Analysis Of Ore-forming Information From High-Dimensional Data

Both scale and dimension of geological data have been expanding.There comes the need of methods to deal with high-dimensional data,such as PCA(principal component analysis),a method to extract features form high-dimensional data,and SVM(support vector machine),a method to classify finite high-dimensional data samples.These two methods have already been tested in geoscience research.In this article,those two methods get combined and used to classify high-dimensional igneous rock data,before which there come standardization and centered log-ratio transformation to definite sum effect of component data.Principal features of data transformed get extracted by PCA and then used in SVM to generate classification hyper plane,which provides prediction for unlabeled samples with confidence probability as basic information for ore forming research.To achieve the purpose above,the application GeoPyTool has been developed.It firstly extends traditional routines,such as TAS and Harker diagram,with contour map and SVM classification.That makes the classic methods capable of measuring similarity of igneous rock quantitatively with corresponding data dimensions.Standardization and centered log-ratio transformation are also used to eliminate the definite sum effect of component data,after which the transformed data get used in PCA and then generate classification boundary by SVM,and finally get used to predict unclassified data.Using these methods on Erdaohe polymetallic deposit as an example,this paper has proved the existence of Tamulangou formation in Erdaohe deposit,and concluded that Manketou Obo formation is the volcanic rock related to mineralization.Log transformation and standardization are also used to show the contours and patterns of in-situ composition data of pyrite.Besides,two new methods to compute distance between high-dimensional data are introduced as a measurement of the correlation between ore forming metal elements.The patterns and the distance results show that the mineralization order of metal elements is Cu>Pb/Ag>Zn,and reveal the existence of multiple hydrothermal fluid interactions containing different ore forming metal elements.Finally,combining the results of high-dimensional data analysis above,this article concludes that both skarn ore forming process and hydrothermal filling effect happened in Erdaohe deposit,as two causes of the ore forming.