Accurate Prediction of Reverse Solute Flux in Forward Osmosis Systems Using Comparative Machine Learning Models

Abstract

Forward osmosis (FO) is a promising technology that can help to recover freshwater from saline and wastewater streams. However, it faces challenges such as reverse solute flux (RSF). RSF is the movement of salts from the draw solution to the feed solution, which can have several negative impacts on FO performance. This paper focuses into the development of machine learning techniques, including MLR, ANN, and ANFIS, to predict RSF in the FO system. Two commercially available membranes (CTA and TFC) were used to avoid the influence of synthesized membranes. The models were evaluated using experimental data obtained from previous lab-scale experiments. The results showed that the ANFIS and ANN models were both accurate, with R² values of 96% and 97.6%, respectively. The ANFIS model performed slightly better than ANN model, while the MLR model displayed inaccurate predictions, with a lower R² (43.46%) and MSE (2.61 × 10^–2) and higher AARE (2.221). The study also identified the most impactful parameters on RSF, such as the DS type, FS concentration, and FS temperature. The study concludes that machine learning techniques can be successfully applied to model RSF in FO systems with high accuracy and recommends their use in the industry to improve the performance of membrane-based systems.