Constructing a kind of fuzzy systems based on neural networks techniques

In this paper, a method to model a kind of nonlinear fuzzy system with fuzzy border is presented. BP networks (BPN) possess efficient capability of approximating nonlinear function and radius base function networks (RBFN) have the fast training speed. These advantages of BPN and RBFN can be combined with clustering techniques to improve system modeling. Firstly, the system structure is obtained by clustering. Secondly the BPN is employed to generate rule base's antecedent function and RBFN to approximate each rule's conclusion function, respectively. So the initial construction of the system can be acquired. Thirdly, structure design and training of networks are discussed in detail. Finally, the structure optimization and overstudy of RBFN are discussed.