Customized sequential designs for random simulation experiments : Kriging metamodeling and bootstrapping

摘要: This paper proposes a novel method to select an experimental design for interpolation in random simulation, especially discrete event simulation. (Though the focuses on Kriging, this approach may also apply other types of metamodels such as non-linear regression models and splines.) Assuming that simulation requires much computer time, it is important with small number observations (or runs). The proposed therefore sequential. Its novelty accounts specific input/output behavior response function) particular at hand; i.e., customized or application-driven. A tool customization bootstrapping, which enables estimation variances predictions inputs not yet simulated. tested through two classic models, namely expected steady-state waiting time M/M/1 queuing model, mean costs terminating (s, S) inventory For these indeed gives better results than popular alternative design, Latin Hypercube Sampling (LHS) prefixed sample.