Optimization and Prediction of Moving Bed Biofilm Reactor (MBBR) Using Surface Response Method (RSM) and Artificial Neural Network (ANN)

 In this study, the optimization and prediction of the efficiency of a moving bed biofilm reactor (MBBR) in the treatment of synthetic wastewater containing organic material including aniline was investigated using response surface methodology and artificial neural network.

Modeling results were applied to a 5-liter volume reactor filled with 30%, 50% and 70% LECA lightweight aggregates as a growth medium for microorganisms and biofilm layer formation. In order to determine the optimum conditions in the experiments results and also to predict the tests not performed, three factors were feed levels at levels of 100 to 3000 mg/L, retention time of 8 to 72 hours and filling percentage of 30 and 50 and 70% were performed using RSM. The accuracy of the presented models was evaluated by ANOVA. Prediction of system removal efficiency using radial basis ANN was also investigated.

Process optimization showed that the optimum conditions for maximum removal were at feed rate of 1700 mg/l and 72 hours at 56.82% filling percentage. The results of the process prediction using radial basis ANN also showed that in the best network structure with Radbas and linear functions (Purelin) with R2 = 0.982 can predict the efficiency.

By comparing the radial basis ANN model and RSM and comparing the error rates of these two methods, it can be concluded that the radial base ANN method predicts the data process more accurately and with lower error.