Scholarlink

<p>The growing demand for lithium (Li) resources in hybrid and electric vehicles drives the discovery of clay-type Li resources. The Li resources found in bauxite deposits represent promising new targets for Li exploration; however, the key factors controlling Li mineralization are still unclear. Besides, traditional exploration is time-consuming and labor-intensive. Here, we propose a data-driven approach to explore the Li mineralization mechanism and quickly identify Li-mineralized samples. Three classification models—K-nearest neighbors, random forests, and support vector machine—are constructed to efficiently classify mineralized and barren samples in the dataset, which is comprised of major compositions (Al<sub>2</sub>O<sub>3</sub>, SiO<sub>2</sub>, Fe<sub>2</sub>O<sub>3</sub>), Al/Si ratio, and mineralization information (mineralization age, Li contents of parent rocks, sea level, pCO<sub>2</sub>). The SMOTE algorithm is employed to resolve the imbalance problem between mineralized and barren samples, while the SHAP method is introduced to enhance the model’s interpretability. The three models achieve high prediction accuracies on the test set (92.64%, 91.59%, and 92.97%, respectively), suggesting their robust performance in classifying Li mineralization. Moreover, the data mining identifies the Al/Si ratio as the most significant feature, suggesting that mineral assemblage is the most crucial factor influencing Li enrichment. As high Al/Si ratios are negatively correlated with Li contents, Li may prefer to be enriched in clay minerals rather than Al-rich minerals. Therefore, although the Li-rich claystones show a close spatial relationship with bauxite, the Li enrichment process may be decoupled from the bauxitization process. Further correlation analyses indicate that clay minerals, including chlorite, kaolinite, illite, and pyrophyllite, may be potential host minerals for Li in the bauxites; however, the type of clay minerals may vary in different mineralization ages. Furthermore, an elevated Li content of parent rocks is considered another important contribution to Li enrichment. The environmental factors, such as pCO<sub>2</sub> and sea level, may positively impact Li enrichment by inhibiting the extent of bauxitization and promoting the preservation of Li-bearing clay minerals. This study highlights that data-driven machine learning models can unravel the genesis of lithium mineralization in bauxite deposits and rapidly evaluate the potential of lithium resources in previously discovered bauxite deposits.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Data-Driven Insights and Prediction of Bauxite-Hosted Lithium Mineralization

  • 期刊论文

  • 发表日期：

    2025年11月20日

查看全文
  • Xiaokang Liang,
  • Guotao Sun,
  • Yong Fu,
  • Bo Tang,
  • Peiwen Chen,
  • Zaiping Shi,
  • Shijiang Yang,
  • Guochen Zhou,
  • Zhonghong Huang,
  • Qiaolin Fu,
  • Tianyi Liu

摘要

The growing demand for lithium (Li) resources in hybrid and electric vehicles drives the discovery of clay-type Li resources. The Li resources found in bauxite deposits represent promising new targets for Li exploration; however, the key factors controlling Li mineralization are still unclear. Besides, traditional exploration is time-consuming and labor-intensive. Here, we propose a data-driven approach to explore the Li mineralization mechanism and quickly identify Li-mineralized samples. Three classification models—K-nearest neighbors, random forests, and support vector machine—are constructed to efficiently classify mineralized and barren samples in the dataset, which is comprised of major compositions (Al2O3, SiO2, Fe2O3), Al/Si ratio, and mineralization information (mineralization age, Li contents of parent rocks, sea level, pCO2). The SMOTE algorithm is employed to resolve the imbalance problem between mineralized and barren samples, while the SHAP method is introduced to enhance the model’s interpretability. The three models achieve high prediction accuracies on the test set (92.64%, 91.59%, and 92.97%, respectively), suggesting their robust performance in classifying Li mineralization. Moreover, the data mining identifies the Al/Si ratio as the most significant feature, suggesting that mineral assemblage is the most crucial factor influencing Li enrichment. As high Al/Si ratios are negatively correlated with Li contents, Li may prefer to be enriched in clay minerals rather than Al-rich minerals. Therefore, although the Li-rich claystones show a close spatial relationship with bauxite, the Li enrichment process may be decoupled from the bauxitization process. Further correlation analyses indicate that clay minerals, including chlorite, kaolinite, illite, and pyrophyllite, may be potential host minerals for Li in the bauxites; however, the type of clay minerals may vary in different mineralization ages. Furthermore, an elevated Li content of parent rocks is considered another important contribution to Li enrichment. The environmental factors, such as pCO2 and sea level, may positively impact Li enrichment by inhibiting the extent of bauxitization and promoting the preservation of Li-bearing clay minerals. This study highlights that data-driven machine learning models can unravel the genesis of lithium mineralization in bauxite deposits and rapidly evaluate the potential of lithium resources in previously discovered bauxite deposits.