Recurrent neuro-fuzzy systems

In this paper we introduce a new architecture called recurrent neuro-fuzzy (RNF) system which enhances the modeling capabilities of fuzzy systems with the dynamic behavior of recurrent neural networks (RNN). In a general sense, the architecture of RNF is similar to other adaptive neuro-fuzzy systems. It has a rule-base, a database, an inference engine, and a learning mechanism. In this paper we will emphasize those portions which are different that other approaches, specifically, the construction and operation of recurrent rules and the learning mechanism which is used in determination and adaptation of system parameters. The fundamental concepts of the RNF system are demonstrated using a two-link robot example.