Detecting the duration of initial transient in steady state simulation of arbitrary performance measures

Abstract

The issue of the initial transient phase in steady state simulation has been widely discussed in simulation literature. Many methods have been proposed for deciding the duration of this phase of simulation, to determine a valid truncation point of the transient portion of output data. However, practically all these methods can only be used in simulations aimed at estimation of mean values. In this paper, we show that analyses of performance measures which do not represent mean values require different solutions, as the rate of convergence to steady state is different for mean values than, for example, for quantiles. We describe and present additional results for a new method of determining the duration of initial transient phase which can be applied in analysis of steady state quantiles and probability distributions. The method appears robust and applicable in analysis of arbitrary performance measures. Copyright 2007 ICST.