Feedback optimization of fed-batch bioreactors via neural net learning

The problem of feedback optimization of the feed-rate for a fed-batch bioreactor is investigated by a neural network approach. Using the nonlinear mapping ability of neural networks, feedback optimal control can be realized as a nonlinear function of the state variables. A neural network trajectory learning technique is proposed for optimisation of the desired product for a fed-batch bioreactor. A simulation study of cell mass production demonstrates the training of the neural network feedback controller to achieve the production objective. The superiority of a feedback optimization scheme over open-loop optimal control when there is either model or initial condition error is illustrated and discussed.<
>