A method for optimizing catalyst preparation conditions based on machine learning and genetic algorithm

Multiphase catalytic technology is an effective method for treating exhaust gases, but the traditional method for optimizing catalyst preparation conditions is inefficient and struggles to achieve multi-dimensional optimization. Aiming at the problems of low efficiency of experimental optimization and tend to fall into local optimum in the development of multiphase catalytic oxidation catalysts, we propose an intelligent optimization strategy integrating machine learning and genetic algorithm (GA). Taking Mn-Ca/γ-Al2O₃ as the research object, the effects of catalyst loading, loading ratio, calcination temperature and time on the degradation of xylene were systematically investigated, and a high-quality dataset comprising 225 sets of experimental data was established. Algorithms such as Random Forest (RF) and Gradient Boosting Tree (GBR) were used to establish a removal rate prediction model, which combined with SHAP values to resolve the key influencing factors, and we used GA to search for the optimal preparation parameters. The results showed that the RF model had the best prediction accuracy (R²=0.949, MAE=0.022 for the test set) and the characteristic importance showed that calcination temperature (43.9%) and load (21.96%) were the key control parameters. The best conditions obtained from GA optimization (5.4% loading, Ca: Mn=1:4, calcination temperature 423°C, time 3.75h) resulted in 93.7% xylene removal, which is a 2.5% enhancement over the conventional experimental optimization. This study provides a new paradigm for green and low-carbon development of catalysts.