Revealing the removal behavior of five neglected microplastics in coagulation-ultrafiltration processes: Insights from experiments and predictive modeling

Abstract

Typical water treatment processes are essential for mitigating the risk of microplastic contamination in drinking water. The integration of experiments and machine learning offers a promising avenue to elucidate microplastic removal behavior, yet relevant studies are scarce. To address this gap, this study combined experimental and artificial neural network (ANN) modeling to explore the removal behavior and mechanisms of five neglected microplastics in typical coagulation-ultrafiltration processes. Experimental results demonstrated that coagulation achieved an optimal removal rate of 37.0–56.0 % for the five microplastics, and subsequent ultrafiltration almost completely removed all residual microplastics. Five ANN models were constructed and optimized by adjusting activation functions and employing batch normalization, accurately predicting microplastic removal, with high R² values of 0.9972–0.9987. X-ray photoelectron spectroscopy elucidated the involvement of Al^IV and Al^VI species, hydrogen bonding, and π-π interaction in coagulation. Two-dimensional correlation spectroscopy explored the sequential formation of six chemical bonds (C–H, Al–O–Al, C–O, COO^-, C=O, and –OH) and potential mechanisms. Moreover, theoretical calculations clarified the interfacial interactions between microplastics and ultrafiltration membrane, highlighting the roles of hydrophobic attraction and acid-base interaction. This study expands our understanding of microplastic removal in drinking water treatment, providing valuable mechanistic and modeling insights.