隐式建模和机器学习算法在西藏巨龙斑岩型铜钼矿床三维成矿预测中的应用研究

doi:10.13745/j.esf.sf.2024.12.82

摘要/Abstract

摘要：

三维成矿预测打破了传统二维预测图件的限制,可以将地学信息在真三维空间中直观地表达出来,因此受到人们越来越多的关注。三维地质建模和成矿预测方法是三维成矿预测过程中非常重要的步骤,但是随着地质数据采集方式的多样化,地质数据来源的多样性,地质数据也逐渐具备大数据的特征,传统的显式建模方式和成矿有利信息提取方法受到诸多限制,主要体现在不能有效地对三维模型进行实时更新和对大量地质勘查数据进行分析与处理。为了提出针对该类问题的一种有效解决方案,笔者选择西藏巨龙超大型斑岩铜钼矿床为研究对象,利用隐式建模方法构建矿区三维地质-地球化学模型,并通过机器学习算法对成矿有利信息进行提取、分析,最后对潜在的有利成矿空间进行预测。建模结果表明:隐式建模方法可以通过插值函数获得整个空间的地质体数据,三维地质体曲面重建算法可以对地质体数据进行三维建模与可视化,自动生成三维可视化模型。隐式建模大大降低了显示建模中人机交互圈定地质界线的烦琐过程,可以实现三维模型的快速动态更新,而且隐式建模在更大程度上能够精确反映地下深部地质体的空间分布特征。成矿预测主要基于三维地质-地球化学模型提取深部有利成矿信息,构建训练集和测试集,利用有监督的机器学习模型(逻辑回归模型、支持向量机模型、人工神经网络模型和随机森林模型)进行训练。训练后的模型对测试集进行预测,并将预测结果绘制成ROC曲线对预测结果进行评估。评估结果显示:4个不同预测模型的AUC值都大于0.6,说明所训练的模型预测精度是优于随机过程的,其中随机森林算法的AUC值最大(0.97),模型预测效果最优。本文选择随机森林模型对矿床深部进行预测,圈定了2个找矿靶区。经过钻探验证,靶区B潜在资源与预测结果相符,证实了该方法具有一定的科学性和可行性。

关键词: 巨龙铜钼矿, 隐式建模, 机器学习, 成矿预测

Abstract:

Three-dimensional metallogenic prediction overcomes the limitations of traditional two-dimensional prediction maps and can represent geo-information directly in true 3D space, so it has attracted more and more attention. 3D geological modeling and metallogenic prediction methods are important steps in the process of 3D metallogenic prediction. However, with the diversification of geological data collection methods and the diversity of geological data sources, geological data has gradually acquired the characteristics of big data. Traditional explicit modeling methods and extraction methods of favorable metallogenic information face many limitations. These limitations are mainly reflected in the inability to effectively update the 3D model in real time and analyze a large number of geological survey data. To address these challenges, the author selects the Julong super-large porphyry copper-molybdenum deposit in Tibet as the research object, employs implicit modeling methods to build a 3D geological-geochemical model of the mining area, and uses the machine learning algorithm to extract and analyze favorable metallogenic information, and finally predicts the potential favorable metallogenic space. The modeling results show that the implicit modeling method can obtain the geological entities of the whole space through the interpolation function, and the 3D geological body surface reconstruction algorithm can model and visualize the geological entities and automatically generate the 3D visualization model. Implicit modeling greatly reduces the cumbersome process of human-computer interaction delineating geological boundaries in explicit modeling, and can realize rapid dynamic update of 3D models. Moreover, implicit modeling can accurately reflect the spatial distribution characteristics of deep underground geological bodies to a greater extent. Metallogenic prediction is mainly based on 3D geological and geochemical models to extract deep favorable metallogenic information, build training sets and test sets, and train supervised machine learning models (logistic regression model, support vector machine model, artificial neural network model, random forest model) respectively. The trained model makes predictions to the test set, and plots the prediction results on Receiver Operating Characteristic (ROC) curves to evaluate the prediction results. The evaluation results show that the Area Under the Curve (AUC) values of the four different prediction models are all greater than 0.6, indicating that the prediction accuracy of the trained model is better than that of the random process, among which the AUC value of the random forest algorithm is the largest (0.97), and the model has the best prediction effect. In this paper, the random forest model is selected to predict the depth extent of the mineralization, and two prospecting targets are delineated. After drilling verification, the potential resources in target area B are consistent with the predicted results, which proves that the method is scientific and feasible.

