Abstract
Determining optimal structure identification of a nonlinear black box system is one of the most significant steps in Fuzzy modelling based on Takagi-Sugeno models. The one parameter that needs to be determined before performing fuzzy clustering and identification algorithms is the optimal number of clusters. In this paper, we present a new approach for automatically predicting the optimal choice of this parameter simply by detecting the general trend of the data structure. First, an approximation model of the system is built by fitting data to a polynomial function. Second, a preliminary decomposition of the data is realized based on detection of the function turning points. Then, a merging method is adopted to reduce the identified number of clusters. The advantage of the generated solution is that it remains in the horizon of the data; hence there is no need to apply heuristic rules or conventional validation tools. The performance of the proposed method is evaluated for both quality of clustering and fuzzy modeling with 1^st order TS systems using GK fuzzy algorithm .

Keywords: Fuzzy clustering, optimal clusters number, nonlinear system, polynomial regression, Takagi-Sugeno models, GK algorithm.

(P1110906635, 584 KB)