Identification of expensive-to-simulate parametric models using Kriging and stepwise uncertainty reduction

This paper deals with parameter identification for expensive-to-simulate models, and presents a new strategy to address the resulting optimization problem in a context where the budget for simulations is severely limited. Based on Kriging, this approach computes an approximation of the probability distribution of the optimal parameter vector, and selects the next simulation to be conducted so as optimally to reduce the entropy of this distribution. The identification of the parameters of a non-uniquely identifiable continuous-time state-space model is used to illustrate the method.