Stochastic suitability measures for nonlinear structure identification

In this paper stochastic suitability measures are introduced as a means of quantifying the ability of a particular nonlinear model class to capture the control relevant I/O-behavior of a nonlinear system to be identified. These suitability measures can be used in the structure identification step that usually precedes the actual parameter identification. Properties of these measures are discussed and compared to their deterministic counterpart and the qualitative dependence on model classes and classes of input sequences is made explicit with two examples.