已核实 DOI 相关产品: 0

A predictive framework for nanofiltration membrane selection to control disinfection by-products in drinking water treatment

Journal 2026 · IF 12.5 · 发表于 2026 年 7 月 26 日
作者: Yongkun Wang,Feng Li,Yuting Di,Mengru Yao,Deyin Hou
阅读全文 在 PubMed 查看

摘要

Predictive framework enables site-specific membrane selection for DBP control • SHAP analysis reveals species-specific pathways converge on precursor passage • Machine learning links membrane properties to DBP formation outcomes • Integrated framework balances rejection performance against energy consumption Controlling disinfection by-product (DBP) formation during drinking water treatment relies on effective removal of their precursors, particularly the low-molecular-weight natural organic matter (LMW-NOM) fraction that conventional processes struggle to remove. Although nanofiltration (NF) membranes can solve this problem effectively, selecting the appropriate membrane involves trade-offs between precursor rejection and energy consumption which depends on multiple factors. Here we report systematic evaluation of four commercial NF membranes with molecular weight cut-offs from 200 to 550 Da using representative NOM compounds spanning 133–620 Da. Cross-flow nanofiltration experiments were combined with chlorination tests at multiple disinfectant doses, and the resulting dataset was augmented with literature data to train XGBoost machine learning models with SHAP interpretability analysis. Size exclusion dominated rejection behavior, with the tightest NF90 membrane achieving >96% TOC removal and > 98% UV254 reduction and correspondingly lowest DBP yields upon chlorination (TCM reduced by 85–92%, DCAA by 78–88%, TCAA by 82–90% compared to unfiltered water). Machine learning models coupled with SHAP analysis revealed that each DBP species responds to different controlling factors, with dichloroacetic acid governed by reaction time, trichloroacetic acid by chlorine dose, and trichloromethane by temperature, yet all pathways ultimately depend on precursor penetrated through the membrane. Integration of these findings into a quantitative selection framework balancing rejection against energy consumption identifies membranes based on characteristics of source waters, providing specific MWCO recommendations with associated energy costs for fourteen representative NOM distribution scenarios. This predictive approach provides a practical, semi-quantitative tool for membrane selection in drinking water treatment plants that minimizes DBP formation while maintaining operational sustainability. Download: Download high-res image (130KB) Download: Download full-size image