High-Efficiency Catalyst Screening Driven by Machine Learning for the Oxidation of 5-Hydroxymethylfurfural to 2, 5-Furandicarboxylic Acid

This study focuses on the catalytic reaction system for the oxidation of 5-hydroxymethylfurfural (HMF) to prepare 2, 5-furanediformic acid (FDCA). This process is regarded as an important transformation path in the field of green chemical engineering due to its high atomic economy and renewable raw materials. Aiming at the problems of relying on trial-and-error experiments, long cycles, and high cost in the development of traditional catalysts, this study proposes a machine learning-driven intelligent optimization strategy. The data set was preliminarily classified through K-means clustering, and the problems of data imbalance and dimension mismatch were solved by adopting the composite minority oversampling technique and the adaptive composite sampling technique, significantly improving the generalization performance of the machine learning model. In the machine learning model optimized by hyperparameters, the neural network shows optimal prediction performance. The interpretability analysis of SHapley Additive exPlanations (SHAP) revealed that the alkali-free conditions and reaction time had a significant impact on the synthesis efficiency of FDCA. Eventually, the promising Ru/Mn6Ce1OY catalyst was screened out based on the multiobjective optimization framework of the genetic algorithm.