Using Neural Network Observers For Bioprocess Control

Abstract

In the present paper a Neural Network based state observer is proposed to be used as a part of a bioprocess control. The production of Saccharomyces cerevisiae, described by a sixth order nonlinear state space model, is used as an example of such a fed-batch bioprocess. The control objective is to ensure the process stability and desirable specifications in the presence of disturbances and lack of reliable state measurements. First, the model of the process and its properties are presented. Next the ability of multi-layer Neural Networks to act as reliable emulators of the system dynamics is tested by simulation results and a discussion on possible better training strategies is made. Finally, a non-linear adaptive observer is designed by means of artificial neural networks.