New approach for nonlinear modelling based on online designing of the fuzzy rule base

Abstract

The problem of online nonlinear modelling emerges among others from limitations of memory. This problem is often solved by using evolving systems. Evolving fuzzy systems play significant role as they are distinguishable by clear representation of knowledge (by fuzzy rules) which allows an interpretation of their behavior. The structure and the parameters of those systems can be selected online. Moreover, the fuzzy rules can represent operating points of modeled object, which can also be identified online. Then, the data from identification can be used for learning. In this paper we proposed an evolving fuzzy system for nonlinear modelling with endless number of steady states and negligible time of non-steady states. It is based on analysis of firing level of the fuzzy rules with possibilities of background learning.