Parameter identification for nonlinear time-varying dynamic system based on the assumption of 'short time linearly varying' and global constraint optimization

摘要

A new identification approach based on a new assumption of "short time linear varying" is proposed for nonlinear time-varying (NTV) dynamic systems. In the identification procedure, the whole period is divided into a series of shifting windows. In each window, the NTV system model, which is known a priori, can be represented by regression equations and all the time-varying (TV) coefficients are determined by a least squares (LS) algorithm. The proposed approach has better identification precision than the traditional assumption of "short time invariant". To enhance the robustness and stability, the problem of parameter identification is solved by means of constrained optimization in the global identification strategy when the noise level increases. The validity and accuracy are verified by applying the method to a single degree of freedom (SDOF) numerical example, and a qualitative analysis on the selection of the window size is carried out in this research.