Takagi-Sugeno recurrent fuzzy neural networks for identification and control of dynamic systems

Abstract

In this paper, we propose a Takagi-Sugeno recurrent fuzzy neural networks (TSRFNN) for identification and control nonlinear dynamic systems. The TSRFNN combines the recurrent multilayered connectionist network with dynamic Takagi-Sugeno (TS) fuzzy model. The temporal information are embedded in the recurrent structure by adding feedback connections between the states layer and inputs layer of the fuzzy neural networks (FNN). Based on the derived dynamic backpropagation (DBP) and recursive least squares (RLS) algorithms, the parameters in the TSRFNN are on-line adjusted. Compared with the traditional recurrent fuzzy neural networks (RFNN), the proposed TSRFNN has not only a smaller network structure and a smaller number of network parameters, but also a fast convergence speed and a better learning performance.