Preparation of Iron-Based Fenton-Like Catalysts Using Sufficient Pyrolysis Method: Rapid Evaluation of Performance, Prediction of Performance by Machine Learning, and Characterization

Abstract

Preparation of Fenton sludge into magnetic biochar is a promising direction for resource utilization. This study proposed iron-based Fenton-like catalysts prepared by sufficient pyrolysis with catalyst performance evaluation based on the volatile matter and low-valent iron content. Additionally, machine learning was further applied to predict the catalyst performance and then analyze key factors. Results showed that the FSMS2:1-800 catalyst could remove 82% of TOC and 98% of chroma in treating real MBR effluent of landfill leachate. EDS and XRD analysis results confirmed a large amount of zero-valent iron in the catalyst. Furthermore, the magnetic saturation value of this catalyst was measured to be 54.89 emu/g by VSM, indicating that it could be easily recycled. The volatile matter and low-valent iron content of each catalyst were linearly correlated to the TOC/chroma removal rate with an R2 approximately 0.83, suggesting that the catalyst performance could be roughly assessed. The SVR and XGBoost models were used to train and predict the Fenton-like experimental data, with the XGBoost model proving to be more appropriate. The R2 and RMSE values of the XGBoost model for predicting the chroma removal rate reached 0.999 and 0.132, respectively, while those for predicting the TOC removal rate reached 0.999 and 0.149. Finally, the marginal contribution of the input features to the predicted results of the XGBoost model was quantified and visualized using the Shapley Additive exPlanations Plot, revealing that pyrolysis temperature was the most important factor.

Keywords: Fenton-like catalyst, sufficient pyrolysis method, Machine learning, rapid assessment