The Shannan region in Xizang, situated within the eastern segment of the Tethyan Himalaya, is a significant area for the concentrated development of gold and antimony deposits in China. Over 60 gold and antimony deposits (occurrences) have been identified within the zone extending from Cuomei County to Longzi County. To enhance the identification of potential deposits (occurrences) and achieve breakthroughs in prospecting, machine learning methodologies can be employed to predict these potential deposits. This study focuses on the area covered by 15 standard 1∶50 000 geological survey sheets in Cuomei County and Longzi County, Shannan, Xizang. It integrates geological, geochemical, geophysical, topographical, and other relevant datasets to construct two XGBoost prediction models for gold and antimony deposits in the region, achieving model accuracies of 0.8807 and 0.9031, respectively. A dataset for predicting mineralization of gold and antimony deposits in the study area has been developed using machine learning techniques. This dataset facilitates the identification of prospective mineralization zones and prospecting targets for gold and antimony deposits. The findings demonstrate that the XGBoost model exhibits satisfactory performance in predicting the mineralization of gold and antimony deposits in Shannan, Xizang. Based on these predictive outcomes, 27 gold prospecting targets and 11 antimony prospecting targets have been identified. These research findings offer valuable insights for guiding gold and antimony prospecting efforts in this region.
西藏山南地处特提斯喜马拉雅构造域东段,属于藏南金锑矿成矿带,是我国金、锑矿床发育较为集中的区域之一,尤其是措美县—隆子县一带。近10年来,研究人员在该区开展了1∶5万区域地质矿产调查项目,发现60余个金、锑矿床和矿(化)点,并采用信息量计算法进行成矿预测。信息量计算法的优势在于解释性强且计算高效,适合小样本数据下的特征筛选和地质标志提取,但无法处理大数据和理解复杂成矿规律。由于矿床(点)与赋矿地层、控矿构造、岩浆活动、变质作用、地球化学异常、地球物理异常具有一定的成因联系,通过集成这些成矿有利信息进行机器学习模型训练,挖掘其与成矿潜在的时空耦合关系(聂虹等,2018;左仁广,2019,2021),能够实现潜在矿床(点)的成矿概率预测(Zuo et al,2020;Xiong et al,2018)。本文选择西藏山南措美县—隆子县15个1∶5万矿调标准图幅范围作为研究区,融合地质、地球化学、地球物理和地形等数据,基于XGBoost算法构建金、锑矿预测模型,利用机器学习方法获得该区金、锑矿成矿预测数据集,并据此圈定金、锑矿找矿靶区,进一步指导找矿工作部署。
本研究利用mIoU(Mean Intersection over Union,mIoU)均值进行模型评估,该指标是预测模型常用的评估手段。它是真实值和预测值2个集合的交集和并集之比的平均值。通过对每个子模型在验证集上的性能加权平均输出权重,对每个像元成矿概率的加权平均值阈值化,若某个像元成矿概率的加权平均值大于70%,则该像元被认为具有成矿潜力,像元成矿概率的加权平均值计算公式为
其中,TP、FP和FN的含义与式(3)相同。模型通过概率排序生成阈值序列,从最高阈值(全样本判为负类)逐步降低至最低阈值(全样本判为正类),逐点计算对应的FPR与TPR并连接成曲线。以曲线下方面积(Area Under the Curve,AUC)作为量化指标,将性能评估简化为单一数值,数值越接近1,代表分类精度越高。经计算,AUC(Au)=0.915,AUC(Sb)=0.923,表明金、锑矿成矿预测模型具有良好的预测性能。金、锑矿预测模型ROC曲线如图6和图7所示。
3.2 变量相对重要性
变量相对重要性是衡量模型变量贡献程度的评判依据。在对数据进行回归分类的同时,计算各个变量的权重指数,评估各变量在预测中所起的作用(Sun et al,2019;蒋海昆等,2023)。在每次实验中,将包含特定变量的子模型全部去除,使用剩余的子模型组成新的加权平均集成模型。本研究利用参与建模的地层(三叠系T、侏罗系J、第四系Q)、岩浆岩(中酸性IA、中基性IB)、变质岩(MR)、构造(近东西GS-EW、近南北GS-SN、北西GS-NW、北东GS-NE和穹窿GS-Dome)、不同元素水系沉积物含量、剩余重力异常(RGA)和地形数据(DEM)作为变量,通过逐个置换变量并量化预测误差,输出各个变量的贡献(图8和图9)。
LaffanS W, NielsenO M, SilcockH,et al,2005.Sparse Grids:a new predictive modelling method for the analysis of geographic data[J].International Journal of Geographical Information Science,19(3):267-292.
