Comparison of statistical methods to predict fouling propensity of microfiltration membranes for drinking water treatment

This paper investigated the feasibility and limitation of statistical models to predict membrane fouling in a pilot-scale system. Operation data from an MF pilot plant with the capacity of 440 m(3)/d were used for the application of these models. Water quality parameters including feed water turbidity, algae concentration, total organic carbon, dissolved organic carbon, and UV254 absorbance were correlated with transmembrane pressure, total resistance, and the rate of resistance change. Model fit equations were derived from multiple linear regression, artificial neural network, genetic programming, and support vector machine. The performances of models were compared in terms of accuracy and prediction capability.