Building fuzzy systems by soft competitive learning

Introducing learning paradigms of neural networks into fuzzy systems is one of the approaches to integrating fuzzy systems with neural networks. Following this guideline, this paper presents an approach to the problem of modeling an unknown system by a fuzzy rule-based model from measured data through soft competitive learning. We address two fundamental issues associated with the rule-based modeling: rule-base construction and rule-base manipulation. By employing fuzzy concepts and competitive learning, a two-step approach consisting of a principal and refining algorithm has been suggested to extract rules from available data set. An optimal algorithm is developed for manipulating the obtained rule-base with novel data. Simulation results on three examples taking from function approximation, time-series prediction, and nonlinear dynamical modeling are given