VARIANCE REDUCTION TECHNIQUES IN STOCHASTIC SHORTEST ROUTE ANALYSIS

FLOYD HENRY GRANT, Purdue University

Abstract

Variance reduction techniques are of much interest to simulation practitioners. However, they typically present formidable difficulties in their implementation, and are, therefore, not widely used. These difficulties stem from the lack of well defined application procedures for the various techniques which are available. This research focuses on the development of application procedures for three variance reduction techniques: antithetic sampling, control variates, and stratified sampling. These procedures are developed for use in stochastic shortest route analysis for the estimation of network completion time and the probability that a path is shortest. The procedures developed are supported by sound theoretical development which assures that no technique will result in a variance increase, and, if any internal correlation exists, reduced variance in our estimates will be realized. Further, one technique guarantees a reduction in variance in the estimates obtained over direct simulation. We also provide extensive application results over a wide range of network characteristics. The results of these applications indicate that all of the variance reduction techniques investigated work well. Further, one technique, antithetic sampling, was found to be far superior to the other techniques and can be generally recommended for use. The procedures developed have been implemented in a generalized program. This program automates the application procedures and makes them transparent to the user. The development of these variance reduction techniques application procedures coupled with their implementation in an easy to use program should be exciting news for the simulation practitioner.

Degree

Ph.D.

Subject Area

Operations research

COinS