投稿时间：2024-05-18
修订日期：2024-10-31
网络发布日期：2025-03-10
中文摘要:进入地质大数据时代，如何深入挖掘与融合多源异构找矿空间大数据，成为当前矿产资源定量预测研究的热点。机器学习为提取和挖掘复杂数据中隐藏的难以识别的矿化信息和致矿异常信息的关联性，以及集成多源地学数据的致矿异常信息提供了有效工具。随机森林作为一种典型的机器学习算法，因其天然的并行特性、良好的模型可解释性、优秀的鲁棒性和泛化特性而被广泛应用于矿产资源预测。下雷-土湖是中国著名的以碳酸盐岩为容矿围岩的沉积型锰矿成矿区，区内产出中国首个超大型锰矿床——下雷锰矿，具有较大的找矿潜力。文章以下雷-土湖地区为研究对象，基于随机森林算法，深入挖掘Mn元素、沉积相、上泥盆统榴江组和五指山组出露、重力、航磁和向斜的空间分布特征及其与锰矿矿床的空间的耦合相关性，以及不同控矿要素之间的相关性，构建二维锰矿资源预测分类模型。在构建模型中，文章加入类权重参数，实现了正负样本的自动平衡。经过验证，该模型的袋外得分为0.998，表明该模型具有较好的泛化能力，且与逻辑回归和支持向量机相比，随机森林在研究区的应用效果更好。应用该模型对未知区进行找矿预测，圈定找矿远景区6处。
Abstract:In the era of geological big data, the in-depth exploration and integration of heterogeneous multi-source exploration spatial big data have become hot topics in current research on quantitative prediction of mineral resources. Machine learning provides effective tools for extracting and mining the correlation between difficult-to-identify mineralization information and mineralization anomaly information hidden in complex data, as well as integrating mineralization anomaly information from multiple sources of geoscience data. As a typical machine learning algorithm, random forest is widely used in mineral resource prediction due to its natural parallel characteristics, good model interpretability, excellent robustness and generalization characteristics. Xialei-Tuhu is a well-known sedimentary manganese mineralization area in China, with carbonate rock as the ore-hosting surroun-ding rock. The area produces China's first super-large manganese deposit—Xialei manganese deposit, which still has great potential for mineral exploration. The paper takes the Xialei-Tuhu area as the research object, and based on the random forest algorithm, deeply explores the spatial distribution characteristics of Mn elements, sedimentary facies, the outcrop of the Upper Devonian Liujiang Formation and Wuzhishan Formation, gravity, aeromagnetism, and syncline, as well as their coupling correlations with the manganese deposit in space. Furthermore, it explores the correlations between different ore-controlling factors to construct a two-dimensional manganese resource prediction and classification model. In the construction of the model, this article added class weight parameters to achieve automatic balancing of positive and negative samples. After verification, the out of bag score of the model is 0.998, indicating that the model has good generalization ability, and compared with logistic regression and support vector machine, the application effect of random forest in the study area is better. This model was used to predict mineralization in unknown areas, and 6 mineralization prospects were delineated.
文章编号：
中图分类号：
文献标志码：
基金项目:本文得到国家重点研发计划项目课题(编号：2022YFC2903404-02、2023YFC2906803)和地质调查二级项目“大数据智能找矿预测”(编号：DD20240004)联合资助
引用文本：
董建辉,刘欢,江沙,贾金典,娄德波,宋国玺,李婉悦.2025.基于随机森林的二维找矿预测——以下雷-土湖地区沉积型锰矿为例[J].矿床地质,44(1):143~158DONG JianHui,LIU Huan,JIANG Sha,JIA JinDian,LOU DeBo,SONG GuoXi,LI WanYue.2025.Two-dimensional mineral exploration prediction based on random forest: A case study of sedimentary manganese deposit in Xialei-Tuhu area[J].Mineral Deposits44(1):143~158
图/表
