Abstract
This paper considers system identification for linearly approximated models. Linear approximation models are useful for identification, but their accuracy may not be estimated by the conventional linear identification methods. This paper proposes a method to evaluate not only the system parameters but also the influence of the linear approximation errors in identification. The method is based on particle filters, which are known for its applicability to a wide class of nonlinear systems. Numerical examples are given to demonstrate the effectiveness of the proposed method in detail. Furthermore, experimental validation is performed for a simple pendulum system.
