STOCHASTIC CO-KRIGING FOR STEADY-STATE SIMULATION METAMODELING

In this paper we present the stochastic co-kriging methodology (SCK) for approximating a steady-state mean response surface based on outputs from both long and short simulation replications performed at selected design points. We provide details on how to construct an SCK metamodel, perform parameter estimation, and make prediction via SCK. We demonstrate numerically that SCK holds the promise of providing more accurate prediction results at no additional computational effort by only externally adjusting the simulation runlength and number of independent replications of simulations through the experimental design of the simulation study.