Abstract
In this paper, a hybrid fuzzy modeling technique is described for an unknown system with a given set of numerical data. Nonlinear systems are difficult to model by conventional fuzzy systems because of problems such as the conflict between overfitting and underfitting, and low reliability. To overcome these problems, a great number of fuzzy rules and very complicated learning algorithms must be used. We propose the hybrid fuzzy modeling technique, which the combination of the fuzzy system and self-organizing approximators (polynomial neural networks: PNN). Fuzzy systems have been used successfully for imprecise data or not well-defined concepts. PNN is an analysis technique used to identify nonlinear relations between system inputs and outputs and build hierarchical polynomial regressions of required complexity. Comparative studies of the proposed approach are presented for both Box-Jenkin data identification system and three-input nonlinear function to show the performance.
The proposed method was efficient and much more accurate than previous other models because it used fewer fuzzy rules and had better generalization ability.