An Integral Plus States Adaptive Neural Control of Aerobic Continuous Stirred Tank Reactor

A direct adaptive neural network control system with and without integral action term is designed for the general class of continuous biological fermentation processes. The control system consists of a neural identifier and a neural controller, based on the recurrent trainable neural network model. The main objective is to keep the glucose concentration, which is considered as external substrate, close to a constant set-point reference using the dilution rate as manipulating function. It is illustrated by simulations that both adaptive neural control schemes (with and without integral-term) work successfully and exhibit good convergence. However, the control system with integral action is able to compensate a constant offset while the scheme without integration term failed. Results are presented which show a favorable behavior of the neural controller in comparison with existing solutions.