Aerobic granulation is a complex process that, while proven to be more effective than conventional treatment methods, has been a challenge to control and maintain stable operation. This work presents a static data-driven model to predict the key performance indicators of the aerobic granulation process. The first sub-model receives influent characteristics and granular sludge properties. These predicted parameters then become the input for the second sub-model, predicting the effluent characteristics. The model was developed with a dataset of 2600 observations and evaluated with an unseen dataset of 286 observations. The prediction R2 and RMSE were >99% and <5% respectively for all predicted parameters. The results of this paper show the effectiveness of data-driven models for simulating the complex aerobic granulation process, providing a great tool to help in predicting the behaviour, and anticipating failures in aerobic granular reactors.
Keywords: Aerobic granulation; Data-driven modelling; Modelling; Neural networks; Wastewater treatment.
Copyright © 2018 Elsevier B.V. All rights reserved.