Comparative Assessment of the Artificial Neural Network and Response Surface Modelling Efficiencies for Biohydrogen Production on Sugar Cane Molasses

This study comparatively evaluates the modelling efficiency of the Response Surface Methodology (RSM) and the Artificial Neural Network (ANN). Twenty-nine biohydrogen fermentation batches were carried out to generate the experimental data. The input parameters consisted of a concentration of molasses (50-150 g/l), pH (4-8), temperature (35-40 A degrees C) and inoculum concentration (10-50 %). The obtained data were used to develop the RSM and ANN models. The ANN model was a committee of networks with a topology of 4-(6-10)-1 structured on multilayer perceptrons. RSM and ANN models gave R (2) values of 0.75 and 0.91, respectively, with predicted optimum conditions of 150 g/l, 8 and 35 A degrees C for molasses, pH and temperature, respectively, with differences in inoculum concentrations (10.11 and 15 %) for RSM and ANN, respectively. Upon validation, 15.12 and 119.08 % prediction errors on hydrogen volume were found for ANN and RSM, respectively. These findings suggest that ANN has greater accuracy in modelling the relationships between the considered process inputs for fermentative biohydrogen production and thus, is more reliable to navigate the optimization space.