The application of evolution strategies to the problem of parameter optimization in fuzzy rulebased systems

Fuzzy logic has become widely acknowledged as an important and useful methodology in the design of rule based systems. It allows the representation of imprecise or incomplete knowledge and offers various mechanisms for reasoning with fuzzy data. In comparison to 'classical' rule based systems, only very few rules are needed to describe difficult problems. Nevertheless, in its current form it has several shortcomings: when it comes to the design of membership functions or to actually attaching priorities to the available rules, the choice of numerical quantities for the different parameters which is indispensable for the reasoning process is generally not justified by the results from knowledge acquisition and, what is worse, demands often a long process of iterative improvement to obtain good results. The use of empirically obtained quantitative representations seems questionable because of its high context dependence. The results are in many cases sub optimal systems. It seems natural to try to use a computer and an algorithmic optimization technique for the final adjustment of the parameters. Evolutionary algorithms seem especially appropriate for this task, partly because the fuzzy reasoning process can hardly be described by means of a closed mathematical formula-not to mention differentiability or other 'convenient' mathematical properties-partly because of the opportunity to apply parallel computation in a very natural way which seems essential in the design of large scale systems.