Smart paradigm to predict copper surface area of Cu/ZnO/Al2O3 catalyst based on synthesis parameters

Abstract

Cu/ZnO/Al2O3 catalyst is used in processes of water-gas shift, methanol steam reforming, and methanol synthesis in the industry. According to various experimental studies, the catalytic activity of this catalyst is directly proportional to its copper surface area (Cu(SA)). In this study, a machine learning approach for predicting Cu(SA) ranges in three classes (low, medium, and high) is introduced based on catalyst preparation factors. Three models of random forest (RF), support vector machine (SVM), and multilayer perceptron artificial neural network (MLP-ANN) classifiers are developed and optimized using grid search 10-fold cross-validation for a 188 sample dataset extracted from 45 experimental studies. It is found that the RF classifier with 90% cross-validation accuracy score and 94.7% test data prediction accuracy score outperforms the other two models. The SHAP (or SHapley Additive exPlanations) analysis is performed to investigate the effects of synthesis factors, such as aging conditions, precipitant type, and pH on Cu(SA). It is concluded that Cu/Zn ratio has the greatest influence on Cu(SA). The optimum synthesis conditions yielding high Cu(SA) are also discovered, which is of great importance for synthesis of Cu/ZnO/Al2O3 catalysts with high catalytic activity.