An entropy-like estimator for robust parameter identification

Abstract

This paper describes the basic ideas behind a novel prediction error parameter identification algorithm exhibiting high robustness with respect to outlying data. Given the low sensitivity to outliers, these can be more easily identified by analysing the residuals of the fit. The devised cost function is inspired by the definition of entropy, although the method in itself does not exploit the stochastic meaning of entropy in its usual sense. After describing the most common alternative approaches for robust identification, the novel method is presented together with numerical examples for validation. © 2009 by the author; licensee Molecular Diversity Preservation International, Basel, Switzerland.