Glycolic Acid Production from Glycerol via Catalytic Oxidation: Machine Learning-Accelerated Multi-Objective Process Optimization

Glycerol derived from the recycling of used cooking oils serves as a renewable and sustainable feedstock for producing glycolic acid (GLYOA), a key monomer for biodegradable polymers, via Mn2O3-catalyzed selective oxidation. While this transformation provides an environmentally friendly alternative to fossil-derived routes, optimizing GLYOA selectivity remains challenging due to complex competing oxidation pathways. Traditional kinetic-based multiobjective optimization can predict reaction behavior but suffers from high computational cost and extensive validation requirements, limiting its scalability. To address these limitations, we propose for the first time a hybrid framework that integrates mechanistic kinetics modeling with a deep residual shrinkage network (DRSN) to accelerate green catalytic optimization. A kinetic model was developed to simulate product distributions under various conditions, generating a data set for training the DRSN surrogate. After structural tuning, the DRSN reduced optimization time by 38.2% compared to direct kinetic-based methods, while maintaining high predictive accuracy. The optimized model was incorporated into an NSGA-II algorithm to identify conditions that maximize GLYOA yield while minimizing CO2 formation. This study demonstrates a scalable, data-driven strategy for optimizing biomass-derived catalytic processes and highlights the potential of machine learning–assisted modeling to advance the sustainable valorization of waste-derived glycerol.