Global approaches to identifiability testing for linear and nonlinear state space models

Whenever a model is not identifiable, there exist several parameter vectors θ which yield the same input-output behavior, so that any further interprtation of θ is questionable. In this paper two methods which can be used to test state-space models for identifiability with a global approach are presented. While the first one is useful for nonlinear and/or time-varying models, the second one is suitable for linear time-invariant models. An appendix provides tools for the solution of he sets of polynomial equations involved in both methods.