Deciphering N-Doped Biochar Design for Non-Radical Pathways through Hierarchical Machine Learning

Abstract

Biochar has been widely employed for the promotion of advanced oxidation processes (AOPs) and when combined with nitrogen doping for charge distribution mediation, N-doped biochar (NBC) can serve as a highly effective catalyst for the degradation of persistent organic pollutants. However, due to the variety of doping and preparation methods, the intrinsic active sites for AOP catalysis have not been clearly identified. Furthermore, the complex relationships between preparation method, material properties, and catalytic degradation pathways remain unclear, impeding the widespread practical application of NBC. Herein, machine learning (ML) was implemented to predict the degradation pathway and identify the vital properties of N-doping required for the acceleration of AOPs. During the process of model training, an innovative method of data set splitting was applied, comparing the results generated from multiple models to enhance model interpretability. We elucidated the correlation between the primary features and nonradical pathway, focusing on the contribution of N species and the regulatory role of pyrolysis temperature. Detailed insights were further provided to enhance the ratio design of NBC for nonradical mediation. Overall, this study offers novel insights into NBC-mediated AOPs for pollution control, underscoring the significant potential of ML for accelerating catalyst applications.