Predicting water flux and reverse solute flux in forward osmosis processes using artificial neural networks (ANN) modelling with structural parameters

Abstract

Artificial neural network (ANN) has gained increasing attention in modelling wastewater treatment processes because it harnesses historical datasets to uncover relationships between input and output data, enabling accurate predictions without the need for complex equations. However, there are few ANN studies that simultaneously predict both water flux (Jw) and reverse solute flux (Js) in the forward osmosis (FO) process. In this work, we present an ANN model designed with only eight input variables, introducing the rarely explored membrane structural parameter (S) as an input variable for the first time. Through the optimisation of hidden layers, neuron counts, and the combination of activation functions, the ANN model achieved high R-squared values of 94.6 % and 95.2 % for Jw and Js prediction, respectively. To assess the model’s versatility across different scenarios, the application scope of the model was further evaluated using three distinct membrane types (namely, substrate-modified, active layer-modified and interlayer-modified). Results demonstrated the ANN model’s capacity to accurately predict the water flux of the substrate-modified membranes. Remarkably, even though the interlayer-modified and active layer-modified membranes were not part of the model’s training, the ANN model still exhibited relative accuracy, effectively capturing the modifications made through the structural parameter input. Compared with the solution diffusion (SD) model, the ANN model showed promising potential as a simplified yet powerful tool to predict the water flux in FO processes. Its capability to deliver accurate results with fewer complexities makes it an attractive choice for the modelling of FO processes.