| 摘要点击次数: 6 全文下载次数: 0 查看全文 查看/发表评论 下载PDF阅读器 |
| 基金项目：国家自然科学基金青年项目“隐伏矿体三维预测不确定性层级传播的层次贝叶斯建模”（42302338），湖南省大学生创新项目“基于MCMC随机模拟的三维地质建模不确定性分析”（S202211527012） |
|
| 摘要： |
| 以新疆西天山阿希金矿床为例，基于断层、管道相和蚀变带的三维地质模型，结合三维空间分析方法，提取控矿地质因素特征值，利用随机森林方法开展三维成矿预测。基于OOB误差最小化原则进行随机森林预测模型参数优化，确定最优模型参数为决策树数量K=550、节点分裂最大特征数M=4。随机森林在分类准确率（ACC=0.947）、AUC值（0.984）、灵敏度（TPR=0.979）及F1分数（F1=0.966）等方面均表现优异，远超AdaBoost、支持向量机等模型，特异度虽略低，但符合矿产勘探“宁可错判、不可漏判”的原则，整体分类性能最优；wrF（断层形态起伏）、dAlt（蚀变带距离场）等控矿地质因素贡献度得分高于其他因子，找矿指示作用更强，基于随机森林预测结果圈定了2处高潜力找矿靶区。研究成果可为阿希金矿的深部矿产勘查提供数据驱动的技术框架，并为未来深部找矿提供参考。 |
| 关键词：随机森林 三维成矿预测 机器学习 阿希金矿床 新疆西天山 |
|
| Abstract: |
| Taking the Axi gold deposit in the western Tianshan Mountains of Xinjiang as an example, this study conducted 3D mineralizationprediction using the random forest method. The research was based on a 3D geological model of faults, pipe facies, and alteration zones, combined with 3D spatial analysis methods to extract characteristic values of ore-controlling geological factors. Parameter optimization of the random forest prediction model was performed following the principle of minimizing Out-of-Bag (OOB) error, and the optimal model parameters were determined as follows: number of decision trees (K) = 550, and maximum number of features for node splitting (M) = 4. The random forest model exhibited excellent performance in terms of classification accuracy (ACC = 0.947), AUC value (0.984), sensitivity (TPR = 0.979), and F1 score (F1= 0.966), which were far superior to those of models such as AdaBoost and Support Vector Machine (SVM). This is consistent with the principle of ″preferring misjudgment over missed judgment″ in mineral exploration, though its specificity was slightly lower, making it the optimal model in terms of overall classification performance. Ore-controlling geological factors such as wrF (fault morphological relief) and dAlt (alteration zone distance field) obtained higher contribution scores than other factors, indicating a stronger role in prospecting indication. Based on the random forest prediction results, two high-potential prospecting target areas were delineated. The research results can provide a data-driven technical framework for deep mineral exploration of the Axi gold deposit and serve as a reference for future deep prospecting. |
| Keywords:random forest 3D mineralization prediction machine learning Axi gold deposit western Tianshan Mountains of Xinjiang |
| 陈霖泓,张 维,高雨杰,等.基于随机森林的新疆西天山阿希金矿床三维成矿预测[J].地质学刊,2025,49(4):417-424 |
|
|
|
|
|