On-line determination of steady state in simulation outputs

Abstract

As simulation output is generally nonstationary and autocorrelated and includes the initialization bias we can't directly apply traditional statistic approaches to its analysis. One of solutions for this problem is to exclude the initial part of simulation output that is affected by this bias. Detection of steady state is one of the most important issues for the automation of simulation output analysis. This paper deals with on-line detection of truncation point in order to estimate efficiently the steady-state mean of single-run simulation by batch means method. Two algorithms are purposed. The first algorithm is based on the Euclidean distance equation and the second utilizes the backpropagation algorithm in neural networks that has been applied to the pattern classification problems.