Set-Theoretic Measures as Evolutionary Fitness Criteria in Nonlinear System Identification*

The paper reports a method for identification of parametric models that are linear and time-invariant in parameters, but arbitrarily nonlinear in signals. Set-bounding solutions are exploited to simultaneously identify and parametrize the model structure. Measures of set-solution quality are used as fitness measures in evolving the structure. Experiments verify effective NARX and NARMAX identification in complex unmodeled disturbances.