Key words: Julong copper-molybdenum deposit, implicit modeling, machine learning algorithm, metallogenic prediction

中图分类号:

P612
P618.41

娄渝明, 康旭, 赖渊平, 龚建生, 周涤非, 窦世荣, 樊炳良, 丁帅, 舒德福, 陈根. 隐式建模和机器学习算法在西藏巨龙斑岩型铜钼矿床三维成矿预测中的应用研究[J]. 地学前缘, 2025, 32(5): 440-455.

LOU Yuming, KANG Xu, LAI Yuanping, GONG Jiansheng, ZHOU Difei, DOU Shirong, FAN Bingliang, DING Shuai, SHU Defu, CHEN Gen. Application of implicit modeling and machine learning algorithm to 3D metallogenic prediction of the Julong porphyry copper-molybdenum deposit, Xizang[J]. Earth Science Frontiers, 2025, 32(5): 440-455.

图/表 21

图1 西藏巨龙铜钼矿床地质简图(a 据文献[31]修改;b据文献[35]补充修改)

Fig.1 Geological schematic map of Julong copper and molybdenum deposit in Xizang. a modified after [31], b modified after [35].

图2 巨龙铜矿体空间分布图

Fig.2 Spatial distribution map of Julong copper orebody

图3 巨龙主矿体开采情况

Fig.3 Mining situation of main orebody in Julong deposit

图4 隐式三维建模流程图

Fig.4 Flow chart of implicit 3D modeling

图5 巨龙铜钼矿床三维地质模型 a—三维成矿预测研究区;b—二长花岗斑岩;c—黑云母二长花岗岩;d—花岗闪长岩;e—石英闪长玢岩;f—隐爆角砾岩;g—叶巴组凝灰岩;h—英安流纹斑岩;i—巨龙铜矿体Cu含量≥0.15%。

Fig.5 3D geological model of Julong copper and molybdenum deposit a—3D metallogenic prediction research area;b—Monzonite granite porphyry;c—Biotite monzonitic granite; d—Granodiorite; e—Quartz diorite porphyrite;f—Cryptoexplosive breccia;g—Yeba Formation tuff;h—Dacite rhyolite porphyry;i—Julong copper orebody Cu≥0.15%.

图6 Sigmoid函数示意图

Fig.6 Sigmoid function diagram

图7 SVM模型训练流程图

Fig.7 SVM model training flow chart

图8 神经网络模型示意图

Fig.8 Schematic diagram of neural network model

图9 随机森林模型示意图

Fig.9 Schematic diagram of random forest model

表1 巨龙铜钼矿床预测要素表

Table 1 Prediction factors of Julong copper and molybdenum deposit

预测要素		要素特征描述	要素分类
地质特征	地层	中-下侏罗统叶巴组凝灰岩	重要
	构造	区域深大断裂构造	不重要
	岩浆岩	中侏罗世英安流纹斑岩	重要
	岩浆岩	中侏罗世花岗闪长岩、中新世黑云母二长花岗岩、中新世二长花岗斑岩、中新世石英闪长玢岩组成的复式杂岩体	必要
	矿体	已知矿体	必要
地球化学特征	元素异常	Cu原生晕异常	重要
地球化学特征	元素异常	Mo原生晕异常	重要

图10 巨龙铜钼矿床地质成矿信息统计分析结果

Fig.10 Statistical analysis results of geological and metallogenic information of Julong copper and molybdenum deposit

图11 巨龙铜钼矿床地球化学成矿信息 a—Cu元素空间插值;b—Mo元素空间插值。

Fig.11 Geochemical metallogenic information of Julong copper and molybdenum deposit a—Cu element space interpolation;b—Mo element space interpolation.