[2]
LiS, ChenJ P, LiuC,et al,2021.Mineral prospectivity prediction via convolutional neural networks based on geological big data[J].Journal of Earth Science,32(2):327-347.
[3]
LiS, ChenJ P, XiangJ,2020.Applications of deep convolutional neural networks in prospecting prediction based on two-dimensional geological big data[J].Neural Computing and Applications,32(7):2037-2053.
[4]
SunT, ChenF, ZhongL,et al,2019.GIS-based mineral prospectivity mapping using machine learning methods:a case study from Tongling ore district,eastern China[J].Ore Geology Reviews,109:26-49.
[5]
XiongY H, ZuoR G, CarranzaE J M,2018.Mapping mineral prospectivity through big data analytics and a deep learning algorithm[J].Ore Geology Reviews,102:811-817.
[6]
YinJ N, LiN,2022.Ensemble learning models with a Bayesian optimization algorithm for mineral prospectivity mapping[J].Ore Geology Reviews,145:104916.
[7]
ZuoR G, WangZ Y,2020.Effects of random negative training samples on mineral prospectivity mapping[J].Natural Resources Research,29(6):3443-3455.
ChenJin, MaoXiancheng, LiuZhankun,et al,2020.Three-dimensional metallogenic prediction based on random forest classification algorithm for the Dayingezhuang gold deposit[J].Geotectonica et Metallogenia,44(2):231-241.
ChengQiuming,2025.A new paradigm for mineral resource prediction based on human intelligence-artificial intelligence integration[J].Earth Science Frontiers,32(4):1-19.
[14]
邓舟,2020.藏南马扎拉金锑矿矿床成因研究[M].成都:成都理工大学.Deng Zhou,2020.Study on genesis of the Mazhala Au-Sb deposit in south Tibet[M].Chengdu:Chengdu University of Technology.
DongFuquan,2012.Fractal characteristics of fractures and its geological significance in Gudui-Longzi region in southern Tibet[J].Gold Science and Technology,20(6):41-45.
[17]
黄小东,2011.藏南拆离系江孜—隆子金—锑成矿带成矿规律与找矿方向研究[M].成都:成都理工大学.Huang Xiaodong,2011.Study on metallogenic regularity and ore-prospecting direction of Gyantse-Lhunze gold-antimony metallogenic belt in the South Tibetan detachment system[M].Chengdu:Chengdu University of Technology.
JiangHaikun, WangJinhong,2023.Discussion on the importance of the features for the judgement of earthquake sequence types applicable to machine learning[J].Journal of Seismological Research,46(2):155-172.
LiCangbai, XiaoKeyan, LiNan,et al,2020. A comparative study of support vector machine,random forest and artificial neural network machine learning algorithms in geochemical anomaly information extraction[J].Acta Geoscientica Sinica,41(2):309-319.
LiuYanpeng, ZhuLixin, ZhouYongzhang,2020.Experimental research on big data mining and intelligent prediction of prospecting target area:application of convolutional neural network model[J].Geotectonica et Metallogenia,44(2):192-202.
