Biochar-based persulfate activation: Rate constant prediction, key variables identification, and system optimization

Abstract

Biochar has emerged as a promising sustainable catalyst for persulfate-based advanced oxidation processes (PS-AOPs), offering synergy between radical and non-radical pathways for superior mineralization. To streamline the traditionally time-consuming and costly biochar selection process, artificial intelligence was employed for a more efficient approach. In this study, an extreme gradient boosting (XGB) model was used to predict the pseudo-first-order kinetic rate constant (k) of a PS-AOPs. Important features were analyzed, and optimal feature values were obtained using particle swarm optimization (PSO). The dataset comprised 671 data points with 18 features extracted from peer-reviewed publications. To address data sparsity and imbalance, machine learning techniques and stratified data splits were utilized, achieving high model performance (R2 = 0.96, RMSE = 0.035). Feature importance analysis using XGB, permutation feature importance, and SHapley Additive exPlanations highlighted the importance of a higher biochar oxygen-to‑carbon ratio and encouraged metal modification. The presence of oxygen functional groups, particularly carbonyl and graphitized biochar structures, was emphasized in kinetic performance. PSO results recommended optimized biochar preparation parameters and operating conditions to enhance the k value, aligning with existing literature. This study predicts kinetic outcomes under preset conditions and guides practical biochar selection, demonstrating its relevance and potential in real-world applications.