State Estimation Technique And Predictive Control Based On Artificial Neural Networks

Abstract

Severe problems occur in the control of fermentation because of the poorly understand nature of the process, its nonlinearity and the wide range of operating states passed through during a batch. During a typical production batch, important variables such as product concentration are determined by slow infrequent 08line laboratory analysis, making this set of variables of limited use for control. In this work, the artijlcial neural network technique has been used for the on-line estimation of the important state variables of the fed batch fermentation process. The neural network tasks include both modelling and state estimation of the residual nitrogen inside the fermenter (Residual nitrogen is one of the key variables required for improved control.) The second part of the paper describes a controller design based on the predictive control approach. The aim of the controller is to maintain residual nitrogen around a desired level by controlling the amount of soluble nitrogen fed.