Machine learning for membrane bioreactor research: principles, methods, applications, and a tutorial

Membrane fouling poses a significant challenge to the sustainable development of membrane bioreactor (MBR) technologies for wastewater treatment. The accurate prediction of the membrane filtration process is of great importance for identifying and controlling fouling. Machine learning methods address the limitations of traditional statistical approaches, such as low accuracy, poor generalization ability, and slow convergence, particularly in predicting complex filtration and fouling processes within the realm of big data. This article provides an in-depth exposition of machine learning theory. The study then reviews advances in MBRs that utilize machine learning methods, including artificial neural networks (ANN), support vector machines (SVM), decision trees, and ensemble learning. Based on current literature, this study summarizes and compares the model input and output characteristics (including foulant characteristics, solution environments, filtration conditions, operating conditions, and time factors), as well as the selection of models and optimization algorithms. The modeling procedures of SVM, random forest (RF), back propagation neural network (BPNN), long short-term memory (LSTM), and genetic algorithm-back propagation (GA-BP) methods are elucidated through a tutorial example. The simulation results demonstrated that all five methods yielded accurate predictions with R-2 > 0.8. Finally, the existing challenges in the implementation of machine learning models in MBRs were analyzed. It is notable that integration of deep learning, automated machine learning (AutoML) and explainable artificial intelligence (XAI) may facilitate the deployment of models in practical engineering applications. The insights presented here are expected to facilitate the establishment of an intelligent control framework for MBR processes in future endeavors.