METAMODELING SAMPLING CRITERIA IN A GLOBAL OPTIMIZATION FRAMEWORK

The use of approximate models or metamodeling has lead to new areas of research in the optimization of computer simulations. Metamodeling approaches have advantages over traditional techniques when dealing with the noisy responses and/or high computational cost characteristic of many computer simulations, most nota-bly those in MDO. While a number of methods in the literature discuss how to exploit the benefits of meta-modeling approaches, one particular algorithm, Effi-cient Global Optimization (EGO), is the focus of this paper. Specifically, we look at the criteria used by the algorithm to select additional points to add to the data set used in fitting the metamodel. In addition to modifi-cations to the original criterion, three criteria originally proposed for use in infill sampling in the field of geosta-tistics are explored. The impact of these criteria on the search strategy of EGO is examined through several analytical examples. In addition, several enhancements to EGO are explored. Finally, a case study is presented using a computer simulation to predict the fuel economy of hybrid vehicles. 1