Determination of coagulant dosages for process control using online UV-vis spectra of raw water

Traditionally, coagulant doses are determined by the operators for the coagulation process at water treatment plants which is a multi-factor approach based on raw and treated water quality and in some situations relies heavily on their decisions. It can be challenging to determine appropriate coagulant doses proactively for tight coagulation control with the traditional method. Therefore, this study looked for alternative approaches for coagulation control and maybe the first to build coagulant dose determination models using only online raw water quality data (UV-Vis spectra) combined with chemometrics to determine coagulant doses for a drinking water treatment plant (WTP). Online UV-Vis spectral data at the raw water intake and alum dose data from a drinking WTP were used for building coagulant dose determination models. Three modelling techniques, including multiple linear regression (MLR), partial least squares (PLS) and artificial neural networks (ANNs), were applied in this work. The results show that MLR and PLS models had almost identical performances with small root mean square errors (RMSE) and high correlation coefficients (R2). Both MLR and PLS had slightly better performance than the ANNs for alum dose predictions. This study shows that the combination of online UV-Vis spectra and a chemometric method (MLR or PLS) was able to mimic operators' decisions in the determination of coagulant doses with a pH target of 6 to achieve a target DOC level of less than 5 mg/L for treated water quality.

Automated summary

This study developed coagulant dose determination models for a drinking water treatment plant using only online raw water quality data combined with chemometrics. Multiple linear regression and partial least squares models showed excellent performances in predicting alum dose, meeting the requirements of treated water quality.