Machine learning-guided prediction of polymeric membrane performance in forward osmosis

Shengnan Hao, Meng Wang, Hao Guan, Yuhao Zhao, Zhanlin Ji, Chenxu Dai,
Machine learning-guided prediction of polymeric membrane performance in forward osmosis,
Separation and Purification Technology,
Volume 379, Part 2,
2025,
135037,
ISSN 1383-5866,
https://doi.org/10.1016/j.seppur.2025.135037.
(https://www.sciencedirect.com/science/article/pii/S1383586625036342)
Abstract: Forward osmosis (FO) offers a promising low-energy approach for seawater desalination and sustainable water purification. However, achieving simultaneous high-water-flux and low-reverse-solute flux in FO polymeric membranes remains a key challenge. To address this issue, we propose a machine learning framework to accurately predict membrane performance and explore the coupling mechanism between water flux and reverse solute flux using advanced machine learning (ML) techniques. Specifically, this study employed several ML models to develop a data driven prediction framework that reveals the impact of membrane type and operating parameters on performance. This work utilized interpretable machine learning methods such as shapley additive explanations and partial dependence plots to quantify the effects of different operational variables and membrane structural parameters on water flux and reverse solute flux. Notably, structural parameters were identified as having the most significant impact on membrane performance, with higher draw solution concentrations leading to an increase in both water flux and reverse solute flux. Additionally, the reliability and stability of the PSO-tuned XGBoost model were evaluated using three metrics (R2, MSE, MAE). This study highlights the significant potential of advanced ML techniques for predicting the performance of forward osmosis polymeric membranes and provides a robust data-driven approach for membrane design and operational optimization in water purification processes.
Keywords: Forward osmosis; Water purification; Polymeric membrane; Machine learning; Performance prediction