Uncertainty equivalent reduced order models for discrete systems

Model reduction invariably involves a trade off between the available information and the simplicity of the retained model. This information loss leads to an uncertainty about the output of the true system given the output of the reduced model for identical inputs. In this paper a novel approach to the reduction problem is given by incorporating this induced uncertainty in the reduced model. Geometrically, the idea is to construct a tube centered on the model output, to which the actual system ouput belongs with a high degree of confidence.