Implementation of fuzzy inference with neural network: the NNFI structure
Shyh-Yeong Shu, Chung-Mu Hwang · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1993
In many fuzzy system applications, the most difficult and time consuming problem is to built the fuzzy rule base. Usually, to build fuzzy rule base depends on a domain expert to reflect his experience. But for a complicated system, it is sometimes difficult for an expert to describe clearly the causal relationships among those linguistic variables. To overcome such a problem, a dense connectionist structure of artificial neural network, called as NN-Fuzzy Inferencer (NNFI), is constructed to implement the fuzzy inference. This NNFI incorporates the effects of neural network and fuzzy inference. It is trainable and gets a more desired output value than backpropagation neural network does. The idea of the NNFI architecture is driven from the traditional fuzzy inference method. It can avoid not only the difficulty that for a designer to define the casual relations between the input variables and output variables, but also determine the membership function for each linguistic value. Furthermore, the system will generate the weighting coefficients in antecedent part and consequent part respectively in every fuzzy rule.