| 机器学习与成矿预测:以闽西南铁多金属矿预测为例 |
| |
| 引用本文： | 张振杰,成秋明,杨玠,武国朋,葛云钊.机器学习与成矿预测:以闽西南铁多金属矿预测为例[J].地学前缘,2021,28(3):221-235. |
| |
| 作者姓名： | 张振杰 成秋明 杨玠 武国朋 葛云钊 |
| |
| 作者单位： | 中国地质大学(北京)地球科学与资源学院,北京100083;中国地质大学(北京)地质过程与矿产资源国家重点实验室,北京100083;中国地质大学(北京)地质过程与矿产资源国家重点实验室,北京100083;中国地质大学(北京)科学研究院,北京100083;中国地质大学(北京)地球科学与资源学院,北京100083 |
| |
| 基金项目： | 国家重点研发计划项目(2018YFE0204204);国家自然科学基金项目(41702075);国家自然科学基金项目(42050103);教育部中央高校基本科研业务费项目(2652018132) |
| |
| 摘 要： | 作为近年来爆炸式发展的方法模型,机器学习为地质找矿提供了新的思维和研究方法。本文探讨矿产预测研究的理论方法体系,总结机器学习在矿产预测领域的特征信息提取和信息综合集成两个方面的应用现状,并讨论机器学习在矿产资源定量预测领域面临的训练样本稀少且不均衡、模型训练中缺乏不确定性评估、缺少反哺研究、方法选择等困难和挑战。进一步以闽西南马坑式铁矿为实例论述基于机器学习方法的矿产预测基本流程:(1)通过成矿系统研究建立成矿模型,确定矿床控矿要素;(2)通过勘查系统研究建立找矿模型,并为评价预测提供相关的勘查数据;(3)通过预测评价系统研究,建立预测模型,并提取预测要素;(4)利用机器学习模型对预测要素进行信息综合集成,获取成矿有利度图;(5)对预测性能和结果进行不确定性评估;(6)找矿靶区/成矿远景区圈定及资源量估算。最后,总结建立以地学大数据和地球系统理论为指导,以“地球系统-成矿系统-勘查系统-预测评价系统”为研究路线的基于地学大数据的矿产资源定量预测理论和方法体系的研究愿景。
|
| 关 键 词： | 矿产预测 机器学习 马坑式铁矿 |
| 收稿时间： | 2021-01-10 |
Machine learning for mineral prospectivity: A case study of iron-polymetallic mineral prospectivity in southwestern Fujian |
| |
| ZHANG Zhenjie,CHENG Qiuming,YANG Jie,WU Guopeng,GE Yunzhao.Machine learning for mineral prospectivity: A case study of iron-polymetallic mineral prospectivity in southwestern Fujian[J].Earth Science Frontiers,2021,28(3):221-235. |
| |
| Authors: | ZHANG Zhenjie CHENG Qiuming YANG Jie WU Guopeng GE Yunzhao |
| |
| Institution: | 1. School of Earth Sciences and Resources, China University of Geosciences(Beijing), Beijing 100083, China2. State Key Laboratory of Geological Processes and Mineral Resources, China University of Geosciences(Beijing), Beijing 100083, China3. Institute of Earth Sciences, China University of Geosciences(Beijing), Beijing 100083, China |
| |
| Abstract: | As a rapidly evolving technology in recent years, machine learning (ML) provides a novel approach for mineral prospecting (MP). In this paper, we discuss the progress on the methodology and theory of machine learning and summarize the applications of ML in MP in the areas of pattern recognition/information mining and information integration. We also point out the difficulties and challenges of ML in MP, such as data imbalance, lack of training data, lack of uncertainty evaluation in model selection, feedback feeding, and method selection. Here, we use mineral prospecting of the Makeng-type iron deposit in southwestern Fujian, China as an example to illustrate the process of using the ML method in MP. A complete prediction procedure should include (1) establishing a metallogenic model and identifying ore controlling factors by studying metallogenic systems; (2) building an exploration model and obtaining relevant data by researching exploration systems;(3) establishing a prediction model and extracting predictive factors by researching prediction evaluation systems;(4) obtaining metallogenic probability through information integration of predictive factors using ML models;(5) evaluating uncertainties of prediction performances and results; and (6) delineating prospecting/target areas and estimating resource reserves. Lastly, a future research roadmap for developing big-data based quantitative mineral prospecting theory and methods is proposed, guided by the geological big data and Earth system theory in following the research route of earth system-metallogenic system-exploration system-prediction evaluation system. |
| |
| Keywords: | mineral prospectivity machine learning Makeng-type iron deposit |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
| 点击此处可从《地学前缘》浏览原始摘要信息 |
| 点击此处可从《地学前缘》下载免费的PDF全文 |
|
