Data-Driven Recursive Kinetic Modeling for Fenton Reaction

The Fenton reaction is a widely used advanced oxidation process for water purification, valued for its simplicity and effectiveness in degrading refractory organic pollutants. However, accurately modeling its degradation kinetics remains challenging due to the complex reaction mechanism and strong sensitivity to multiple operational variables. Here, we develop a multiple estimation recursive machine learning (MERML) framework that is capable of predicting kinetic profiles from initial reaction conditions, including pollutant type, initial pollutant concentration, and Fenton reagent dosages. The MERML, without requiring prior mechanistic knowledge, captures recursive kinetic patterns directly from experimental data by establishing a data-driven mapping from preceding reaction states to future concentrations. This framework demonstrates superior performance in terms of accuracy, few-shot learning capability, robustness, interpretability, and application scope on an experimental data set from Fenton reactions of 12 phenolic compounds. Moreover, MERML enabled data-driven kinetic analysis, including reaction condition optimization, analysis of rate-influencing variables, and identification of representative kinetic patterns. This work provides a novel tool for pollutant degradation modeling, treatment optimization, and kinetic analysis of environmental reaction systems.