Qualitative System Description Based on Knowledge and Numerical Data and Its Application to the Design of a Fuzzy Controller

Abstract

Fuzzy modeling, based on input-output(I/O) data pairs, is a method for identifying an unknown 'black-box system'. With a fuzzy model humans can qualitatively understand a system's I/O relationship. Usually we have some knowledge about a system we wish to model. Hence we can say that almost all systems are 'grey-boxes'. In effect fuzzy modeling enables a human to achieve a deeper understanding of grey-box system. In this paper we propose a method of qualitative system description based on both knowledge and numerical data. First we identify a numerical data-based I/O model and a knowledge-based model. Then the two models are combined by comparing each others structure. At this point we can make an abduction to settle our lack of knowledge. A description of the system will be generated, with the users register level in consideration, by simulating the combined model. We apply our method to a fuzzy controller design by first constructing a control model of the combined model and then designing a fuzzy feedback controller based on some control description. Finally we apply our technique to a helicopter system.