Artificial neural network modeling on trichloroethylene biodegradation in a packed-bed biofilm reactor and its comparison with response surface modeling approach

Abstract

An artificial neural network (ANN) model was developed for predicting trichloroethylene (TCE) biodegradation via co-metabolism in a packed-bed biofilm reactor using Pseudomonas putida F1 as the bio-catalyst and toluene as the primary substrate. The model uses an architecture with three hidden layers (3–2–1) and resilient back-propagation algorithm to obtain optimal weights for the associated neurons. ANN model predictions of TCE degradation efficiency have not only successfully validated those obtained from a Response Surface Model (RSM), but the model also outperformed RSM with a better predicting power that the ANN showed a lower root mean square error (RMSE) 9.07 and a higher determination of coefficient (R2) 80.32 % than those of RSM (RMSE=9.93, 10.88 after corrected for model degree of freedom; R2 = 76.4 %). Further model analysis revealed that this ANN model correctly interpreted the experimental data with the underlying biological mechanisms and would be able to provide guidance on local operations to achieve optimal TCE degradation. Sensitivity analysis showed that the ANN model was still a robust model when an extreme TCE data point was added during the model development. A simulation study also indicated that on average, ANN models had better fit to the experimental data than the RSM.