| Title |
Performance Prediction of Forward Osmosis Membrane Module Using Multiple Linear Regression and Artificial Neural Network Models |
| Authors |
이해룡(Haelyong Lee) ; 미타 누르하야티(Mita Nurhayati) ; 이승윤(Sungyun Lee) |
| DOI |
https://doi.org/10.15681/KSWE.2025.41.6.539 |
| Keywords |
Artificial neural network; Forward osmosis; Machine learning; Multiple linear regression; Performance prediction |
| Abstract |
Sustainable desalination technologies are gaining attention, with forward osmosis (FO) emerging as a promising alternative to reverse osmosis due to its low energy consumption and reversible fouling. However, accurately predicting FO module performance remains a challenge. This study developed and compared multiple linear regression (MLR) and artificial neural network (ANN) models to predict the performance of FO membrane modules using 69 datasets from pilot-scale plate-and-frame systems operating under varied conditions (membrane areas: 7?63 m2; feed concentrations: 10?30 g/L; draw concentrations: 70?150 g/L; flow rates: 5?20 L/min). Variable importance analysis revealed that membrane area and feed concentration are the primary factors affecting water flux. Both models exhibited high predictive accuracy (R2 > 0 .95). The MLR model demonstrated an R² of 0.9577 and a root mean square error (RMSE) of 0.6550 L m-2 h-1, with statistical validation (F = 228.74, p < 10-32) and clear interpretability of variables. The ANN model achieved a slightly higher accuracy with an R2 of 0.9886 and an RMSE of 0.3498 L m-2 h-1, along with improved generalization stability. For predicting recovery rates, both models reached an R2 greater than 0.95, with the ANN model (0.9928) performing marginally better than the MLR model (0.9525). These results indicate that both methodologies provide reliable frameworks for predicting FO performance, with MLR offering interpretability and ANN delivering greater accuracy, making them suitable for different aspects of FO process design and scale-up. |