Evolutionary versus inductive construction of neurofuzzy systems for bioprocess modelling

The control and optimization of biotechnical processes is a complex task of industrial relevance, due to the growing importance attached to biotechnology. Therefore, there is an increasing use of intelligent data analysis methods for the development and optimization of bioprocess modelling and control. Since a clear understanding of the underlying physics does not exist, nonlinear learning systems, which can accurately model exemplar data sets and explain their behaviour to the designer, are an attractive approach. This paper investigates applying neurofuzzy construction algorithms to this problem and in particular compares a Genetic Programming structuring approach with a more conventional forwards inductive learning-type algorithm. It is shown that for simple problems, the inductive learning techniques generally outperforms the Genetic Programming, although for large complex problems, the latter may prove beneficial.