Application of Al Techniques to Baker's Yeast Fermentation: A Simulation Study

摘要

The cultivation can be divided in three or more separate phases according to thedynamics of the process. To define and recognize the cultivation phases a fuzzy expert system was employed. On the basis of fuzzified measurements of the process variables the current phase was inferred using rules describing the process. The rules were of the type: If S is higher than 'critical' and E is 'increasing' and DO is higher than 'critical' then the phase is First ethanol production phase. During different phases different substrate feedrate control strategies are required. For each process phase a separate neural network structure was designed for state variable estimation and all the three neural networks were separately trained according to the process behavior during the corresponding process phase. By changing the parameters in the model, simulation studies were performed in order to identify the neural networks and the fuzzy phase recognition system. The present study shows that fed-batch baker's yeast cultivation can be controlled by using a hybrid fuzzy-neural system with separate neural networks trained for different process phases.