1 引言
2 结果与讨论
2.1 数据分析
图2 由乙醇转化率、正丁醇选择性和正丁醇收率组成的散点图(收率随数值大小由不同颜色表示)Figure 2 A scatter plot of ethanol conversion rate, n-butanol selectivity, and n-butanol yield (where yield values are represented by different colors according to their magnitude) |
2.2 数据预处理和特征选取
图5 所有输入特征之间的相关性分析: (a)皮尔森相关性; (b)斯皮尔曼相关性Figure 5 Correlation analysis among all input features: (a) Pearson correlation; (b) Spearman correlation |
2.3 正丁醇合成催化剂的收率预测模型构建与模型性能评估
表1 不同机器学习算法回归结果对比Table 1 Comparison of regression results for different machine learning algorithms |
| ML model | R2-train | R2-test | RMSE(%)-train | RMSE(%)-test |
|---|---|---|---|---|
| LR | 0.350 | 0.373 | 7.39 | 7.62 |
| RR | 0.350 | 0.372 | 7.39 | 7.62 |
| LLR | 0.311 | 0.307 | 7.61 | 8.01 |
| KNN | 0.856 | 0.753 | 3.48 | 4.78 |
| DTR | 0.851 | 0.707 | 3.54 | 5.20 |
| RFR | 0.942 | 0.854 | 2.21 | 3.68 |
| ETR | 0.987 | 0.868 | 1.04 | 3.50 |
| GBDT | 0.975 | 0.869 | 1.46 | 3.48 |
| XGBR | 0.986 | 0.897 | 1.07 | 3.09 |
图7 不同机器学习模型的验证结果. LR: 线性回归; RR: 岭回归; LLR: 套索回归; KNN: K近邻回归; DTR: 决策树回归; RFR: 随机森林回归; ETR: 极端随机树回归; GBDT: 梯度提升回归; XGBR: 极端梯度提升回归Figure 7 Validation results of different machine learning models. LR: Linear Regression; RR: Ridge Regression; LLR: Lasso Regression; KNN: K-Nearest Neighbor; DTR: Decision Tree Regression; RFR: Random Forest Regression; ETR: Extremely-randomized Trees Regression; GBDT: Gradient Boosting Decision Tree; XGBR: eXtreme Gradient Boosting Regression |
2.4 模型可解释性分析
2.4.1 催化反应条件贡献分析
2.4.2 催化剂特征贡献分析
2.4.3 金属组分特征贡献分析
2.5 高性能正丁醇合成催化剂的预测
表2 高性能正丁醇合成候选催化剂的模型预测结果Table 2 Model prediction results for candidate catalysts for high-performance n-butanol synthesis |
| No. | Metal | Support | Reaction temperature/℃ | Butanol yield/% |
|---|---|---|---|---|
| 1 | Au | MgAlOx | 334 | 57.1 |
| 2 | Ag | NaX | 322 | 53.9 |
| 3 | Cu | NC | 344 | 51.1 |
| 4 | Ca | NC | 332 | 48.7 |
| 5 | Be | NC | 327 | 46.9 |