基于多源地球物理数据的机器学习方法在地质体分类中的应用——以黑龙江多宝山矿集区为例  

Application of machine learning method based on multi-source geophysical data to geological body classification-A case study of Duobaoshan ore concentration area(Heilongjiang,China)

作　　者：李希元 崔健 胡望水[1] 李成立 LI XiYuan;CUI Jian;HU WangShui;LI ChengLi(School of Earth Science,Yangtze University,Wuhan 430100,China;Jiangsu Eastern China 814 Geophysical Exploration Co.,Ltd,Nanjing 210007,China;Shenyang Center,China Geological Survey,Shenyang 110034,China)

机构地区：[1]长江大学地球科学学院,武汉430100 [2]江苏华东八一四地球物理勘查有限公司,南京210007 [3]中国地质调查局沈阳地质调查中心,沈阳110034

出　　处：《地球物理学报》2022年第9期3634-3649,共16页Chinese Journal of Geophysics

基　　金：国家重点研发计划项目“北方东部复合造山成矿系统深部结构与成矿过程”(2017YFC0601300)资助。

摘　　要：为适应多源地球物理大数据地质解释的需要,同时也为了在地质体物性存在交叠情况下快速有效地实现多源地球物理数据的地质解释,本文提出应用机器学习的支持向量机方法对多源地球物理数据进行地质解释的新思路,并给出了利用物性三体(密度、磁化率、电阻率)进行地质体圈定与分类方法.阐明了参数归一化、参数寻优对模型训练与学习及分类结果的影响与作用.本文将黑龙江多宝山矿集区物性三体与矿集区的区域地质、矿床地质及钻井资料相结合,利用所提出的方法对地下地质体进行了圈定与分类,对分类结果经过交叉检验,正确率达81.6%,表明了训练模型具有较高的可信度.经对预测模型填充已知物性参数正演的重磁异常与实测重磁异常对比,证明两者在整体和细节上均有高度的相似性,间接说明对地质体圈定与分类结果的可靠性,进一步表明利用多源地球物理数据,采用支持向量机方法圈定地质体及对地质体进行岩性识别方面的合理性与有效性.多源地球物理数据机器学习的支持向量机方法在多宝山矿集区地质体圈定与分类所取得较好的应用效果,为多源综合地球物理的地质解释提供了可借鉴的成功经验,也提供了多源地球物理资料地质解释的一种新型的技术手段,开辟了应用人工智能方法进行多源地球物理资料地质解释的新途径.As the overlap of the physical properties of geological bodies,research for fast and geological interpretation of multi-source geophysical data is important for geological interpretation of multi-source geophysical big data.The support vector machine method based on machine learning is proposed to be used for geological interpretation of multi-source geophysical data,and a method for geological body delineation and classification by using three physical properties(density,susceptibility and resistivity)is given.The influence and function of parameter normalization and parameter optimization on model training and learning and classification results can be explained by this method.Using this method,the underground geological bodies are delineated and classified.The research data are regional geology,deposit geology and drilling data in the three-body combination area of Duobaoshan ore concentration area.The results indicated that the training model has high reliability,which correct rate of cross validation of classification results is 81.6%.The comparison between the forward gravity and magnetic anomalies filled with the known physical parameters in the prediction model and the measured gravity and magnetic anomalies shows that they are highly similar in the whole and details.The consistency of the comparison results can indirectly confirm the correctness of the geological body delineation and classification results,and then shows the rationality and effectiveness of this method in lithology identification of target specific geologic body.The efficient and accurate application effect of support vector machine method for multi-source geophysical data machine learning in delineation and classification of geological bodies in Duobaoshan ore concentration area can provide successful experience for geological interpretation of multi-source integrated geophysics,which greatly improves application prospect of artificial intelligence method in geological interpretation of multi-source geophysical data.

关 键 词：多源地球物理数据 机器学习 支持向量机 物性三体 地质体分类 

分 类 号：P631[天文地球—地质矿产勘探]