Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
410535 | Neurocomputing | 2009 | 6 Pages |
It is known that the performance of an optimal control strategy obtained from an off-line computation is degraded under the presence of model–plant mismatch. In order to improve the control performance, a hybrid neural network and on-line optimal control strategy are proposed in this study and demonstrated for the control of a fed-batch bioreactor for ethanol fermentation. The information of unmeasured state variables obtained from the neural network as an on-line estimator is used to modify the optimal feed profile of the fed-batch reactor. The simulation results show that the neural network provides a good estimate of unmeasured variables and the on-line optimal control with the neural network estimator gives a better control performance in terms of the amount of the desired ethanol product, compared with a conventional off-line optimal control method.