Data-driven neural networks for biological wastewater resource recovery: Development and challenges

Abstract

Recovering resources from wastewater has received increasing attention due to the requirement of carbon neutrality. The mathematical simulation of biological resource recovery processes and the intelligent control of wastewater treatment plants (WWTPs) are crucial for transforming traditional WWTPs into water resource recovery facilities (WRRFs). Although mechanistic models such as the activated sludge model and anaerobic digestion model have been widely applied, data-driven models, especially neural networks, outperform the mechanistic models in modeling intricate microbe-driven wastewater resource recovery processes with unknown mechanisms. Therefore, this review focuses on the development and current applications of neural networks including shallow and deep neural networks in the field of biological resource recovery from wastewater. The basic development and structures of neural networks are introduced first. Then, the current applications of neural networks in recovering biogas, volatile fatty acids, biofuel, electricity and bioplastic from wastewater are critically reviewed. The important input variables related to resource production are analyzed and the importance of preparing datasets for neural networks is highlighted. Moreover, the complexity of neural networks is discussed to guide the interdisciplinary development of neural networks in recovering resources from wastewater. Finally, the current limitations and perspectives of neural networks in this interdisciplinary field are proposed. The implementation of neural networks in full-scale WRRFs remains limited, necessitating further research and intensified efforts to enhance their practical applications. The combination of neural networks with mechanistic models presents great potential to further address practical modeling issues in this interdisciplinary field. This review would provide guidance for utilizing shallow and deep neural networks to model and optimize biological wastewater resource recovery processes.