Modelling of the extractive membrane bioreactor process based on natural fluorescence fingerprints and process operation history

Abstract

A method for non-mechanistic and non-linear modelling of complex biological processes is presented, using the example of the extractive membrane bioreactor (EMB). The model is based on artificial neural networks (ANN), which are able to predict the state of the process from a combination of reactor operational parameters and natural fluorescence fingerprints. Current as well as historic process operation is included in the ANN input vector, in order to account for lag-times within the reactor system and for biofilm dynamics that are dependent on process history. The model is especially relevant for practitioners, as it does not require assumptions on underlying process mechanisms, and it relies on routinely available operational data and on an easy-to-install, non-invasive, in-situ, on-line monitoring method. Moreover, it focuses on the prediction of overall process performance parameters, which are of immediate relevance in practice. The developed model was able to predict the process state very well. Sensitivity analysis revealed that the main impact on process performance stems from process operation rather than the physiological state of the biological culture, and that in the EMB configuration employed process operation history decisively impacts on the process outcome.