In the present study, the performance of a membrane bioreactor (MBR) was modelled using a hybrid approach based on the activated sludge model number 3 (ASM3) combined with projection to latent structures (PLS) to predict the residuals of the ASM. The application of ASM to MBRs requires frequent re-calibration to adjust the model to variations in influent characteristics, determined through time-consuming analysis and batch tests. Considering this problem, the objective of this study was to improve ASM prediction ability with minimal additional monitoring effort. Hybrid models were developed to predict three MBR performance parameters: mixed liquor suspended solids (MLSS), COD in the permeate (CODp) and nitrite and nitrate concentration in the permeate (NOxp). For PLS modelling of ASM residuals three input strategies were used: (1) analytic and operating data; (2) operating data plus 2D fluorescence spectroscopy; (3) all the data. The first input strategy improved ASM prediction of the three selected outputs, and highlighted the lack of detailed and real-time information from wastewater and operating parameters in the ASM used in this study. In the second input strategy, the incorporation of updated data from 2D fluorescence spectroscopy resulted on better model fitting than in the first input strategy, for all the output parameters studied. Through the hybrid modelling approach it was possible to significantly improve the ASM predictions in real-time using 2D fluorescence measurements and other relevant parameters acquired on-line, without requiring further laboratory analysis. Furthermore, the third input strategy, incorporating all the collected data, did not significantly improve the prediction of the outputs beyond the second strategy. This shows that 2D fluorescence spectroscopy is a comprehensive monitoring tool, able to capture on-line the required information to complement, through hybrid modelling, the mechanistic information described by an ASM.
- 2D fluorescence spectroscopy
- Hybrid modelling