Internal model control framework using neural networks for the modeling and control of a bioreactor

Abstract

An internal model control (IMC) framework using neural networks for the modeling and control of a nonlinear bioreactor is presented. Unlike existing IMC design techniques, this approach needs no mathematical models. It is shown that, to obtain an accurate inverse model, one needs to use steady-state data in addition to the transient data for training the networks. An integration of neural networks and an unstructured math model of the bioreactor is also proposed to improve the neural networks' modeling accuracy. This hybrid approach shows significantly better performance than the “black box” method, and almost as good a performance as a nonlinear IMC based on an exact mathematical model. The hybrid method also has other advantages, such as the use of only steady-state data and the need for only one neural network that can be used for both the process model and the inverse process model. Simulation results show that the neural-net strategy is superior to a PI controller.