2001 Volume 13 Issue 3 Pages 322-333
A modified Takagi-Sugeno fuzzy model that makes the estimation error eventually close to zero by adding compensation rules is proposed. In order to achieve a given estimation error (arbitrary small), input space is divided into a set of linear domains in which the estimation error is acceptable, and others called a set of fuzzy domains. As for the fuzzy domain, the compensation rules with a partially linear membership function are added so that the given estimation error can be achieved as a whole. The effectiveness of the proposed method is confirmed by the experiment on a widely used 2-input non-linear function by comparing the results given by conventional 3 methods. Then the experiment with Box and Jenkins gas furnace data also shows that the proposed fuzzy model achieves a MSE (mean squared error) of 10^<-2> order with 4 rules, which is equivalent to the results of conventional methods, and that MSE of 10^<-17> order is achieved with 44 rules.
