I. Introduction
The feasibility of using adaptive fuzzy control in the most practical applications is largely dependent on computational demands of the associated on-line learning scheme. The computational demands can be reduced by basing the controller on a fuzzy model with a fixed structure and predefined fuzzy sets [9] and by using a fuzzy identification scheme that avoids explicit optimization [5]. Unfortunately, fuzzy models based on predefined fuzzy sets can be computationally inefficient when little prior knowledge is available [2] and non-optimising identification schemes are very sensitive to the quality of the training data [8]. Adaptive fuzzy control based on an evolving fuzzy model [3] may be attractive in such circumstances [1].