APPROACHES TO AID DESIGN AND DEVELOPMENT OF FUZZY RULE BASED SYSTEMS

Abstract

The initial stages of fuzzy rule based design depends to some extent on intuition and the designer's feeling of the nature of the problem. It often becomes difficult for the designer to put these "feelings" on paper as a proper set of rules and variables. Thus, the fuzzy model of the system under consideration will be modified quite often during design, and for each case the system has to be analysed for Its performance, which is again a tedious task. An attempt is made in this dissertation work to evolve some ways to simplify the task of designer, or to help him in his task. Two ways have been evolved - first in the form of a fuzzy system development tool SOFTFUZZ and second in the form of an algorithm which converts a set of numerical data into a set of rules. SOFTFUZZ allows the user to define the structure of his fuzzy system in a simple and convenient manner, while providing the facility to modify the information frequently and easily, It thus allows the user to generate the knowledge base, and the rest of the components of a fuzzy rule based system are incorporated in-it. Thus it becomes a complete fuzzy system which can be run and observed for its performance. The case studies undertaken to test the algorithm which generates rules from numerical data cleary prove the effectiveness of the algorithm. The algorithm equips fuzzy systems with the ability to learn by examples so that they can compete with neural networks in areas where direct acquisition of rules is difficult or impossible.