In the field of catalyst design, machine learning is gaining significant attention, especially in situations where data is limited. Facing this challenge, we have developed a neural network model that enhances predictive accuracy through the careful selection of catalyst descriptors and feature engineering, aiming to predict the conversion rate and selectivity of the acetylene semi-hydrogenation reaction. Our model has identified promising metal oxide catalysts, such as CuO, ZnO, and V2O5, for the acetylene semi-hydrogenation reaction, and these predictions have been experimentally validated. Among them, CuO achieved an acetylene conversion of 99.6 % and an ethylene selectivity of 90 % at 50–100°C, which is unprecedented at the reaction space velocity we tested. This high activity at relatively low temperatures indicates a more promising industrial application. Looking ahead, machine learning will play a pivotal role in catalyst design, accelerating the discovery and industrial application of new materials.
Read full abstract