Parameter identification for nonlinear models from a state-space approach ?

摘要

A new approach to parameter estimation of dynamical models is proposed. Its objective is to approximate at best the different dynamics of the system, instead of approximating at best the system output in time. This leads to a weighting of the error depending on the samples location in the state-space and input space. A possible implementation is proposed and applied for estimating the parameters of a two degrees of freedom vehicle dynamics model. The proposed approach is shown to better approximate the fast transient dynamics, at the cost of a degraded performance on steady-states.