An efficient fuzzy neural modeling approach using the fuzzy curve concept

A novel modeling technique based on the fuzzy curve concept is suggested in this paper, for generating fuzzy models composed of Takagi-Sugeno rules. This method exhibits a number of significant attributes, such as effective input space searching, computational simplicity and high accuracy of the resulting fuzzy models. The premise space partitioning problem is effectively solved by segmenting the fuzzy curves into a certain number successive, linear segments. Then, an ordered tree is generated which provides the number of rules and the proper rule co-ordinates along each axis. The rule output hyper-planes are correctly oriented in the output space using the RLSE method. The validity of the suggested modeling approach is demonstrated using a simple static example and the well known gas furnace problem