| 张锦,徐京城.机器学习辅助的过渡金属掺杂MoS2材料结构筛选：从单原子到双原子掺杂[J].有色金属材料与工程,2025,46(6):18-23. |
| 机器学习辅助的过渡金属掺杂MoS2材料结构筛选：从单原子到双原子掺杂 |
| Machine learning assisted material structure screening for transition metal doped MoS2: from single-atom to double-atom doping |
| |
| DOI：10.13258/j.cnki.nmme.20250219001 |
| 中文关键词: 机器学习 金属掺杂 气体吸附 MoS2 |
| 英文关键词:machine learning metal doping gas adsorption MoS2 |
| 基金项目:东方科软（北京）科技有限公司横向项目（H-2023-369-039） |
|
| 摘要点击次数: 600 |
| 全文下载次数: 250 |
| 中文摘要: |
| 为了加快对传感材料结构的设计及气敏性能调控，使用高通量计算结合机器学习的方法研究过渡金属掺杂MoS2气敏材料与吸附气体之间的构效关系。通过构建单原子和双原子过渡金属掺杂MoS2体系（TM-MoS2、TM2-MoS2，TM=V、Mn、Co、Mo、Ru），基于第一性原理计算了3种典型的室内气体污染物HCHO、NH3和H2S的吸附能，并采用4种算法构建了机器学习回归预测模型。使用逐步滤波的特征工程策略，筛选出合理的特征集合提高了机器学习模型的泛化能力。结果表明，XGBoost算法最佳，其在训练集和测试集上的R2分别为0.91和0.78，RMSE值分别为0.15、0.34 eV。使用SHAP分析各特征对回归模型的贡献，其中Ng和$ \varepsilon $d的贡献值最大，特征Dg的贡献最小。 |
| 英文摘要: |
| In order to accelerate the design of sensing material structures and control of gas sensing performance, high-throughput computing combined with machine learning was used to study the structure-activity relationship between transition metal doped MoS2 gas sensing materials and adsorbed gases. By constructing single atom and double atom transition metal doped MoS2 systems (TM-MoS2, TM2-MoS2, TM=V, Mn, Co, Mo, Ru), the adsorption energies of three typical indoor gas pollutants HCHO, NH3, and H2S were calculated based on first principles, and four algorithms were used to construct machine learning regression prediction models.The use of a stepwise filtering feature engineering strategy to select a reasonable set of features has improved the generalization ability of the machine learning model. The results indicate that the XGBoost algorithm is the best, with R2 values of 0.91 and 0.78 on the training and testing sets, and RMSE values of 0.15 and 0.34 eV, respectively. Use SHAP to analyze the contribution of each feature to the regression model, where Ng and $ \varepsilon $d have the highest contribution values, and feature Dg has the smallest contribution. |
| HTML 查看全文 查看/发表评论 下载PDF阅读器 |
