Nonlinear dynamic system identification based on wavelet approximation

Proposes a multiscale approach for nonlinear dynamic system identification. Based on wavelet multiresolution analysis theory, model parameters are estimated step by step from coarse to finer scales. Some criteria and empirical formulae are discussed that explain how to select initial scale, approximating resolution and size of identification samples. Further, the fundamental frame and identification algorithms of the proposed method are described. The computing complexity and local maximum of neural network methods can be avoided. Finally, the convergence and precision of the identification algorithm are analyzed. The simulating case study on a nonlinear dynamic system shows the high efficiency and effectiveness of this methodology