基于机器学习的深海多金属结核成因分类  

Genetic Classification of Deep-sea Polymetallic Nodules Based on Machine Learning

作　　者：尹浩文 成秋明 YIN Hao-wen;CHENG Qiu-ming(School of Earth Science and Engineering,Sun Yat-sen University,Zhuhai 519000,China;State Key Labrtory of Geological Processes and Mineral Resources,China University of Geosciences,Beijing 100083,China;Frontier Science Center for Deep Time Digital Earth,Ministry of Education,China University of Geosciences,Beijing 100083,China)

机构地区：[1]中山大学地球科学与工程学院,珠海519000 [2]中国地质大学地质过程与矿产资源国家重点实验室,北京100083 [3]中国地质大学教育部深时数字地球前沿科学中心,北京100083

出　　处：《科学技术与工程》2024年第25期10605-10619,共15页Science Technology and Engineering

基　　金：大数据-数学地球科学创新研发团队和极端地质事件团队项目(2021ZT09H399)。

摘　　要：铁锰结核广泛分布于深海平原,储量巨大,具有商业开采潜力。利用1128个铁锰结核样本的地球化学数据和8种地质与海洋要素,采用随机森林机器学习方法,探讨结核成因分类。首先,基于Mn、Fe、Cu、Co、Ni、Mn/Fe和Fe/Co地球化学数据使用高斯混合模型聚类方法对1128个样本进行成因分类,并作为训练数据。其次,基于海底沉积速率、海水底部溶氧量和海水表面生物初级生产力等地质-海洋特征建立预测模型,将结核划分为水成型、成岩型和混合型,结果显示,模型对水成型和成岩型结核的分类精度分别为91%和66%,对混合型的分类精度较低,仅为23%。应用该模型对全球4119个铁锰结核进行成因分类,结果表明,水成型结核占71.8%,混合型占21.8%,成岩型占6.2%。水成型结核广泛分布于各大洋,而成岩型和混合型则集中在大洋中纬度地区,如东太平洋的克拉里昂-克里帕顿断裂带和东南太平洋的秘鲁海盆等。这些地区的沉积物速率、海底生物量和含氧量显著影响结核分布。尽管基于地球化学数据的分类方法更可靠,研究表明,利用地质和海洋要素及机器学习方法也可有效分类。Manganese nodules are widely distributed across deep-sea plains and have significant commercial mining potential due to their vast reserves.Based on the geochemical data of 1128 iron and manganese nodule samples and 8 geological and marine elements,the genetic classification of nodule was discussed by using random forest machine learning method.Firstly,based on Mn,Fe,Cu,Co,Ni,Mn/Fe and Fe/Co geochemical data,1128 samples were classified by Gaussian mixture model clustering method and used as training data.Secondly,a prediction model was established based on the geological and marine characteristics such as seabed deposition rate,dissolved oxygen at the bottom of seawater,and biological primary productivity on the surface of seawater,and the nodules were divided into hydroforming,diagenetic and mixed types.The results show that the classification accuracy of the model for hydroforming and diagenetic nodules is 91%and 66%respectively,while the classification accuracy of the mixed type is only 23%.The genetic classification of 4119 ferromanganese nodules in the world by this model shows that hydroforming nodules account for 71.8%,mixed type 21.8%and diagenetic type 6.2%.Hydroforming nodules are widely distributed in the oceans,while diagenetic and mixed nodules are concentrated in the mid-latitudes,such as the Clarion-Clipaton fault zone in the eastern Pacific Ocean and the Peru Basin in the Southeast Pacific Ocean.Sediment development,seafloor biomass and oxygen content in these areas significantly affect nodule distribution.Although classification methods based on geochemical data are more reliable,studies have shown that the use of geological and marine elements and machine learning methods can also be effective.

关 键 词：海洋矿产资源 铁锰结核 成因分类 空间分布 机器学习 

分 类 号：P744[天文地球—海洋科学]