Neuro-Fuzzy Modeling for strongDynamic Systemstrong Identification - Fuzzy.pdfVIP

NEURO-FUZZY MODELING FOR DYNAMIC SYSTEM IDENTIFICATION JYH-SHING ROGER JANG CS Dept., Tsing Hua Univeristy, Hsinchu, Taiwan This paper presents the continued work of a previously proposed ANFIS (Adaptive Neuro-Fuzzy b z z y Inference System) architecture with emphasis on the applica- tions to dynamic system identification. We demonstrate the use of ANFIS for the hair dryer modeling problem and compare its performance with the ARX model. 1 Introduction System identificatiod is the process of constructing a model to predict the behavior of a target system. Conventional system identification techniques are mostly based on linear models with fast computation and rigorous mathemat- ical support. On the other hand, neuro-fuzzy modeling represents nonlinear identification techniques that require massive computation but without math- ematical proofs of convergence to global minimaor the like. This paper applies two representative approaches (ANFIS and ARX) from both disciplines and compare their performance on a classic system identification problem of hair dryer modeling!. This paper is organized into five sections. In the next section, the basics of ANFIS are briefly introduced. Section 3 explains the problem of hair dryer modeling and how to use the ARX model to find a linear model. Section 4 exhibits the use of ANFIS for the same problem and compare the results with the ARX model. Section 5 gives concluding remarks. 2 ANFIS A first-order Sugeno fuzzy inference system? with two fuzzy rules can be ex- pressed as Rule 1: If X is AI and Y is B1, then fl =p l z + q1y + TI, Rule 2: If X is A2 and Y is B2,then fz =pax + qzy + rz