污染物
催化作用
可转让性
人体净化
多相催化
化学
可重用性
化学工程
环境化学
环境科学
水污染物
工作(物理)
水处理
经验模型
鉴定(生物学)
工艺工程
材料科学
联轴节(管道)
废物管理
萘普生
降级(电信)
作者
Sisi Wu,Ruixing Huang,Kunren Zhu,Xiaoling Zhang,Yuheng Huang,Jingrui Wang,Xiaoliu Huangfu,Hongxia Liu,Qiang He,Jun Ma
标识
DOI:10.1021/acs.est.6c04918
摘要
Abstract Heterogeneous Fenton-like catalysis offers an effective approach for organic pollutant removal, yet catalyst development remains largely empirical and lacks a closed-loop design-synthesis-validation framework. Here, we report a synthesis-oriented inverse-design framework for Fe-based catalysts. We first developed ML_Removal to predict pollutant removal, achieving a test-set R2 of 0.970, while SHAP analysis identified specific surface area (SSA) and Fe loading (Fe wt %) as actionable catalyst descriptors. We then built two synthesis-to-property models to predict SSA and Fe wt %, with SMOTER augmentation enhancing model performance and yielding test-set R2 of 0.782 and 0.856, respectively. Coupling these models with NSGA-II enabled the identification of candidate synthesis recipes targeting the SSA and Fe wt % windows. Two catalysts synthesized according to the NSGA-II-derived recipes exhibited measured SSA of 538.40 and 468.63 m2/g and Fe wt % of 8.19 and 7.01 wt %, aligning with the design targets. The catalysts achieved broad pollutant removal, retained activity in real waters matrices, and maintained >71% naproxen removal after five cycles. External validation of ML_Removal using the experimental data sets yielded an R2 of 0.784, demonstrating transferability to pollutants absent from the training set. Overall, this work establishes a closed-loop design–synthesis–validation framework for data-driven catalyst development and experimental realization.
科研通智能强力驱动
Strongly Powered by AbleSci AI