Research Of Comprehensive Prospecting Prediction Based On Compositional Data Analysis And Machine Learning Model In Ashele Copper-Zinc Deposit,Xinjiang,NW China

The Ashele copper-zinc deposit,located in the Ashele basin on the Altai orogenic belt's southwestern margin,is a typical volcanic massive sulfide deposit(VMS).Numerous studies,including essential geological characteristics,ore-forming material sources,ore-forming mechanism,and ore-forming prediction of the deposit,have been made by Predecessors and accumulated a large number of geological data and prospecting results.Since the Ashele copper-zinc deposit has the characteristics of deep burial,significant difficulty in mining,and high mining maintenance cost,along with the continuous consumption of proved mineral reserves,the grade of the deep-margin orebody decreases,the prospecting task for the deep and peripheral parts of the Ashele copper-zinc deposit is exceptionally urgent.Under the guidance of the quantitative evaluation system of mineral resources and based on the comprehensive collection of geological data and previous research results in the study area,this paper summarizes the ore-forming geological patterns and ore-controlling geological features of the mining area.The compositional data analysis approach is first applied to preprocess rock geochemical data collected from mining and periphery areas of Ashele district.Moreover,the geological anomalies are separated,identified,and extracted by fractal and singularity theory.A comprehensive information prospecting model in the study area is constructed based on the quantification of the ore-controlling geological features in the mining area and geochemical prospecting indexes.After that,three different machine learning algorithms are chosen to carry out a prospecting prediction in the periphery of the mining area,and the prediction results of different learning models are evaluated.Finally,the borehole primary halo's vertical zoning characteristics are analyzed,and the prospecting potential in the deep part of the mining area is assessed.The achievements and understandings of this paper are as follows:(1)The rock geochemical data of nine elements in the study area were analyzed using the compositional data analysis method to obtain the elements' natural spatial distribution.The elemental assemblages' characteristics were explored by the robust principal component method,and two sets of mineralization assemblages,namely,Cu-Zn-Co and Pb-Mo-Ag-Au-Sb,corresponding to the two mineralization stages of exhalative sedimentation and metamorphic hydrothermal superimposed transformation,respectively,were derived.(2)Fractal-multifractal method is applied to separate element geochemical anomalies and background distribution pattern at the space domain and frequency domain to extract mineral primary halo anomalies in the study area.Weak anomalies,which are difficult to be identified by conventional geochemical data processing methods,are identified and extracted by singularity theory to fully mine the weak anomaly closely related to mineralization in geochemical data.(3)The study area's ore-controlling geological features are summarized by analyzing the mining area's metallogenic regularity.Taking the GIS information system as the medium and the ore-beared boreholes in the mining area as the reference,the "distance distribution method" is used to determine the optimal buffer distance between all kinds of ore-controlling geological features and orebodies.A geological-geochemical comprehensive information prospecting model in the study area was put forward by combining with the surficial primary halo geochemical comprehensive anomalies and all kinds of quantified ore-controlling geological features information which closely related to mineralization.(4)Based on the geological-geochemical comprehensive information prospecting model,three kinds of supervised learning algorithms are chosen to carry out a prospecting prediction in the study area.After that,all kinds of machine learning models are evaluated,and the prediction results of each model correspond to the ore boreholes in the mining area.It is concluded that the prospecting prediction effect of the three kinds of machine learning models is remarkable.Therefore,it is proposed to combine three machine learning algorithms to construct an integrated prospecting prediction model based on machine learning.(5)Combining the result of integrated machine learning model and weak anomalies delineated by rock geochemical data,along with the geological background of the study area and the evaluation index of the significance of ore-controlling geological features in the mining area,a total of 9 prospecting prediction areas of three types are delineated in the periphery of Ashele copper-zinc mining area.Furthermore,the potential for deep-seated mineralization is recognized with the primary halo drilling dataset.