Jaume Giralt1, Gabriela Espinosa1, Alexandre Restrepo1, and Francesc Giralt2. (1) Departament d'Enginyeria Química, Universitat Rovira i Virgili, Av. dels Països Catalans 26, Tarragona, 43007, Spain, (2) Fenòmens de Transport, Departament d'Enginyeria Química, Universitat Rovira i Virgili, Av. dels Països Catalans 26, Tarragona, 43007, Spain
A controller based on Neuro-dynamic programming coupled with an adaptive neural network has been developed to optimally produce cloned invertase in Saccharomyces Cerevisiae yeast in a fed-batch bioreactor. There are many candidate feed rate profiles and testing all of them experimentally is not a feasible option. Thus, a model able to identify the best feed rate profile would be of interest for the above mentioned or similar reactions. Different experiments have been conducted to simultaneously optimize reaction time and feeding trajectory. Different objective functions, including product removal and reactor sterilization time, have been considered for training the adaptive neural system. The optimal glucose feed rate profile needed to achieve the highest fermentation profit in this reactive system, where the enzyme expression is repressed at high glucose concentrations, has been determined, with the controller updating in time an optimized control action that incremented the fed-batch bioreactor profitability.