Abstract:

We have developed a Neuro-Fuzzy-Perceptron which combines the capability of neuronal networks to learn with the rule system of a Fuzzy Controller. Expert knowledge and company strategies for analyses can easily be modeled which -despite their complexity- always remain transparent. This article focuses on the human interface which enables the expert to build complex models. We have learned that fuzzy logic only will be accepted as a controlling tool, if an easy to use graphical interface is provided for the decision makers. In most cases, gaining expert knowledge is a try and error process, in which the expert reflects his rule base for the first time. While trying out some little changes in his initial settings, he gets a feeling of the influences of the changes. An abstraction of his knowledge is built in steps. It is important to have an interface in which the expert intuitively finds the spots where to he has to change the initial setting to satisfy his wishes.