Design and Operation of Copper-Based Heterogeneous Catalysts for CO2 Conversion to Methanol Using Extreme Gradient Boosting

Copper-based catalysts are widely utilized as active metals in the hydrogenation of CO2 to methanol due to their high activity and selectivity. However, optimizing the catalyst design (i.e., material design) and operating conditions for such systems remains challenging; as a result, the optimization processes may potentially generate more CO2 than it reduces. To mitigate this issue, machine learning has emerged as a powerful tool, offering intelligent prediction capabilities and insightful analysis of catalyst systems. In this approach, key parameters were classified into two primary categories: catalyst composition and operating conditions. A comprehensive data set was compiled from the literature (698 data points), covering a range of Cu-based catalysts with various supports, promoters, and catalyst surface areas, as well as reaction conditions. This data set was used to train 12 different machine learning models, including linear models, tree-based models, and others, with hyperparameter tuning conducted via GridSearch and 10-fold cross-validation. Among these, the Extreme Gradient Boosting (XGBoost) model achieved the highest performance, with an R2 of 0.89 and a mean absolute error (MAE) of 0.131. The feature importance was analyzed using SHAP (Shapley Additive Explanations) and further supported by FPI (Feature Permutation Importance) and PDP (Partial Dependence Plot) analysis. The results indicated that approximately 60% of the influence on STY (space-time yield) is attributed to reaction conditions, while 40% relates to catalyst composition. The ranked importance of features affecting STY is determined as follows: GHSV > Cu loading > SBET > temperature > pressure > H2/CO2 ratio > promoter > support. This approach, leveraging the XGBoost model, provided a systematic and data-driven framework for understanding and optimizing Cu-based catalysts in a heterogeneous system design.