|
| 基于无监督学习的地球化学异常的识别与应用——以内蒙古图古日格地区为例 |
| Identification and Application of Geochemical Anomalies Based on Unsupervised Learning: A Case Study of Tugurige Area, Inner Mongolia |
| 投稿时间：2025-04-30 修订日期：2025-08-01 |
| DOI： |
| 中文关键词: 地球化学异常 不确定性 无监督机器学习 狼山成矿带 图古日格 |
| 英文关键词:geochemical anomaly uncertainty unsupervised machine learning Langshan area Tugurige |
| 基金项目: |
|
| 摘要点击次数: 14 |
| 全文下载次数: 0 |
| 中文摘要: |
| 在矿产勘探过程中，识别矿化相关地球化学异常并降低不确定性是实现找矿突破的核心科学问题。本研究采用集成无监督机器学习算法，在构建稳健综合异常的同时实现不确定性的量化。以狼山成矿带北段图古日格地区为研究对象，利用水系沉积物地球化学数据，通过主成分分析和地理探测器确定与Au矿化相关的元素组合。采用单类支持向量机（OCSVM）、孤立森林（IF）和深度自编码器（DAE）三种无监督机器学习模型开展地球化学异常识别，基于模型平均方法和动态权重方法构建集成综合异常，并引入指数衰减函数对模型预测的不确定性进行定量评估。结果表明：1）研究区内与Au矿化相关联的元素组合为Au-Ag-Sb；2）通过动态权重方法确定的集成综合异常可以有效识别弱异常，并与已知矿床（点）空间套合度更高；3）经过指数衰减法计算的不确定性具有排除非矿异常的可靠性；4）圈定了四个成矿有利地段。 |
| 英文摘要: |
| Identifying mineralization-related geochemical anomalies and reducing uncertainties are core scientific issues for achieving breakthroughs in mineral exploration. This study employs integrated unsupervised machine learning algorithms to construct robust comprehensive anomalies while quantifying uncertainties. Taking the Tugurige area in the northern segment of the Langshan metallogenic belt as the research object, stream sediment geochemical data were used to determine element associations related to Au mineralization through principal component analysis (PCA) and geographical detector. Three unsupervised machine learning models—one-class support vector machine (OCSVM), isolation forest (IF), and deep autoencoder (DAE)—were applied to identify geochemical anomalies. Integrated comprehensive anomalies were constructed based on the model averaging method and dynamic weighting method, and an exponential decay function was introduced to quantitatively evaluate the uncertainty of model predictions. The results show that: 1) the element association related to Au mineralization in the study area is Au-Ag-Sb; 2) the integrated comprehensive anomalies determined by the dynamic weighting method can effectively identify weak anomalies and have a higher spatial coincidence with known ore deposits (occurrences); 3) the uncertainty calculated by the exponential decay method is reliable for excluding non-mineral anomalies; and 4) four prospective ore-prospecting areas have been delineated. |
| 查看/发表评论 下载PDF阅读器 |
| 关闭 |