图12 巨龙铜钼矿床预测变量回归系数

Fig.12 Regression coefficient of predictor variables in Julong copper and molybdenum deposit

图13 巨龙铜钼矿床支持向量机建模参数图

Fig.13 Regression coefficient of predictor variables in Julong copper and molybdenum deposit

图14 不同隐藏层误差曲线

Fig.14 Error curves of different hidden layers

图15 不同batch_size误差曲线

Fig.15 Error curve of different batch_size

图16 不同优化器误差曲线

Fig.16 Error curves for different optimizers

图17 巨龙铜钼矿床随机森林建模参数图 a—OOB误差与预测器数量之间的关系;b—树的数量。

Fig.17 Random forest modeling parameters of Julong copper and molybdenum deposit

图18 不同机器学习模型ROC曲线

Fig.18 ROC curves of different machine learning models

图19 巨龙铜钼矿床深部三维预测结果

Fig.19 Results of 3D deep prediction of Julong copper and molybdenum deposit

图20 巨龙铜钼矿床深部潜在资源量验证

Fig.20 Verification of deep potential resources of Julong copper and molybdenum deposit

参考文献 64

[1]	李文昌, 李建威, 谢桂青, 等. 中国关键矿产现状、研究内容与资源战略分析[J]. 地学前缘, 2022, 29(1): 1-13.
[2]	翟明国, 吴福元, 胡瑞忠, 等. 战略性关键金属矿产资源: 现状与问题[J]. 中国科学基金, 2019, 33(2): 106-111.
[3]	HAGEMANN S G, LISITSIN V A, HUSTON D L. Mineral system analysis: quo vadis[J]. Ore Geology Reviews, 2016, 76: 504-522.
[4]	耿瑞瑞. 鹿井铀矿床深部和外围三维成矿预测研究[D]. 北京: 核工业北京地质研究院, 2021.
[5]	COLLON P, PICHAT A, KERGARAVAT C, et al. 3D modeling from outcrop data in a salt tectonic context: example from the Inceyol minibasin, Sivas Basin, Turkey[J]. Interpretation, 2016, 4(3): SM17-SM31.
[6]	HILLIER M J, SCHETSELAAR E M, DE KEMP E A, et al. Three-dimensional modelling of geological surfaces using generalized interpolation with radial basis functions[J]. Mathematical Geosciences, 2014, 46(8): 931-953.
[7]	周邓, 曾广亮, 王洪荣, 等. 江西桃山罗布里南部地区三维地质建模与成矿预测[J]. 地质通报, 2022, 41(12): 2256-2264.
[8]	李青元, 张洛宜, 曹代勇, 等. 三维地质建模的用途、现状、问题、趋势与建议[J]. 地质与勘探, 2016, 52(4): 759-767.
[9]	李楠, 曹瑞, 叶会寿, 等. 内蒙古浩尧尔忽洞金矿三维建模与深部成矿预测[J]. 地学前缘, 2021, 28(3): 170-189.
[10]	赵鹏大, 陈建平, 张寿庭. “三联式” 成矿预测新进展[J]. 地学前缘, 2003, 10(2): 455-463.
[11]	赵鹏大. “三联式” 资源定量预测与评价: 数字找矿理论与实践探讨[J]. 地球科学:中国地质大学学报, 2002, 27(5): 482-489.
[12]	宋建军. 新常态下地质工作创新发展的思考[J]. 中国国土资源经济, 2017, 30(3): 4-8.
[13]	吴磊. 新形势下地质矿产勘查及找矿技术的分析[J]. 世界有色金属, 2022(23): 73-75.
[14]	毛先成, 王琪, 陈进, 等. 胶西北金矿集区深部成矿构造三维建模与找矿意义[J]. 地球学报, 2020, 41(2): 166-178.
[15]	陈麒玉, 刘刚, 何珍文, 等. 面向地质大数据的结构-属性一体化三维地质建模技术现状与展望[J]. 地质科技通报, 2020, 39(4): 51-58.
[16]	谷浩, 杨泽强, 高猛, 等. 河南围山城金银矿集区三维地质建模与成矿预测[J]. 地学前缘, 2023, 31(3): 245-259.
[17]	MAYER-SCHÖNBERGER V, CUKIER K. Big data: a revolution that will transform how we live, work, and think[M]. Boston: Houghton Mifflin Harcourt, 2013.
[18]	王娜. 地质大数据功能分析及其分类算法研究[D]. 长春: 吉林大学, 2019.
[19]	李章林, 吴冲龙, 张夏林, 等. 地质科学大数据背景下的矿体动态建模方法探讨[J]. 地质科技通报, 2020, 39(4): 59-68.
