Global derivative based sensitivity method for parameter estimation


Sulieman H., Kucuk I.

JOURNAL OF THE FRANKLIN INSTITUTE-ENGINEERING AND APPLIED MATHEMATICS, cilt.348, ss.1556-1573, 2011 (SCI İndekslerine Giren Dergi) identifier identifier

  • Cilt numarası: 348 Konu: 7
  • Basım Tarihi: 2011
  • Doi Numarası: 10.1016/j.jfranklin.2010.05.014
  • Dergi Adı: JOURNAL OF THE FRANKLIN INSTITUTE-ENGINEERING AND APPLIED MATHEMATICS
  • Sayfa Sayısı: ss.1556-1573

Özet

In nonlinear parameter estimation local sensitivity assessment; conventionally measured by the first-order derivative of the predicted response with respect to a parameter of interest fails to provide a representative picture of the prediction sensitivity in the presence of significant parameter co-dependencies and/or nonlinearities. In this article we derive the profile-based sensitivity measure developed by Sulieman et al. (2001, 2004) [1,21 in the context of model re-parameterization. In particular, the so-called predicted response re-parameterization is shown to ultimately lead to the profile-based sensitivity coefficient defined by the total derivative of the model predicted response with respect to a parameter. Although inherently local, the profile-based measure is shown to handle simultaneous perturbations in parameter values while accounting for their co-dependencies. Thus the proposed measure possesses a central property of a global sensitivity measure and so it is considered hybrid local global measure.