中国科技核心期刊
美国化学文摘社（CAS）数据库
美国EBSCO学术数据库
日本科学技术振兴机构数据库(JST)
基于二维卷积神经网络的智能金矿找矿预测方法——以青海五龙沟地区为例
- 作者：
李金龙 1，李 华 2，薛林福 1*，丁 可 1，燕 群 1
- 作者单位：
- （1. 吉林大学地球科学学院；2. 青海省地质调查局）
基金项目：
- 青海省地勘资金科研项目（QDDJ/DKWX（SJZX）2023-08）
详细信息：
- 作者简介：
- 李金龙（2000—），男，硕士研究生，研究方向为智能找矿预测；E‑mail：2273947628@qq. com*
- PDF下载
Intelligent gold prospecting prediction based on 2D convolutional neural networks —A case study of the Wulonggou area, Qinghai
- English Author：
Li Jinlong1, Li Hua2, Xue Linfu1, Ding Ke1, Yan Qun1
- Unit：
- (1. College of Earth Sciences, Jilin University; 2. Qinghai Geological Survey)
- 摘要
- 在线预览
- 参考文献
摘要:
随着新一代人工智能技术的突破，深度学习为矿产资源预测提供了新的技术范式。传统找矿方法在处理海量多元异构地质数据时，会面临空间分布不均、非线性关系复杂、特征提取效率低等难题。以五龙沟地区为研究对象，提出一种基于CNN2D模型的智能找矿预测方法，旨在解决多源数据融合与非线性特征挖掘难题。该方法融合了地质、物探、化探 3种多元异构数据，设计并训练了二维卷积神经网络金矿找矿预测模型。结果表明：融合地质、化探、物探 3种数据的预测效果最优，模型的准确率较高；预测区面积占全区面积 10.13 %，圈定的 P03预测区、P05预测区、P07预测区具有良好找矿条件，可作为找矿靶区进一步布设探矿工程。通过野外调查并对比前人研究成果，认为预测结果符合成矿规律，具有良好的找矿潜力，进一步证明了本方法的有效性。本研究实现了CNN2D模型在高原复杂构造区的找矿应用，为深部矿产预测提供了可解释性强、泛化能力强的智能解决方案。
关键词:
二维卷积神经网络；智能找矿；找矿预测；相对属性网格化；数据增强；五龙沟地区；参数对比
Abstract:
With breakthroughs in next‑generation artificial intelligence technologies, deep learning has introduced anovel paradigm for mineral resource prediction. Traditional prospecting methods often encounter challenges when handlinglarge‑scale, heterogeneous geological datasets, such as uneven spatial distribution, complex nonlinear relationships,and low feature extraction efficiency. Taking the Wulonggou area as a case study, this paper proposes an intelligentgold prospecting prediction method based on the CNN2D model, aiming to address the integration of multi‑source dataand the mining of nonlinear features. The method integrates 3 types of heterogeneous data—geological, geophysical,and geochemical—and designs and trains a CNN2D model for gold prospecting prediction. Results show that the modelachieves the best prediction performance when all 3 data types are integrated, with high accuracy. The predicted areaaccounts for 10.13% of the total study area, with the delineated targets P03, P05, and P07 exhibiting favorable metallogenicconditions, making them viable targets for further exploration. Field investigations and comparison with previous researchindicate that the prediction results are consistent with known metallogenic patterns and demonstrate strong prospectingpotential, further validating the method’s effectiveness. This study marks application of the CNN2D model for prospectingin a complex tectonic region on the plateau, offering an interpretable and generalizable intelligent solution for deep mineralprediction.
Keywords:
2D convolutional neural network; intelligent prospecting; prospecting prediction; relative attributegridding; data augmentation; Wulonggou area; parameter comparison
