Membrane Bioreactors (MBRs) are preferred for treating high-strength industrial wastewater due to their exceptional treatment efficiency and compact design. However, membrane fouling remains a significant challenge that limits their broader application. Since transmembrane pressure (TMP) is closely associated with membrane fouling mechanisms, its accurate prediction is essential for anticipating fouling events. This study focuses on predicting TMP in a bench-scale MBR by employing advanced regression models such as Lasso, support vector machines, and random forest. The bioreactor is operated using synthetic textile wastewater, with a primary focus on critical sludge parameters. A total of 60 datasets were compiled for training and testing of the models. The random forest model attained an R2 of 0.95 and 0.86 and root mean square error values of 1.75 kPa and 3.3 kPa for the training and test datasets, respectively, demonstrating the best predictive accuracy. Extra polymeric substances were the most influential parameter, while particle size distribution was the least. The random forest and lasso regression models predicted high values more accurately, while the support vector regression model was inclined towards making more conservative predictions for extreme values. The effectiveness of the proposed models showcases the significance of utilizing machine learning in advancing MBR technology for wastewater treatment. Accurate TMP predictions through machine learning can help in maintaining key parameters within optimal ranges, thereby reducing the likelihood of membrane fouling and enhancing system efficiency.