[24]
娄渝明,2023.基于地质—地球化学信息的西藏雄村矿集区地表和深部成矿预测研究[M].成都:成都理工大学.Lou Yuming,2023.Study on surface and deep metallogenic prediction based on geological-geochemical information in Xiongcun district,Tibet[M].Chengdu:Chengdu University of Technology.
LouYuming, KangXu, LaiYuanping,et al,2025.Application of implicit modeling and machine learning algorithm to 3D metallogenic prediction of the Julong porphyry copper-molybdenum deposit,Xizang[J].Earth Science Frontiers,32(5):440-455.
LouYuanlin, ChenWu, YangTao,2019.Metallogenic model and prospecting pattern of the Bangzhuoma gold deposit in Longzi County,Tibet[J].Geological Bulletin of China,38():449-461.
LouYuanlin, QianJianli, ZhuZhiping,et al,2024.Application of integrated geophysical,geochemical and remote sensing prospecting methods in Lajiu area,Longzi County,Tibet[J].Gold Science and Technology,32(2):241-257.
MaYao, ZhaoJiangnan,2021.Advances in the application of machine learning methods in mineral prospectivity mapping[J].Bulletin of Geological Science and Technology,40(1):132-141.
NieHong, ZhuYueqin, ChangLiheng,et al,2018.Research on construction method of data-driven minerals prediction model[J].China Mining Magazine,27(9):82-87.
OuyangYuan, LiuHong, LiGuangming,et al,2023.Mineral search prediction based on Random Forest algorithm:a case study on porphyry-epithermal copper polymetallic deposits in the western Gangdise meatallogenic belt[J].Geology in China,50(2):303-330.
YuanFeng, LiXiaohui, ZhangMingming,et al,2014.Three dimension prospectivity modelling based on integrated geoinformation for prediction of buried orebodies[J].Acta Geologica Sinica,88(4):630-643.
ZhangBimin, WangXueqiu, ZhouJian,et al,2025.Copper mineralization pattern and machine learning-based copper prospectivity prediction in Laos[J].Earth Science Frontiers,32(1):61-77.
[43]
张刚阳,2012.藏南金锑多金属成矿带成矿模式与找矿前景研究[M].武汉:中国地质大学(武汉).Zhang Gangyang,2012.Metallogenic model and prospecting potential in Southern Tibet Au-Sb polymetallic belt[M].Wuhan:China University of Geosciences.
ZhangYu, WeiJunhao, ShiWenjie,et al,2022.Soil geochemical anomaly information extraction and metallogenic prediction of the Buzhu gold(antimony) deposit in the Kangma County,south Tibet[J].Gold Science and Technology,30(1):1-18.
ZhangZhenjie, ChengQiuming, YangJie,et al,2021.Machine learning for mineral prospectivity:a case study of iron-polymetallic mineral prospectivity in southwestern Fujian[J].Earth Science Frontiers,28(3):221-235.
[48]
周永章,陈烁,张旗,等 ,2018a.大数据与数学地球科学研究进展:大数据与数学地球科学专题代序[J].岩石学报,34(2):255-263.ZhouYongzhang,ChenShuo,ZhangQi,et al,2018a.Advances and prospects of big data and mathematical geoscience[J].Acta Petrologica Sinica,34(2):255-263.
ZhouYongzhang, LiPeixing, WangShugong,et al,2017.Research progress on big data and intelligent modelling of mineral deposits[J].Bulletin of Mineralogy,Petrology and Geochemistry,36(2):327-331,344.
[51]
周永章,王俊,左仁广,等 ,2018b.地质领域机器学习、深度学习及实现语言[J].岩石学报,34(11):3173-3178.ZhouYongzhang,WangJun,ZuoRenguang,et al,2018b.Machine learning,deep learning and Python language in field of geology[J].Acta Petrologica Sinica,34(11):3173-3178.
ZuoRenguang,2019.Deep learning-based mining and integration of deep-level mineralization information[J].Bulletin of Mineralogy,Petrology and Geochemistry,38(1):53-60.