| The study area is located in the southeast of Athabasca basin,Canada.It is rich in unconformity type uranium resources.The unconformity type uranium deposit is famous for its high grade and large reserves.The research degree of early deposit in the study area is high and the research foundation is good.Based on the collection and sorting of multi-source geological data in the study area,this paper summarizes the uranium metallogenic model,extracts the prospecting information,analyzes and processes the geochemical and geophysical information respectively,extracts the multi-source comprehensive information,uses the supervised learning theory and machine learning algorithm to study the prospecting and prediction of unconformity type uranium deposits,evaluates the prediction model,and selects the optimal prospecting and prediction model,According to the optimal model,the metallogenic prospect areas are further divided.The main achievements are as follows:（1）The discontinuity layer in the metallogenic process of unconformity type uranium deposit provides a flow channel for ore bearing fluid,and the unconformity provides a storage location for uranium deposit.Illite kaolinite chlorite silicification alteration produces minerals,and the graphite substrate drives convection through conduction of deep heat source to promote the centralized precipitation of uranium hydrothermal solution.（2）The correlation analysis of geochemical element content shows that u element has high correlation with Th,Ti O2,Ta,Nb,V,SC,Sn,Pb and W,which is closely related to uranium mineralization;The elements screened by ROC analysis are analyzed by principal component analysis,and the component data RC2 is obtained after inversion;U,Th,K,gravity and aeromagnetism are analyzed by principal component analysisΔT geophysics obtains the first principal component（PC1）.（3）Fault and comprehensive alteration are determined as the prediction variables of geological elements,and aeromagnetic parameters are selected through ROC curveΔT.Taking PC1,Pb,RC2,U,Sn and Mo as prediction variables,the importance of prediction elements is ranked according to the random forest method.The order of importance is:u,comprehensive alteration,aeromagnetism,PC1,Pb,RC2,fault,Sn and Mo.（4）Random forest,support vector machine and BP neural network model are used to predict the mineralization of unconformity surface type uranium deposits in Athabasca basin.The prediction and evaluation are carried out with confusion matrix and ROC curve.The effect of random forest model is the best,the accuracy and recall rate of confusion matrix and AUC value of ROC curve are the highest,and the area of favorable metallogenic area is the smallest.Based on the prediction results of random forest model,the metallogenic prospects are delineated,and three types and seven metallogenic prospects are obtained:three type I metallogenic prospects;2 class Ⅱ scenic spots;Two class Ⅲ scenic spots. |
