Simulation of Micropollutant Behaviour in Forward Osmosis Membrane-Based Process Using Machine Learning Algorithms

Abstract

Simulation of the MP behaviour in forward osmosis (FO) membrane-based processes may provide a better understanding and design of the process to facilitate the highest performance. This study explored the feasibility of applying machine learning (ML)–based models to simulate MP behaviour in the FO-membrane bioreactor (FOMBR) process. The information obtained on 97 MPs revealed that FO demonstrated extremely low rejection efficiency (2-80%) for low molecular weight MPs. The pre-evaluation of the dataset indicated that a higher number of input variables resulted in a higher performance of the ML models in the prediction of MP rejection. Among eight investigated ML models, ensembles of trees (ET), adaptive-neuro fuzzy inference system (ANFIS), and Gaussian Process Regression (GPR) were the most effective approaches for the prediction of MP behaviour. Further optimisation of the ANFIS with a substractive clustering radius of 0.1 (ANFIS-SC) showed an excellent performance in forecasting MP removal (R = 0.99 and RMSE = 0.56%). In addition, the developed ANFIS-SC was feasible for simulating the influence of operational parameters on the elimination of MPs by FOMBR, contributing notably to the better design and more efficient operation of the system to achieve the highest elimination of the target MPs in the future.

Keywords: Artificial intelligence, forward osmosis, modelling, micropollutants, osmotic membrane bioreactor