Neural network modelling and predictive control of yeast starter production in champagne

The second fermentation is one of the most important steps in Champagne production. For this purpose, yeasts are grown on a wine based medium to adapt their metabolism to ethanol. A recurrent neural network combined with a stoichiometric reaction scheme were identified as a state model of yeast growth fermentation process. This model was used to perform an open-loop or a closed-loop control of the final yeast concentration (after a fermentation time of 21 hours) following a predictive mode control. Industrial application of the control let to a 4% error between the desired and the measured final yeast concentration. This was good enough to guaranty a constant production of yeast with an efficient physiological state.