TWO NOVEL METHODS OF OPTIMAL REPETITIVE CONTROL IN FED-BATCH BAKER'S YEAST FERMENTATION USING NEURAL NETWORKS

Two novel methods of optimal repetitive control in fed-batch baker's yeast fermentation using neural net are presented. In one method, feedback control law is used. Neural net parameters of process and controller are learned by backpropagation. In the other method, direct optimal control law is used. Neural net parameters of process and neural net input (that is control action) are learned. Two methods all have self-adaptive capability. It takes the form of dynamic repetitive control in implementation that the methods have the advantages similar to predictive control. The robustness is very strong. The methods are applied to the optimal control of a baker's yeast fed-batch fermentation. Simulation results are satisfactory. Based on the optimal control strategies for substrate feed rate, more economic profile is obtained by experiments. The productivity is increased by 26% and the cosumption of molases is decreased by 4%, compared to industrial scale production. This method provides a new way to solve optimal control of biochemical engineering process for which it is difficult to model by conventional method.