[20]	周永章, 左仁广, 刘刚, 等. 数学地球科学跨越发展的十年: 大数据、人工智能算法正在改变地质学[J]. 矿物岩石地球化学通报, 2021, 40(3): 556-573.
[21]	ZUO R G. Machine learning of mineralization-related geochemical anomalies: a review of potential methods[J]. Natural Resources Research, 2017, 26(4): 457-464.
[22]	周永章, 陈烁, 张旗, 等. 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, 2018, 34(2): 255-263.
[23]	刘艳鹏, 朱立新, 周永章. 卷积神经网络及其在矿床找矿预测中的应用: 以安徽省兆吉口铅锌矿床为例[J]. 岩石学报, 2018, 34(11): 3217-3224.
[24]	肖克炎, 孙莉, 李楠, 等. 大数据思维下的矿产资源评价[J]. 地质通报, 2015, 34(7): 1266-1272.
[25]	刘艳鹏, 朱立新, 周永章. 大数据挖掘与智能预测找矿靶区实验研究: 卷积神经网络模型的应用[J]. 大地构造与成矿学, 2020, 44(2): 192-202.
[26]	成秋明. 什么是数学地球科学及其前沿领域?[J]. 地学前缘, 2021, 28(3): 6-25.
[27]	HOU Z Q, GAO Y F, QU X M, et al. Origin of adakitic intrusives generated during mid-Miocene east-west extension in southern Tibet[J]. Earth and Planetary Science Letters, 2004, 220(1/2): 139-155.
[28]	徐志刚, 陈毓川, 王登红, 等. 中国成矿区带划分方案[M]. 北京: 地质出版社, 2008.
[29]	HOU Z Q, COOK N J. Metallogenesis of the Tibetan collisional orogen: a review and introduction to the special issue[J]. Ore Geology Reviews, 2009, 36(1/2/3): 2-24.
[30]	YANG Z M, HOU Z Q, WHITE N C, et al. Geology of the post-collisional porphyry copper-molybdenum deposit at Qulong, Tibet[J]. Ore Geology Reviews, 2009, 36(1/2/3): 133-159.
[31]	YIN A, HARRISON T M. Geologic evolution of the Himalayan-Tibetan Orogen[J]. Annual Review of Earth and Planetary Sciences, 2000, 28: 211-280.
[32]	郑有业, 薛迎喜, 程力军, 等. 西藏驱龙超大型斑岩铜(钼)矿床: 发现、特征及意义[J]. 地球科学:中国地质大学学报, 2004, 29(1): 103-108.
[33]	佘宏全, 丰成友, 张德全, 等. 西藏冈底斯铜矿带甲马夕卡岩型铜多金属矿床与驱龙斑岩型铜矿流体包裹体特征对比研究[J]. 岩石学报, 2006, 22(3): 689-696.
[34]	HU Y B, LIU J Q, LING M X, et al. The formation of Qulong adakites and their relationship with porphyry copper deposit: Geochemical constraints[J]. Lithos, 2015, 220: 60-80.
[35]	西藏巨龙铜业有限公司. 西藏自治区墨竹工卡县巨龙矿区铜矿资源储量核实报告[R]. 拉萨: 西藏巨龙铜业有限公司, 2022.
[36]	孟祥金, 侯增谦, 李振清. 西藏驱龙斑岩铜矿S、Pb同位素组成: 对含矿斑岩与成矿物质来源的指示[J]. 地质学报, 2006, 80(4): 554-560.
[37]	HAN Z H, WANG R, TONG X S, et al. Multi-scale exploration of giant Qulong porphyry deposit in a collisional setting[J]. Ore Geology Reviews, 2021, 139: 104455.
[38]	字艳梅, 田世洪, 陈欣阳, 等. 埃达克岩与热液成矿过程中钾镁同位素分馏及其指示意义: 以驱龙斑岩铜矿床为例[J]. 地学前缘, 2024, 31(3): 150-169.
[39]	孟祥金, 侯增谦, 高永丰, 等. 碰撞造山型斑岩铜矿蚀变分带模式: 以西藏冈底斯斑岩铜矿带为例[J]. 地学前缘, 2004, 11(1): 201-214.
[40]	王登芳, 蒋宗洋, 岳宁飞, 等. 西藏知不拉和浪母家果矽卡岩铜矿地质特征及驱龙-知不拉-浪母家果成矿系统的确定[J]. 地质与勘探, 2015, 51(4): 619-633.
[41]	曲焕春. 西藏驱龙斑岩铜钼矿床UST成因: 对大型斑岩矿床成因的暗示[D]. 北京: 中国地质大学(北京), 2016.
[42]	孙伟涛. 西藏驱龙斑岩型铜钼矿床成矿元素特征及找矿指示意义[D]. 北京: 中国地质大学(北京), 2019.
[43]	CHUNG C F, AGTERBERG F P. Regression models for estimating mineral resources from geological map data[J]. Journal of the International Association for Mathematical Geology, 1980, 12(5): 473-488.
