Scholarlink

<p>This study introduces a machine learning approach to address the challenges in the rational design of photocatalysts and the optimization of process parameters for uranium extraction. A Particle Swarm Optimization (PSO)-enhanced XGBoost regression model is developed to elucidate the synergistic effects of the intrinsic properties of g-C₃N₄-based materials and reaction conditions on extraction performance. The model achieves high predictive accuracy with a coefficient of determination (R<sup>2</sup>) of 0.95 on both training and testing datasets. By leveraging SHAP-based interpretability analysis, the study transcends mere prediction to conduct in-depth process analysis. Initial uranium concentration, catalyst dosage, and adsorption ratio are identified as the most influential features. Crucially, this interpretable framework delineates optimal operational windows for enhanced performance, including a valence band above 1.85&#xa0;eV, a conduction band above -1.0&#xa0;eV, a specific surface area of 38–126 m<sup>2</sup>/g, an average pore width of 15–21&#xa0;nm, an adsorption ratio below 15%, a pH range of 4.3–6.1, and a catalyst dosage below 0.4&#xa0;g/L. Furthermore, the model exhibits robust generalization in extrapolation tests on non-g-C₃N₄ systems, highlighting its broad applicability. This work provides an interpretable roadmap for catalyst screening and process engineering, underscoring the transformative potential of machine learning in advancing nuclear environmental remediation technologies.</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

PSO-Optimized Machine Learning Accelerates the Discovery of g-C₃N₄-Based Photocatalysts for Enhanced Uranium Extraction

  • 期刊论文

  • 发表日期：

    2026年4月7日

查看全文
  • Sen Kang,
  • Yi Duan,
  • Jiahang Zhang,
  • Youchen Zheng,
  • Miaobin Ye,
  • Shuai Zhou

摘要

This study introduces a machine learning approach to address the challenges in the rational design of photocatalysts and the optimization of process parameters for uranium extraction. A Particle Swarm Optimization (PSO)-enhanced XGBoost regression model is developed to elucidate the synergistic effects of the intrinsic properties of g-C₃N₄-based materials and reaction conditions on extraction performance. The model achieves high predictive accuracy with a coefficient of determination (R2) of 0.95 on both training and testing datasets. By leveraging SHAP-based interpretability analysis, the study transcends mere prediction to conduct in-depth process analysis. Initial uranium concentration, catalyst dosage, and adsorption ratio are identified as the most influential features. Crucially, this interpretable framework delineates optimal operational windows for enhanced performance, including a valence band above 1.85 eV, a conduction band above -1.0 eV, a specific surface area of 38–126 m2/g, an average pore width of 15–21 nm, an adsorption ratio below 15%, a pH range of 4.3–6.1, and a catalyst dosage below 0.4 g/L. Furthermore, the model exhibits robust generalization in extrapolation tests on non-g-C₃N₄ systems, highlighting its broad applicability. This work provides an interpretable roadmap for catalyst screening and process engineering, underscoring the transformative potential of machine learning in advancing nuclear environmental remediation technologies.