Fuzzy modeling by machine learning and its application to prediction of heater temperature

Knowledge acquisition is a critical stage in developing expert systems. ID3 is an approach to overcome it. ID3 can create crisp IF-THEN rules automatically based on the entropy of cases. However the number of rules increases when there are many cases even if they have slightly different values. The fuzzy modeling technique is applied to many fields because it can represent fuzzy knowledge. In order to apply fuzzy modeling to the real world problems, identification of variables in the premise part is essential. It is, however, difficult because it involves many numbers of variables in the real world. Therefore this paper proposes a practical method to identify effective variables in a premise part of fuzzy IF-THEN rules. This method is to: (1) Calculate priorities of variables and boundary values for fuzzy sets by information contents and identify effective variables in a premise part by the priorities and operators’ experience. (2) Identify consequence part using multiple regression analysis in subspace determined by a premise part. This method is applied to predicting heater outlet temperature by crude methods and their rates and/or feed rate and so on. And we have a good result, that is, (1) We created practical fuzzy model to predict heater outlet temperature and (2) It has shown that enough accuracy without adding new rules or without modifyng rules for two and a half years. [In Japanese.]