A Rule-Based Fuzzy Model for Nonlinear System Identi cation

This article discusses a rule-based fuzzy model for the identi cation of nonlinear MISO (multiple input, single output) systems. The dis cussed method of fuzzy modeling consists of two parts: structure modeling, i.e. determing the num ber of rules and input variables involved respec tively, and parameter optimization, i.e. optimizing the location and form of the curves which describe the fuzzy sets. For structure modeling we use a modi ed TSK-model. The TSK-model was rst proposed by Takagi, Sugeno, and Kang in [1], [2]. For parameter optimization we propose the use of RPROP, a powerful optimization technique orig inally designed for Neural Network training (see also [3], [4]). We applied RPROP to the modi ed version of the TSK-model, implemented the algo rithm, and tested its performance [5], [6]. In this article we focus on the structure modeling part and show by an example how this structuring algorithm performs an input space partitioning.