Comparing new perspective of hybrid approach and conventional kinetic modelling techniques of a submerged biofilm reactor performance

This study dealt with modelling a submerged biofilm reactor using three modelling approaches including Artificial Neural Network (ANN), Evolutionary Polynomial Regression (EPR), and Modified Stover Kincannon (MSK). Generally, eight experimental runs were conducted under different operational conditions, i.e., influent COD (at four levels of 200, 300, 400, and 500), aeration rates (at two levels of 4 and 8 L/min), and operating run time (20 days). Regardless of specific conditions, the predicted fluctuation of effluent COD by MSK, ANN, and EPR were in the range of 22-104, 12-112, 21-115, respectively while from the experimental data, this values were obtained between 13 and 126. Based on the MSK model, an increasing at aeration rate from 4 to 8 L/min causes doubling in the maximum COD removal rate constant (Rmax) from 10.5 to 20.3 (gCOD/L day). Among the applied models, ANN with minimum normalized Mean Square Error (MSE) of 0.1 showed an accurate prediction effluent COD (R2 = 0.95); however, it gave no information about reactions occurring in the system. The results suggested that, EPR method with a good coefficient of determination (R2 = 0.93), the representation of mathematical model expressions and the interpretation of the underlying phenomena is promising modelling technique for prediction of biofilm processes.