中国含铝岩系中稀土元素的赋存形式: 主量元素数据驱动的集成机器学习分析*

周锦涛; 余文超; 杜远生; 邓旭升; 翁申富; 雷志远

doi:10.7605/gdlxb.2025.045

古地理学报 ›› 2025, Vol. 27 ›› Issue (2) : 307 -320. DOI: 10.7605/gdlxb.2025.045

沉积矿产资源专题

中国含铝岩系中稀土元素的赋存形式: 主量元素数据驱动的集成机器学习分析^*

周锦涛 ¹^,² ,
余文超 ¹^,² ,
杜远生 ¹^,² ,
邓旭升 ²^,³ ,
翁申富 ²^,⁴ ,
雷志远 ²^,⁵

作者信息 +

Occurrence forms of rare earth elements in aluminum-bearing rock series in China: ensemble machine learning analysis driven by major element data

Jintao ZHOU ¹^,² ,
Wenchao YU ¹^,² ,
Yuansheng DU ¹^,² ,
Xusheng DENG ²^,³ ,
Shenfu WENG ²^,⁴ ,
Zhiyuan LEI ²^,⁵

Author information +

文章历史 +

PDF (5632K)

摘要

铝土矿及含铝岩系中的稀土元素是重要的潜在供应源,随着技术进步和市场需求增长,其勘探、开发和利用变得日益重要。目前,对含铝岩系中稀土元素的主要赋存形式和富集机制仍有争议,尤其是重稀土(HREE)。本研究汇编了中国含铝岩系的地球化学数据,利用机器学习方法预测轻稀土(LREE)和重稀土(HREE)含量,探讨其赋存形式。研究构建了随机森林(RF)、极限梯度提升(XGBoost)、支持向量机(SVM)和多层感知器(MLP)4种模型,并结合最优的2种模型构建集成模型,以提高预测性能。结果表明,集成模型在预测HREE和LREE含量方面表现更佳,能较好预测剖面尺度的稀土变化规律,并对高稀土含量剖面更敏感。特征重要性评估显示,P₂O₅和TFe₂O₃是预测HREE含量的关键变量,表明铁相矿物和磷酸盐矿物是HREE的主要载体。HREE可能通过类质同象替代和内环络合形式在磷酸盐矿物和铁相矿物中富集。含铝岩系中铁相矿物的存在是赋存HREE的基础,而磷酸盐矿物则是富集HREE的关键。对于LREE,P₂O₅的重要性显示含磷矿物是其重要赋存载体,即含磷稀土矿物为代表的独立稀土矿物是中国含铝岩系LREE的重要赋存形式。黏土矿物和铝矿物的离子吸附形式作为一种背景现象同样存在,但对稀土异常富集影响较小。研究建议,在找矿勘查中,针对LREE应关注P₂O₅含量较高的层位,含铝岩系的岩性应关注贫铁铝土矿和黏土质铝土矿。对于HREE,应首先关注TFe₂O₃含量较高的层位,若同时有较高的P₂O₅含量,则HREE富集可能性更大,建议关注的岩性包括富铁铝土矿和铝土质黏土岩。本研究建立的稀土含量预测模型在中国含铝岩系伴生稀土预测方面展现出潜力,需进一步验证、优化与应用。

Abstract

Bauxite and associated aluminum-bearing rock series,which host rare earth elements(REE),are considered significant potential sources of REE. With the advancement of new technologies and the increasing global market demand,the exploration,development,and utilization of REE associated with aluminum-bearing rock series have become increasingly critical. However,the dominant occurrence forms of REE in these rock series remain controversial. Moreover,most studies have primarily focused on Light Rare Earth Elements(LREE),while the occurrence forms and enrichment mechanisms of Heavy Rare Earth Elements(HREE)remain poorly understood. To address these issues,this study compiled geochemical data from Chinese aluminum-bearing rock series and attempted to utilize machine learning methods based on major elemental compositions to predict LREE and HREE contents,as well as to explore the occurrence forms of REE in these rock series. Four models were constructed for this purpose: Random Forest(RF),Extreme Gradient Boosting(XGBoost),Support Vector Machine(SVM),and Multi-Layer Perceptron(MLP). Additionally,two ensemble models were developed by combining the two best-performing models among the four. The results demonstrated that the ensemble models exhibited superior predictive performance for both HREE and LREE contents. The two ensemble models successfully predicted the variation of REE at the profile scale and delineated the REE enrichment horizons. These models demonstrated enhanced sensitivity to profiles with higher overall REE content,leading to improved predictive performance. Feature importance analysis revealed that P₂O₅ and TFe₂O₃ are critical variables for predicting HREE content,indicating that iron-bearing and phosphate minerals are primary carriers of HREE. The enrichment of HREE in phosphate minerals likely occurs via a substitution mechanism,while iron-bearing minerals enrich HREE through inner-sphere complexation. The presence of iron-bearing minerals in aluminum-bearing rock series is fundamental to HREE occurrence,and their absence may result in lower HREE content. Phosphate minerals play a crucial role in HREE enrichment,complementing iron-bearing minerals to facilitate high HREE content in aluminum-bearing rock series. The high importance of P₂O₅ in predicting LREE content suggests that phosphorus-bearing minerals are significant carriers of LREE. Independent REE minerals,particularly phosphorus-bearing REE minerals,represent an important form of LREE occurrence in Chinese aluminum-bearing rock series. While clay and aluminum minerals contribute to ion adsorption as a background phenomenon,they have minimal impact on abnormal REE enrichment. This study recommends focusing on strata with higher P₂O₅ content for potential LREE enrichment in practical mineral exploration. For aluminum-bearing rock series,attention should be given to low-iron bauxite and clayey bauxite. For HREE,strata with higher TFe₂O₃ content should be prioritized,especially when accompanied by elevated P₂O₅ levels. Lithologies suggested for attention include iron-rich bauxite and bauxitic claystone. In addition,the prediction model developed in this study shows potential in predicting REE associated with aluminum-bearing rock series in China,and is in urgent need of further optimization for validation and application.