Modelling of submerged membrane flocculation hybrid systems using statistical and artificial neural networks methods

Hybrid membrane filtration processes involve complex physical, chemical and biological phenomena, thus their mechanistic modelling is challenging. The chief advantages of statistical and artificial neural networks (ANN) models (data-driven models) are that they do not require assumptions and simplifications to establish relationships from data. This paper investigates the characteristics and performance of several data-driven methods to model a hybrid membrane system. The focus is on the application of regression analysis and artificial intelligence based methods to a steady-state system. Among empirically based approaches, ANN neural networks methods were found to be very useful to predict permeate quality and membrane fouling. In the past multivariate nonlinear regression had barely been investigated for process modelling in water and waste water treatment. In this study polynomial multivariate nonlinear regression showed a superior performance. Multivariate parametric nonlinear models could match the performance of the nonparametric ANN models in the empirical modelling of complex systems, especially when combined with advanced optimization methods. This paper gives the methodology of how one could optimize a membrane hybrid system using ANN, validating it with one set of data. The same procedure/methodology can be applied to similar systems.