A C-Means Clustering Based Fuzzy Modeling Method - Fuzzy Systems, 2000. FUZZ IEEE 2000. The Ninth IEEE International Conference on

This paper proposes a new neuro-fuzzy method to model the dynamic behavior of com'plex system.s based on real experimental data. First, we investigate the firing strength of rules by a fuzzy C-means clustering method. Then, we retrieve the membership functions of input variables by a neuro-fuzzy network. Finally, we identify the parameters of linear local models by recursive least squares. I n particular, we applied this method to construct the dynamics of a boiler combustion process. I . INTRODUCTION . The dynamics of the most industrial plants are highly nonlinear and uncertain. Therefore, it is very difficult to investigate their dynamic behavior 'by using traditional modeling approaches. For several years, different fuzzylogic-based models have been developed to cope with nonlinearity and uncertainty[ 1-51. Of them, fuzzy TakagiSugeno-Kang (TSK) models [2-31 are widely used in control engineering. Basically, a fuzzy TSK model can be expressed by a set of following typical rules Rule, : lFx,isA,,,ANDx,isA,., A N D . . . A N D X , ~ S A / . ~ THEN -v/ = a/o + u / , x ~ + u , , x ~ + ... + u ~ K x K (1) I = I , 2, "., L For I= I , 2, a * * , L. Where A , , is one of M Gaussian membership functions associated with input variable x, (kl, 2;-., K ) , defined by. A global model can be expressed by: (2) G(x) = e ( x a ) ' / ~ 2