Membrane bioreactors (MBRs) are widely implemented in many wastewater treatment plants (WWTPs), contributing to a sustainable urban water cycle. However, they are often challenged by the problem of membrane fouling. Developing tools capable of monitoring and providing early warning of fouling events is therefore essential to maintain the operational stability of MBRs. Here we propose a spectral sensing empowered machine learning strategy for prediction and providing early warning of fouling events and validate it in a full-scale MBR during long-term operation. Specifically, integrated with ultraviolet-visible and fluorescence spectral fingerprints that can sensibly profile the key molecular structures responsible for fouling, a data–knowledge codriven machine learning model was established. The model can accurately ( R 2 > 0.80) predict fouling tendency within the next 5‒7 days, which is superior to purely data-driven learning. Spectral fingerprints contributed 17‒30% of interpretability across models, and process conditions and other membrane/foulant material descriptors explained the rest. By exploring the spectral fingerprints of specific processes, our strategy could also be generalized to various WWTPs, offering an intelligent solution (for example, with the aid of spectral early warning) to achieve stable operation of WWTPs and ultimately contribute to a cleaner environment.
Lai et al. (Mon,) studied this question.