Application Of Random Forest Algorithm For Anomaly Identification Of Sandstone-type Uranium Deposits Based On Well Logging Data

In order to quickly and effectively obtain the distribution information of uranium anomalies and the basic data of lithology distribution in Sandstone-type uranium mining area.Based on the theory of Sandstone-type uranium anomalies and geophysical logging response characteristics of different lithologies,this paper uses the integrated machine learning method random forest algorithm with its ability of non-linear fitting and decision-making,takes the known uranium anomaly information and the characteristic information of lithologic logging as training samples,and constructs the uranium anomaly recognition model and lithologic classification model based on random forest algorithm for the study area.In this paper,the abnormal layer and mineralized layer are identified and extracted from the borehole logging data of a uranium mine in the southern end of Daqing placanticline in Songliao Basin,and the lithology is classified and identified.The recognition results of random forest algorithm and the characteristic information of known mineralized layer are verified and analyzed.In order to solve the problem of the influence of super parameters,we use the idea of traversal optimization to screen the super parameters of random forest algorithm,and use the combination of the optimal algorithm parameters to classify and identify.The results show that when there are 563 random forest decision trees and 3 decision tree nodes,the algorithm structure is the best and the recognition accuracy is the highest.The error verification accuracy of out of bag data of training samples is 93.39%,and the prediction accuracy of prediction samples is 93.31%.The prediction accuracy of random forest algorithm is better than that of support vector machine and BP neural network.The prediction accuracy of nine boreholes in the study area is 92.25% on average.According to the parameter importance analysis of random forest algorithm,the contribution of resistivity and natural potential to the lithologic classification of the study area is greater,and the importance of density and radioactive logging is lower.In uranium anomaly recognition,the error verification accuracy of out of bag of random forest algorithm is 97.63%,and after variable selection,it is increased to 97.97%.The prediction accuracy is 99.21%,and the application effect is good.The recognition results of mineralized layer and abnormal layer coincide with the characteristics of known mineralized layer.Under the condition of equal proportion data,the accuracy of out of bag of training data and prediction accuracy of the algorithm model are 93.96% and 94.8%.Compared with the traditional methods of uranium anomaly recognition,it is more close to the true shape of uranium anomaly distribution.The results show that the stochastic forest algorithm is a reliable new method in the application of lithology identification and uranium anomaly identification,and it has a good prospect in uranium exploration and related geological work.The main contents of this paper are as follows:(1)Based on the in-depth analysis of the regional geological situation and logging data in the study area,8 characteristic parameters,such as resistivity,natural gamma,density,acoustic time difference,quantitative gamma,are selected to analyze the logging response law of Sandstone-type uranium deposits and different lithology types.(2)A random forest algorithm model for borehole logging data in the study area is designed,and the lithology of borehole logging data in the study area is automatically classified,and the results are analyzed.(3)A random forest algorithm model is designed for the ore bearing target layer in the study area,and the machine recognition and extraction of uranium anomalies are carried out for the borehole logging data in the study area,and the results are analyzed.(4)simultaneous interpreting results of random forest algorithm are compared with traditional BP neural network and support vector machine classification prediction algorithm.The differences and adaptability of the three methods are analyzed.(5)applying this method to the actual work of the research area,we can improve the performance of the algorithm through variable selection,simultaneous interpreting with the traditional uranium exploration method,and achieved good results.