Hybrid model development for fed-batch bioprocesses; combining physical equations with the metabolic network and black-box kinetics ∗

A hybrid modeling technique is proposed to make models during the development phase of a fed-batch bioprocess. A limited amount of data is available and experimental data are therefore combined with a priori information about the mass balances and the chemical reaction network inside the cell (the metabolic network). This information highly constraints the model, and only a few degrees of freedom remain in the static description. The dynamic part (kinetics) is described by fuzzy models or neural networks. The final model consists of differential equations based on mass balances partly composed of fuzzy or neural functions. Some preliminary results are shown for the clavulanic acid production by Streptomyces clavuligerus.