| Mineral resources are indispensable and important resources which support the country’s economy,people’s livelihood,security,and other aspects.In order to ensure the sustained and stable supply of mineral resources,it is necessary to strengthen the exploration of mineral resources.With the advancement of prospecting work,the demand for finding deep hidden minerals is becoming increasingly urgent.It is necessary to combine the development of cutting-edge information technology and use new algorithms in the field of artificial intelligence to realize the prediction of deep minerals.Based on the comprehensive analysis of geological,remote sensing,geophysical,geochemical and other geological information,the copper mine in the west of Dongchuan,Yunnan,Panzhihua and Liangshan of Sichuan is taken as the research area to establish the copper mine prospecting model.The quantitative prediction of mineral resources is studied by machine learning.The main research contents and progress achieved are as follows:(1)Combined with the idea of comprehensive information prediction of mineral resources,the geological characteristics such as strata,structure and resource distribution are analyzed and deeply interpreted.S-A fractal and other algorithms are used to fully interpret geological,remote sensing,geophysical,and geochemical data in the study area,analyze the characteristics of geological phenomena,and summarize their impact on the formation of copper deposits.(2)A comprehensive information metallogenic model for copper deposits in the study area is designed.By analyzing the geographical location,structural zoning,and ore-bearing strata of the 7 typical copper deposits in the region,the metallogenic conditions of the typical deposits are summarized,and a metallogenic model covering geotectonic location,metallogenic epoch,structure,geophysical characteristics and geochemical characteristics is constructed.(3)The fuzzy evidence weight,support vector machine,random forest and BP neural network are used to predict mineral resources.Firstly,14 evidence layers are constructed and fuzzy evidence weight is used to predict mineral resources.Secondly,according to the known mineral distribution,the research area is divided into 138651 statistical units,and a training sample set with equal amounts of positive and negative samples is constructed.Then,SVM,BP neural networks,and random forests are used for prediction.(4)4 kinds of prediction results are evaluated comprehensively,and the prediction result of fuzzy evidence weight is determined to be the best.The prediction results of the 4 algorithms are divided into 9 potential areas,and the prediction accuracy,reliability and efficiency are compared from 3 aspects: the predicted number of deposits,the scale of deposit reserves,the ratio of the number of ore points and the area.The experiment proves that the fuzzy weight evidence has high prediction accuracy,reliable prediction results,and can predict the maximum known mineral points in the minimum area,with much higher prediction efficiency than other algorithms.It is suitable for the subsequent delineation of prospective areas and other work,and has the most practical application significance.(5)The favorable prospecting target area was delineated based on the prediction results of fuzzy evidence weight,with a total of 10 Class I prospective areas,3 Class II prospective areas,and 8 Class III prospective areas delineated.The metallogenic conditions,known deposits and prospecting potential of each target area are evaluated comprehensively,which points out the work direction of copper prospecting in the study area. |