[44]	HOSMER D W, LEMESBOW S. Goodness of fit tests for the multiple logistic regression model[J]. Communications in Statistics-Theory and Methods, 1980, 9(10): 1043-1069.
[45]	BISHOP C M. Pattern recognition and machine learning[M]. 2 ed. Berlin: Springer Google Scholar, 2006.
[46]	娄渝明. 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[D]. 成都: 成都理工大学, 2023.
[47]	SUYKENS J, VANDEWALLE J. Least squares support vector machine classifiers[J]. Neural Processing Letters, 1999, 9(3): 293-300.
[48]	VAPNIK V N. The nature of statistical learning theory[M]. New York: Springer, 1995.
[49]	MÜLLER B, REINHARDT J, M T S. Neural networks[J]. Physics of Neural Networks, 1992, 293(4): 2588-2592.
[50]	HAYKIN S S. Neural networks: a comprehensive foundation[M]. 2nd ed. Upper Saddle River, NJ: Prentice Hall, 1998.
[51]	RUSSELL S J, NORVIG P. Artificial intelligence:a modern approach[M]. London: Pearson, 2016.
[52]	BREIMAN L. Random forests[J]. Machine Learning, 2024, 45: 5-32.
[53]	BREIMAN L. Bagging predictors[J]. Machine Learning, 1996, 24(2): 123-140.
[54]	GORDON A D, BREIMAN L, FRIEDMAN J H, et al. Classification and regression trees[J]. Biometrics, 1984, 40(3): 874.
[55]	CARRANZA E J M. Geochemical anomaly and mineral prospectivity mapping in GIS[M]. Amsterdam: Elsevier, 2009.
[56]	ALLISON P D. Logistic regression using the Sas system: theory and application[M]. Cary, NC: SAS Publishing, 2001.
[57]	XIANG J, XIAO K Y, CARRANZA E J M, et al. 3D mineral prospectivity mapping with random forests: a case study of Tongling, Anhui, China[J]. Natural Resources Research, 2020, 29(1): 395-414.
[58]	马瑶, 赵江南. 机器学习方法在矿产资源定量预测应用研究进展[J]. 地质科技通报, 2021, 40(1): 132-141.
[59]	花旗, 夏庆霖, 刘奇锋. 基于数据驱动的南岭地区花岗岩岩体含矿性判别[J]. 地质科技通报, 2025, 44(1): 332-345.
[60]	NYAKILLA E E, SILINGI S N, SHEN C B, et al. Evaluation of source rock potentiality and prediction of total organic carbon using well log data and integrated methods of multivariate analysis, machine learning, and geochemical analysis[J]. Natural Resources Research, 2022, 31(1): 619-641.
[61]	OBUCHOWSKI N A. Receiver operating characteristic curves and their use in radiology[J]. Radiology, 2003, 229(1): 3-8.
[62]	NYKÄNEN V, LAHTI I, NIIRANEN T, et al. Receiver operating characteristics (ROC) as validation tool for prospectivity models: a magmatic Ni-Cu case study from the Central Lapland Greenstone Belt, Northern Finland[J]. Ore Geology Reviews, 2015, 71: 853-860.
[63]	李程. 深部地质地球化学三维定量矿产预测方法研究: 以西秦岭早子沟金矿为例[D]. 成都: 成都理工大学, 2021.
[64]	陈建平, 史蕊, 王丽梅, 等. 基于数字矿床模型的陕西潼关县Q8号金矿脉西段三维成矿预测[J]. 地质学刊, 2012, 36(3): 237-242.