The long-term running continuous microfiltration-submersed (CMFS) reactor was built and spectral indicators, organic matter, and fouling potential were monitored. The specific filtration volume decline rates (V0.5 and V0.25) and unified membrane fouling index (UMFI) were characterized as pore blocking, gel layer, and cake layer fouling potential, respectively. Spectral analysis results showed the parameters of both S275–295 (slope of log-transformed absorption coefficient with the range of 275–295 nm) and SR (the ratio of S275–295 and S350–400) were significant correlations with V0.5, V0.25, and UMFI. The humification index (HIX) had stronger correlations with UMFI, proving the humic acids (HA) had more significant impacts on the cake layer. The V0.5 and V0.25 had correlations with the ratio of peak T to peak C (fT/C) because the aromatic proteins (PN) were more involved in the pore blocking. Statistical analysis results showed that V0.5 and V0.25 were mainly contributed by the polysaccharide (PS) and the interactions of PS × PN × HA, while UMFI was mainly contributed by the HA and PS × PN × HA interactions. The multivariate linear regression (MLR), backpropagation neural network (BPNN), and support vector regression (SVR) models were successfully employed to achieve multiple signals fusion and prediction potential, with the SVR model having better accuracy, especially for UMFI.
Read full abstract