System identification, model simplification and the design objective

Models and model quality are prime concerns for most design issues in control and system analysis. The success of a simulation study hinges upon the reliability of the model used. In this contribution we discuss how to build mathematical models that given certain constraints, are of optimum quality for a prespecified application. We then taken into account the influence of both bias errors and random errors on the model. It turns out that for a fairly broad class of identification methods in the prediction error family, the optimal choices of design variables can be given in an explicit form.