Empowerment of accurate modeling of anaerobic membrane bioreactors by automated machine learning

Machine learning (ML) is a promising approach for anaerobic membrane bioreactor (AnMBR) modeling. However, its complexity poses challenges for researchers who lack expertise in ML. To address the challenges of model selection and hyperparameter optimization in ML, we employed automated machine learning (AutoML) to model the removal of chemical oxygen demand (COD) in AnMBR treating municipal wastewater. The results show that AutoML achieved better modeling performance than previously applied deep neural models, with a mean absolute percentage error of 3.11%. We further investigated the impact of expanding the feature set and increasing the amount of data on modeling performance. The results demonstrate that the features of operation time can improve performance. However, adding more data does not markedly contribute to ML-based modeling. An analysis of ensemble feature importance indicates that the concentration of COD in influent is the most important feature for predicting the removal of COD. This study demonstrates the efficacy of AutoML for modeling AnMBRs and provides insights into the modeling of wastewater treatment processes with small datasets.