Characterization of Coagulant-Induced Membrane Fouling by Multi-Spectral Fusion : Dom Properties and Model Prediction Based on Machine Learning

This study used the ultraviolet-visible (UV–vis), synchronous fluorescence, and excitation-emission matrix (EEM) spectroscopy coupled with zeta potential and particle size to explore the DOM solutions properties with the addition of two kinds of coagulants, including aluminum chloride and ferric chloride, in humic solutions and surface water. The results of zeta potential and particle size showed the size of aggregates became larger with the addition of coagulant. The binding sequence of coagulant to DOM was followed from aromatic amino acids to fulvic-like and humic-likes fractions. The results of the unified membrane fouling index (UMFI) were showed negatively correlated with the concentrations of coagulant due to the cake layers formed by aggregates being looser and more permeable. Among all the 12 spectral parameters, slope radio (SR) and specific fluorescence intensities (SFI) had the most significant correlation to UMFI. The multi-spectral fusion for predicting UMFI was established via multiple linear regressions and backpropagation neural network by using the parameters with both SR and SFI with good accuracy and robustness. Sensitivity analyses results showed that SFI was more sensitive than SR in solutions with high protein-like substances, while SR was more sensitive than SFI with high humic-like substances