Abstract: A new method is presented to identify the structure and parameters of nonli near systems in this paper. Because of the specialties of time-variable systems, an efficient on-line method for both criterion function and parameter identification is obtained by means of the recursive least square method with the forgetting factor, at the same time, the model structure is determined. To reduce the quantity of computation, the Cholesky factorization is used in model estimation, and the proof of the convergence of the algorithm is presented. The new method is applied to the on-line estimation of the F-8 Crusader aircraft longitudinal kinetics model. The results show that the models represented by polynomial and splint functions can be determined efficiently.

Key words: on-line estimation, identification of nonlinear systems, Cholesky factorization