W.C.M. van Beers

Application-driven sequential designs for simulation experiments: Kriging metamodelling

This paper proposes a novel method to select an experimental design for interpolation in simulation. Although the paper focuses on Kriging in deterministic simulation, the method also applies to other types of metamodels (besides Kriging), and to stochastic simulation. The paper focuses on simulations that require much computer time, so it is important to select a design with a small number of observations. The proposed method is therefore sequential. The novelty of the method is that it accounts for the specific input/output function of the particular simulation model at hand; that is, the method is application-driven or customized. This customization is achieved through cross-validation and jackknifing. The new method is tested through two academic applications, which demonstrate that the method indeed gives better results than either sequential designs based on an approximate Kriging prediction variance formula or designs with prefixed sample sizes.

Kriging for interpolation in random simulation

Whenever simulation requires much computer time, interpolation is needed. Simulationists use different interpolation techniques (eg linear regression), but this paper focuses on Kriging. This technique was originally developed in geostatistics by DG Krige, and has recently been widely applied in deterministic simulation. This paper, however, focuses on random or stochastic simulation. Essentially, Kriging gives more weight to ‘neighbouring’ observations. There are several types of Kriging; this paper discusses—besides Ordinary Kriging—a novel type, which ‘detrends’ data through the use of linear regression. Results are presented for two examples of input/output behaviour of the underlying random simulation model: Ordinary and Detrended Kriging give quite acceptable predictions; traditional linear regression gives the worst results.

Constrained Optimization in Simulation: A Novel Approach

This paper presents a novel heuristic for constrained optimization of random computer simulation models, in which one of the simulation outputs is selected as the objective to be minimized while the other outputs need to satisfy prespeci¯ed target values. Besides the simulation outputs, the simulation inputs must meet prespeci¯ed constraints including the constraint that the inputs be integer. The proposed heuristic combines (i) experimental design to specify the simulation input combinations, (ii) Kriging (also called spatial correlation mod- eling) to analyze the global simulation input/output data that result from this experimental design, and (iii) integer nonlinear programming to estimate the optimal solution from the Krig- ing metamodels. The heuristic is applied to an (s, S) inventory system and a realistic call-center simulation model, and compared with the popular commercial heuristic OptQuest embedded in the ARENA versions 11 and 12. These two applications show that the novel heuristic outper- forms OptQuest in terms of search speed (it moves faster towards high-quality solutions) and consistency of the solution quality.

Kriging metamodeling in constrained simulation optimization: an explorative study

This paper describes two experiments exploring the potential of the Kriging methodology for constrained simulation optimization. Both experiments study an (s, S) inventory system with the objective of finding the optimal values of s and S. The goal function and constraints in these two experiments differ, as does the approach to determine the optimum combination predicted by the Kriging model. The results of these experiments indicate that Kriging offers opportunities for solving constrained optimization problems in stochastic simulation; future research will focus on further refining the methodology.

Monotonicity-Preserving Bootstrapped Kriging Metamodels for Expensive Simulations

Kriging metamodels (also called Gaussian process or spatial correlation models) approximate the Input/Output functions implied by the underlying simulation models. Such metamodels serve sensitivity analysis, especially for computationally expensive simulations. In practice, simulation analysts often know that this Input/Output function is monotonic. To obtain a Kriging metamodel that preserves this characteristic, this article uses distribution-free bootstrapping assuming each input combination is simulated several times to obtain more reliable averaged outputs. Nevertheless, these averages still show sampling variation, so the Kriging metamodel does not need to be an exact interpolator; bootstrapping gives a noninterpolating Kriging metamodel. Bootstrapping may use standard Kriging software. The method is illustrated through the popular M/M/1 model with either the mean or the 90% quantile as output; these outputs are monotonic functions of the traffic rate. The empirical results demonstrate that monotonicity-preserving bootstrapped Kriging gives higher probability of covering the true outputs, without lengthening the confidence interval.