Analysis of nonstationary stochastic simulations using classical time-series models

Abstract

This article extends the use of classical autoregressive and moving average time-series models to the analysis of a variety of nonstationary discrete-event simulations. A thorough experimental evaluation shows that integrated and seasonal time-series models constitute very promising metamodels, especially for analyzing queueing system simulations under congested or cyclical traffic conditions. In some situations, stationarity-inducing transformations may be required before this methodology can be used. Our approach for efficient estimation of meaningful performance measures of selected responses in the target system is illustrated using a set of case studies taken from the simulation literature.