Abstract: This paper proposed SVD-TLS and EKF based dynamic self-organizing fuzzy neural network (STD_DSFNN)for optimizing fuzzy rules and Adjusting nonlinear and linear parameters reasonably. Firstly, the structure and meanings of each layer arc given. hhen nonlinear parameters arc learned by using EKF algorithm, the linear parameters are learned by using SVD_TLS algorithm which also extract important rules at the same time. At last, the STE_DSFNN is verified through the typical Machcy-Glass time series prediction examples. The results show that the STE_DSFNN network structure is more compact and has better generalization ability compared with the DFNN, ANFIS and UKF_DFNN.

Key words: SVD_TLS algorithm, EKF algorithm, Dynamic self-organizing fuzzy neural network