A methodology for coagulant virtual testing to improve dissolved organic matter removal in surface water treatment

Coagulation is one of the most crucial steps in a Drinking Water Treatment Plant (DWTP). The coagulant dose required for the removal of particles and natural organic matter (NOM) is typically determined through jar tests. However, this method is time-consuming and not well-suited for rapid changes in raw water quality, such as those occurring during and after rainfall events. We propose a methodology for estimating the combined coagulant doses needed for NOM removal (represented by UV absorbance at 254 nm, UV₂₅₄) using a machine learning technique called Support Vector Regression (SVR) in a full-scale DWTP that does not conduct independent controls of coagulation pH. The methodology involves virtual testing of combined coagulant dose performances on UV₂₅₄ removal and coagulation pH. Performance metrics demonstrated the high capacity of the models to predict UV₂₅₄ removal and coagulation pH in the test dataset. Furthermore, our proposed methodology includes a strategy to evaluate whether the predicted coagulation pH limits residual aluminum in the produced water. The proposed framework can assist decision-making for coagulation operation practices in DWTPs.