ag^2和Ag×Pb为沟槽样品简化二次多项式模型的重要指示变量,而阈值变量为钻孔样品模型的重要指示变量。验证组沟槽样品和钻孔样品简化二次多项式模型的R^2(pred)分别为74.9%和60.62%,R^2分别为73.9%和60.9%。
The correspondence analysis will describe elemental association accompanying an indicator samples.This analysis indicates strong mineralization of Ag,As,Pb,Te,Mo,Au,Zn and to a lesser extent S,W,Cu at Glojeh polymetallic mineralization,NW Iran.This work proposes a backward elimination approach(BEA)that quantitatively predicts the Au concentration from main effects(X),quadratic terms(X2)and the first order interaction(Xi×Xj)of Ag,Cu,Pb,and Zn by initialization,order reduction and validation of model.BEA is done based on the quadratic model(QM),and it was eliminated to reduced quadratic model(RQM)by removing insignificant predictors.During the QM optimization process,overall convergence trend of R2,R2(adj)and R2(pred)is obvious,corresponding to increase in the R2(pred)and decrease of R2.The RQM consisted of(threshold value,Cu,Ag×Cu,Pb×Zn,and Ag2-Pb2)and(Pb,Ag×Cu,Ag×Pb,Cu×Zn,Pb×Zn,and Ag2)as main predictors of optimized model according to288and679litho-samples in trenches and boreholes,respectively.Due to the strong genetic effects with Au mineralization,Pb,Ag2,and Ag×Pb are important predictors in boreholes RQM,while the threshold value is known as an important predictor in the trenches model.The RQMs R2(pred)equal74.90%and60.62%which are verified by R2equal to73.9%and60.9%in the trenches and boreholes validation group,respectively.
基金
support of the IMIDRO(Iranian Mines and Mining Industries Development & Renovation Organization) for our research
关键词
相关性分析
一阶交互作用
简化二次多项式模型
优化模型
降阶和验证
强烈遗传效应
correspondence analysis
first order interaction
reduced quadratic model (RQM)
optimized model
order reduction and validation
strong genetic effects