| 艾启兴,邓浩,毛先成,宋嘉璇,邓春芳,段新明,张展悦,张镜,邓选伦.基于深度条件随机场的金川铜镍硫化物矿床三维成矿预测[J].矿产勘查,2026,17(8):1333-1344 |
| 基于深度条件随机场的金川铜镍硫化物矿床三维成矿预测 |
| Three-dimensional mineral prospectivity modeling for the Jinchuan Cu-Ni sulfide deposit using deep conditional random fields |
| 投稿时间：2026-07-16 |
| DOI：10.20008/j.kckc.202608004 |
| 中文关键词: 三维成矿预测 空间相关性 深度条件随机场模型 金川铜镍硫化物矿床 深部矿体资源 |
| 英文关键词: three-dimensional mineral prospectivity modeling spatial correlation deep conditional randomfields Jinchuan Ni-Cu sulfide deposit deep mineral resources |
| 基金项目:本文受深地国家科技重大专项项目(2025ZD1007102)、国家自然科学基金项目(42472363、42272344)和湖南省科技创新计划(2021RC4055)联合资助。 |
|
| 摘要点击次数: 51 |
| 全文下载次数: 21 |
| 中文摘要: |
| 金川铜镍硫化物矿床是世界级岩浆硫化物矿床，经济与战略价值极高，但浅部资源日趋减少，深部隐伏矿体预测勘查迫在眉睫。传统三维成矿预测方法，由于采用成矿信息体元化表达，而忽略成矿地质体、矿化分布的空间连续性与关联性，建立的预测模型存在成矿规律空间连续性特征难以保真的问题。针对该问题，本文提出开展基于深度条件随机场(DCRF)的金川矿床深部矿体三维智能预测研究。该方法融合概率图模型内置势函数约束，将三维体元单元之间的空间平稳特征、矿化作用协同演化特征统一纳入全域推算流程，能够有效避免仅依靠单一体元独立建模预测带来的连续性保真缺陷，实现局部多元地质指标与全域空间关联特征的深度融合。 DCRF模型的线下面积(AUC)达到 0.917，明显优于支持向量机(SVM)、随机森林(RF)及多层感知机(MLP)等传统机器学习模型。基于该模型，在金川矿床的深边部成功圈定了 4处高潜力找矿靶区，为金川矿床深边部工程靶区优选提供了定量参考。本文提出的方法显著提升了矿山深边部隐伏矿体立体定位预测的客观性与准确性，为同类复杂岩浆硫化物矿床隐伏矿体精细化立体定位预测提供了可行的参考方案。 |
| 英文摘要: |
| The Jinchuan Cu-Ni sulfide deposit is a world-class magmatic sulfide deposit with exceptionallyhigh economic and strategic significance. However, its shallow resources are gradually depleting, making the predic.tion and exploration of deep concealed ore bodies an urgent task. Traditional three-dimensional (3D) metallogenicprediction methods rely on voxelized representation of mineralization information, which disregards the spatial conti.nuity and correlation of ore-forming geological bodies and mineralization distribution. Consequently, the predictionmodels constructed suffer from difficulties in preserving the spatial continuity characteristics of metallogenic laws.To address this issue, this study proposes a 3D intelligent prediction approach for deep ore bodies in the Jinchuandeposit based on deep conditional random fields (DCRF). By integrating the inherent potential function constraintsof probabilistic graphical models, this method incorporates the spatial stationarity among 3D voxels and the co-evolutionary characteristics of mineralization processes into a global inference framework. It effectively avoids theloss of continuity caused by independent modeling of single voxels, enabling deep fusion of local multi-source geo.logical indicators and global spatial correlation features. The DCRF model achieves an area under the curve (AUC)value of 0.917, significantly outperforming traditional machine learning models such as support vector machines(SVM), random forests (RF), and multilayer perceptrons (MLP). Using this model, four high-potential prospectingtargets were successfully delineated in the deep and peripheral parts of the Jinchuan deposit, providing a quantita.tive reference for the optimization of engineering targets in these regions. The method proposed in this paper signifi.cantly improves the objectivity and accuracy of 3D positioning prediction for concealed ore bodies in deep and pe.ripheral parts of mines, and offers a feasible solution for the refined 3D positioning prediction of concealed ore bod.ies in other complex magmatic sulfide deposits. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|