Predictive control in biotechnology, using fuzzy and neural models.

New developments in process modeling, identification, measurement and control are likely to cause some major breakthroughs in process control in the next decade. Especially black box modeling techniques based on Artificial Neural Networks and Fuzzy Set theory are opening new horizons for modeling and controlling non-linear processes in biotechnology. The link between accurate dynamic process models and actual process control is provided by the concept of Model-based Predictive Control (MBPC). A model serves here as process output predictor so that the effect of (future) control actions can be evaluated automatically, before the process is activated. This chapter presents a brief introduction to modeling with fuzzy sets and artificial neural nets. To demonstrate the practical applicability, laboratory experiments are described where MBPC was applied to a non-linear pressure control problem in a fermentor. Both fuzzy and neural models were developed and identified for this process and as the results show the fuzzy and neural MPBC outperform the classical PI controller. Controller tuning was very easy compared to classical (linear) techniques.