Artificial intelligence modeling to predict transmembrane pressure in anaerobic membrane bioreactor-sequencing batch reactor during biohydrogen production. (15th August 2021)

• Record Type: Journal Article
• Authors: Taheri, Ensiyeh; Amin, Mohammad Mehdi; Fatehizadeh, Ali; Rezakazemi, Mashallah; Aminabhavi, Tejraj M.
• Is Part Of: Journal of environmental management. Volume 292(2021)
• Publication Date: 2021-08-15
• Subjects: Anaerobic membrane bioreactor -- ANN -- ANFIS -- Fuzzy system -- Transmembrane pressure; Environmental policy -- Periodicals; Environmental management -- Periodicals; Environment -- Periodicals; Ecology -- Periodicals; 363.705
• DOI: 10.1016/j.jenvman.2021.112759
• Languages: English
• ISSNs: 0301-4797

Abstract

Abstract: The complex nature of wastewater treatment has led to search for alternative strategies such as different artificial intelligence (AI) techniques to model the various operational parameters. The present work is aimed at predicting the transmembrane pressure (TMP) as a key operational parameter in the case of anaerobic membrane bioreactor-sequencing batch reactor (AnMBR-SBR) during biohydrogen production using the adaptive neuro-fuzzy inference systems (ANFIS) and artificial neural network (ANN). In both the models, organic loading rates (OLR) ranging from 0.5 to 8.0 g COD/L/d, effluent pH (3.6–6.9), mixed liquor suspended solid (4.6–21.5 g/L) and mixed liquor volatile suspended solid (3.7–15.5 g/L) were used as the input parameters to test TMP as an output parameter. The ANFIS model was trained using the hybrid algorithms for TMP prediction. The higher prediction performance was obtained by using the Gauss membership function with four membership numbers. A back-propagation algorithm was also employed for the feed forward training of ANN model; the best structure was a Levenberg-Marquardt training algorithm with nine neurons in the hidden layer. By employing ANFIS and ANN models, relatively a good prediction of TMP was obtained with the R 2 values of 0.93 and 0.88, respectively while the calculated mean square error for TMP in the ANFIS model (7.3 × 10 −3 ) was lower than that of ANN model (8.02 × 10 −3 ). The higher R 2 and lower MSE values for the ANFIS model exhibited a better TMP prediction performance than the ANN model. Finally, it was observed that in the sensitivity analysis of ANN model, OLR was the most important input parameter on the variation of TMP. Graphical abstract: Image 1 Highlights: TMP variation during long-term operation of AnMBR-SBR for biohydrogen production was investigated. AI approach including ANFIS and ANN and ANFIS were used for TMP modeling. ANN and ANFIS well define the relations between TMP and input parameters. ANFIS model showed better prediction capability than the ANN model. OLR was the dominant parameter for TMP prediction in ANN model.