• captcha

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)

期刊导读

基于SMOTified-BRF模型的地球化学金矿化异常预测研究——以陕西汉滨—旬阳地区为例

  • 作者:
  • 徐正林|王晰|薛林福|冉祥金|燕群|?钣朗于晓飞
  • 作者单位:
  • 吉林大学地球科学学院|吉林大学地球科学学院|吉林大学地球科学学院|吉林大学地球科学学院|吉林大学地球科学学院|中国地质调查局发展研究中心&自然资源部矿产勘查技术指导中心|中国地质调查局发展研究中心&自然资源部矿产勘查技术指导中心
  • 基金项目:

  • 中国地质调查局矿调项目(DD20190159)
  • 详细信息:

  • 作者简介:
  • 徐正林(2000—),男,硕士研究生,研究方向为人工智能找矿预测、人工智能油气应用等;E-mail:2765018131@qq.com
  • 通讯作者:
  • 薛林福(1962—),男,教授,博士,从事地学数据分析、三维地质建模与地质过程模拟等方面的研究工作;E-mail:xuelf@jlu.edu.cn
  • 此文档没有上传!

Research on geochemical gold mineralization anomaly prediction based on the SMOTified-BRF model-A case study of the Hanbin-Xunyang area, Shaanxi Province

  • English Author:
  • College of Earth Sciences,Jilin University|College of Earth Sciences,Jilin University|College of Earth Sciences,Jilin University|College of Earth Sciences,Jilin University|College of Earth Sciences,Jilin University|Development and Research Center,China Geological Survey&Mineral Exploration Technology Guidance Center,Ministry of Natural Resources|Development and Research Center,China Geological Survey&Mineral Exploration Technology Guidance Center,Ministry of Natural Resources
  • Unit:
  • 摘要
  • 在线预览
  • 参考文献

摘要:

矿产勘查既是新质生产力的重要组成,又是推动新质生产力发展的重要力量。为了改善矿产勘查中金矿化预测面临的由于已知矿化样本数量少导致的样本类不平衡问题及矿化信息稀缺问题,提出SMOTified-BRF模型,该模型使用SMOTE方法对极少数已知矿化样本进行数量增强并使用平衡随机森林方法进行预测。以汉滨—旬阳地区为研究区,对水系沉积物地球化学数据分别使用SMOTified-BRF模型和BRF模型进行金矿化预测效果和模型性能对比研究。研究结果表明:SMOTified-BRF模型的AUC值(0.987 5)高于BRF模型的AUC值(0.972 6),且在约登指数指示的最优阈值下SMOTified-BRF模型预测的矿化面积占比(1.95 %)相较于BRF模型(12.23 %)更小,说明SMOTified-BRF模型相比于BRF模型具有更准确高效的金矿化异常预测性能。

关键词:

金矿化;异常;预测;平衡随机森林法;SMOTE;地球化学;汉滨—旬阳地区

Abstract:

Mineral exploration is not only a critical component of new quality productive forces but also a driving force for their development.To address the challenges of sample imbalance caused by the limited number of known mineralization samples and the scarcity of mineralization information in gold mineralization prediction during mineral exploration,this study proposes the SMOTified-BRF model.This model applies the SMOTE method to augment the extremely limited known mineralization samples and employs the balanced random forest (BRF) method for prediction.Using the Hanbin-Xunyang area as the study area,geochemical data from stream sediments were analyzed using the SMOTified-BRF and BRF models and compared the gold mineralization prediction outcome and model performance.The results show that the SMOTified-BRF model achieves a higher AUC value (0.987 5) compared to the BRF model (0.972 6).Additionally,at the optimal threshold indicated by the Youden index,the predicted mineralized area ratio of the SMOTified-BRF model (1.95 %) is significantly smaller than that of the BRF model (12.23 %),demonstrating that the SMOTified-BRF model offers more accurate and efficient performance in predicting gold mineralization anomalies.

Keywords:

gold mineralization;anomaly;prediction;balanced random forest;SMOTE;geochemistry;Hanbin-Xunyang area