초록
<P><B>Abstract</B></P> <P>This study reveals similar kinetic patterns among batch indoor photo-fermentations using tungsten light and batch outdoor photo-fermentations irradiated by solar light, only considering the lighting period. The potential of Artificial Neural Networks (ANN) as a modeling technique has been evidenced by simulating the biohydrogen production by photo-fermentation using an immobilized consortium of photo-bacteria. The ANN model was constructed with a set of indoor experimental fermentations operated on batch at 30 °C and under different conditions of light intensity, initial pH and metals concentrations (Fe, V and Mo) added to the medium. After that, the model was cross-validated on indoor photo-fermentations as well. Different ANN architectures were evaluated to develop the best data-based model. The chosen architecture can render the maximum correlation between the real bio-hydrogen production and the outputs provided by the ANN model. Experimental kinetics were contrasted with the modeled kinetics, evidencing the reliability of the model for predicting the biohydrogen production by supplying sampling times, and initial operating conditions such as metals concentration, light intensity and pH as input data. The ANN-based model was successfully validated on an outdoor fermentation, where light intensity changed along the process time, which demonstrated its veracity and generalization capacity.</P> <P><B>Highlights</B></P> <P> <UL> <LI> H<SUB>2</SUB> production by immobilized photo-bacteria was modeled with data-based models. </LI> <LI> Artificial Neural Networks (ANN) were constructed using indoor data. </LI> <LI> The best ANN architecture was 6-9-1, trained with the Bayesian Regulation algorithm. </LI> <LI> ANN model was randomly cross-validated using 80% batches for calibration and 5 models. </LI> <LI> ANN model accurately predicted experimental outdoor H<SUB>2</SUB> production. </LI> </UL> </P>