Scholarlink

Transition metal nitrides, phosphides, sulfides, and selenides are potential alternatives to platinum group metal catalysts in the hydrogen evolution reactionHydrogen evolution reaction (HER). Moreover, doping with transition metal atoms is expected to further enhance catalytic performance. In this work, we propose a machine learningMachine learning (ML) technique for predicting the HER activity of transition metal-doped transition metal nitrides, phosphides, sulfides, and selenides. To achieve this, we establish a multi-step workflow utilizing tools from the ML algorithm toolbox tailored to the specific context of our study, aiming to build a well-trained data-driven model. This model is designed to predict the HER activity of 360 different transition metal-doped systems across these materials. One-third of these materials (120 systems) were randomly selected for initial evaluation using density functional theoryDensity functional theory (DFT) calculations to assess HER performance. Subsequently, through feature importance analysis, correlation analysis, and Recursive Feature Elimination (RFE), we identify highly correlated and low-importance features. Finally, ensemble and individual models are trained and tested. The results indicate that the RF model achieves a mean absolute error (MAE) of 0.1306 and a Root Mean Square Error (RMSE) of 0.1738.

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

Machine Learning for the Efficient Identification of High-Performance Metal-Doped Transition Metal Compounds for Hydrogen Evolution Catalysis

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Lu Xue,
  • Jie Dang

摘要

Transition metal nitrides, phosphides, sulfides, and selenides are potential alternatives to platinum group metal catalysts in the hydrogen evolution reactionHydrogen evolution reaction (HER). Moreover, doping with transition metal atoms is expected to further enhance catalytic performance. In this work, we propose a machine learningMachine learning (ML) technique for predicting the HER activity of transition metal-doped transition metal nitrides, phosphides, sulfides, and selenides. To achieve this, we establish a multi-step workflow utilizing tools from the ML algorithm toolbox tailored to the specific context of our study, aiming to build a well-trained data-driven model. This model is designed to predict the HER activity of 360 different transition metal-doped systems across these materials. One-third of these materials (120 systems) were randomly selected for initial evaluation using density functional theoryDensity functional theory (DFT) calculations to assess HER performance. Subsequently, through feature importance analysis, correlation analysis, and Recursive Feature Elimination (RFE), we identify highly correlated and low-importance features. Finally, ensemble and individual models are trained and tested. The results indicate that the RF model achieves a mean absolute error (MAE) of 0.1306 and a Root Mean Square Error (RMSE) of 0.1738.