Implement TSK Model Using a Self-constructing Fuzzy Neural Network

This paper proposes an approach to implement TSK model by using a self-constructing fuzzy neural network (SCFNN). This network is built based on ellipsoidal basis function (EBF), which can be divided into two parts. The first hidden layer composed of EBF units is considered as IF-part, and the output layer which consists of the connect weights is the THEN-part. The structure of SCFNN can adjust adaptively by a new structure learning algorithm based on the proposed crucial factor which denotes the importance of a fuzzy rule. Thus, a rule can be generated or pruned automatically according to both the firing strength of the rule and the performance of SCFNN. Simulation results show that the SCFNN has the powerful capability to extract fuzzy rules in the network. Comprehensive comparisons with other approaches indicate that the proposed method is better considering the learning efficiency and actual effect.

I. Introduction

Fuzzy models have been used widely because they are able to handle the complex nonlinear problems.