Recurrent neural network model of a fed-batch Saccharomyces cerevisiae fermentation process

The recurrent neural network model is applied for identification and prediction of the fed-batch fermentation process of yeast Saccharomyces cerevisiae which is a microorganism that can be cultivated under aerobic and anaerobic conditions, obtaining in each case completely different products. Under aerobic conditions, a biomass is produced, while under anaerobic conditions, the principal product is ethanol, which yield depends on the initial biomass concentration. Thus, a good control of the fermentation parameters is needed to assure the efficiency of this fermentation process. The data used to train the RTNN has been obtained from mathematical models, derived from simple material and energy balances, and from the Monod equation. The output data generated by the RTNN are compared with those generated by the nonlinear process model and used for the NN training.