COMPARISON OF SPACE-FILLING DESIGNS IN DISCRETE DOMAINS

Abstract eng:
In order to properly explore response of a model, one needs to perform simulations for a set of design points. When dealing complex non-linear models, the simulations are usually very timeconsuming, hence the number of simulations performed within a limited time is rather low. Randomly chosen design points do not ensure the observed properties will be captured properly. Therefore, the design points must be chosen carefully. The motivation of the presented contribution is to investigate methods, which are suitable for generating designs in discrete parameter space, where each parameter can attain different number of levels, because commonly used software based on Latin Hypercube Sampling fails in solving such a situation. Hence, we compare here several well-known metrics for assessing optimal designs as for instance the Euclidean maximin distance, the maximum pairwise correlation or the D-optimal criterion. The resulting optimal designs are consequently employed for the evaluation of the stochastic sensitivity analysis so as to investigate their ability in prediction of the ‘parameter-response’ correlations.

• Export as: BibTeX | MARC | MARCXML | DC | EndNote | NLM | RefWorks
• Add to your basket