Fuzzy Model Based Control of Biotechnological Processes

The use of biotechnological processes in modern industry is of increasing importance. Food production, pharmacology, and waste management are best-known elds microorganisms have already been successfully applied in. Like for any other industrial process, optimal running of fermentation processes requires at least the following criteria to be maximized: productivity, prootability, and security. To achieve these goals, eecient control mechanisms have to be applied, which on their parts should be based on a-priori knowledge of the process behaviour in order to lead to acceptable results. As far as mathematical models can be developed analytically for these processes, they are mostly characterized by extreme complexity and highly nonlinear equations. Moreover, there usually remains an amount of unknown parameters which can only be identiied with a rather low degree of accuracy. For these reasons, biotechnological processes are well suited for beeing modelled by fuzzy systems taking into account the uncertainty in complex processes by their basic conception of handling vagueness 3]. In contrast to traditional fuzzy systems serving as fuzzy controllers, the requirements to be satissed by fuzzy models for real processes are much stronger. Instead of rather qualitative information they have to provide quantitative results simulating the process behaviour as good as possible. For this reason, the well-known way of developing fuzzy systems by heuristic means is no longer practicable. It has to be substituted by an identiication procedure for fuzzy systems deriving the essential system components from the process input and output data. According to the classical way in system modelling, fuzzy modelling consists of the principal phases of structure and parameter identiication. Additionally, these phases are supplemented by the task of deening appropriate fuzzy operators, i.e. the selection of a suitable inference and defuzziication method. The latter problem can be solved successfully on the basis of preliminary considerations which recommend the selection of linear fuzzy operators. Thus, undesirable, additional nonlinearities, introduced by the fuzzy inference method itself, can be avoided. The problem of structure and parameter identiication, however, proves to be the crucial part in fuzzy modelling. Fuzzy structure identiication consists of the deenition of the linguistic variables for the inputs and outputs of the fuzzy system. Additionally, it includes the determination of the number of linguistic values, i.e. the fuzzy sets to be deened over each variable. In contrast, the intention of fuzzy parameter identiication is to nd out the actual position of the fuzzy sets and, in case …