Regenerative Simulation Using Internal Controls.
Abstract
A new procedure is developed to exploit the asymptotic properties of appropriately defined controls in order to improve the efficiency of system performance estimators obtained from a simulation experiment . Two types of standardized variates are formulated and their joint convergence in distribution to multivariate normality is established. This result is the basis for applying multinormal regression theory to the construction of valid confidence intervals for steady-state parameters. A two-stage procedure is developed. In the first stage, standardized flow variates are applied to the denominator of the classical regenerative ratio estimator to reduce its inherent bias. In the second stage, standardized service-time variates are applied to the numerator of the new ratio to reduce the variance of the resulting double multivariate ratio estimator. A new multivariate test for normality is also developed based upon the univariate Shapiro-Wilk test. The complete variance reduction procedure is validated on an analytically tractable closed Jackson queueing system. On the basis of experimental results together with the established theoretical characteristics of the technique, guidelines are presented for the practical application of the method.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1982
- Accession Number
- ADA119740
Entities
People
- Jan Denise Eakle-cardinal
Organizations
- Air Force Institute of Technology