A mechanistic model supported by data-based classification models for batch hydrogen production with an immobilized photo-bacteria consortium

A mechanistic model for batch hydrogen production through photo-fermentation with a consortium of purple non-sulfur bacteria is proposed and developed. Kinetic parameters were obtained from experimental photo-fermentations using fluorescent lamps as artificial light source. Three different light intensities (I) were tested in order to evaluate the influence of this process variable on growth rate and hydrogen production. Optimal conditions for state variables, such as I, pH and Temperature, were experimentally obtained and then used as parameters in the mechanistic model. Kinetics from either the proposed mechanistic model or the experimental photo-fermentations were compared, obtaining similar time trajectories for the process variables when operating at optimal conditions. However, this similarity decreased slightly at different light intensities, when the pH oscillated during the fermentations. Simulations under different initial conditions of substrate concentration, pH, and light intensity were run so as to generate data sets, used for constructing data-based classification models. A complete methodology, which entails a feature extraction step with Multi-way Principal Component Analysis technique and a proper classification stage using Support Vector Machines as algorithm, was proposed and addressed in order to develop such data models. These models were validated on both simulated and experimental data, which supported the robustness not only of the data-based models but also the developed mechanistic model. One-hundred percent classification performance was obtained after validating the data-based models, which envisages a further application of these models to other fields such as process monitoring, process control or fault diagnosis.