Research On Anomaly Identification Methods Of Gold Mineralization In Wulaga Area Of Heilongjiang Province Based On Ensemble Learning Algorithms

The exploration geochemical method is one of the most effective methods to quickly delineate regional prospecting areas.In the past few decades,exploration geochemical methods have played an important role in prospecting work in China.Many newly discovered mineral deposits in China have been found by exploration geochemical methods.At present,the general method of exploration geochemical prospecting is to analyze the frequency histogram of each geochemical element,use the mean±(1～2)×σ method to determine the lower limit of anomaly,delineate each geochemical element anomaly,and overlay the anomaly maps of different elements to determine the geochemical element combination anomaly.Although this method can quickly delineate geochemical prospecting areas,However,since this method ignores the spatial variability of geochemical background,it is inevitable to omit weak geochemical anomalies.The ensemble learning algorithm can accurately describe the complex geochemical background,and it is not easy to miss the weak geochemical anomalies.Therefore,in recent years,a series of ensemble learning algorithms have been successfully applied to the identification of multivariate geochemical anomalies and the delineation of prospecting areas,and have achieved good results.This paper takes the Wulaga area of Heilongjiang Province as the research area,and the geological and geochemical data obtained by the previous geological and mineral survey work as the main analysis and processing object,and carries out the research of multivariate geochemical anomaly identification method based on an ensemble learning algorithm.Based on Scikit-learn code,a Python program of multi-geochemical anomaly identification algorithm based on isolation forest algorithm,extended isolation forest algorithm and,generalized isolation forest Algorithm was developed and applied to the study of 1: 50000 stream sediment anomaly identification in the study area.The effect of multi-geochemical anomaly recognition of three anomaly recognition models was compared and analyzed.The research work of this paper has mainly achieved the following new opinions:(1)Isolation forest algorithm,extended isolation forest algorithm,and generalized isolation forest algorithm are effective multivariate geochemical anomaly identification methods.The corresponding area under the ROC curve(AUC)is above 0.7,indicating that the multivariate geochemical anomalies delineated by the three algorithm models have a strong spatial correlation with known gold deposits(points),which can reflect the metallogenic regularity of gold deposits in the study area to a certain extent.(2)There is a close correlation between the multi-geochemical anomaly areas identified by the three algorithms and the spatial distribution of the intermediate-acid magmatic intrusions,which are the main controlling factors of gold mineralization in the study area.It shows that the three ensemble learning algorithms can well identify the mineralization-related information in geochemical data.The anomaly identification results can be used as a scientific basis for delineating geochemical prospecting areas.(3)The three ensemble learning algorithms have certain limitations.The robustness of the three multivariate geochemical anomaly recognition models is poor.Under the same initialization parameters,the same algorithm model is run many times,and the anomaly recognition results are different.How to increase the robustness of the three algorithms is one of the problems that need further study.