Data-driven recursive kinetic modeling for chemical reactions

Kinetic models are crucial for analyzing reaction mechanisms and optimizing conditions. However, traditional models suffer from limitations such as lack of accuracy, narrow applicability, and difficulty in handling complex reaction conditions. Here, we develop a data-driven recursive kinetic model capable of predicting kinetic profiles from initial reaction conditions. The model captures reaction kinetics by leveraging recursive relationships between reactant or product concentrations at different times, which is learned through a machine learning algorithm, rather than traditional concentration-time equations. Moreover, we integrate a multiple estimation strategy into the model for performance enhancement. This model demonstrates superior accuracy, broad application scope, robustness, and few-shot learning capability on a simulated dataset including 18 chemical reaction types. Furthermore, its applicability to real-world chemical reactions is confirmed on the datasets of four practical reactions with complex kinetics. In addition, its strong capacity for mechanism interpretation and condition optimization is showcased using the experimental dataset of Fenton reaction. This work provides inspiration for the development of chemical kinetic models, potential to accelerate chemical research via advanced kinetic analysis.