An empirical evaluation of a factor effects screening procedure for exploring complex simulation models
Loading...
Authors
Bosché , Kerry N.
Subjects
Advisors
Sanchez, Susan M.
Date of Issue
2006-06
Date
Publisher
Monterey California. Naval Postgraduate School
Language
Abstract
Screening experiments are procedures designed to identify the most important factors in simulation models. Previously proposed one-stage procedures such as sequential bifurcation (SB) and controlled sequential bifurcation (CSB) require factor effects to be arranged according to estimated sign or magnitude prior to screening. FF-CSB is a two-stage screening procedure for simulation experiments proposed by Sanchez et al. (2005) which uses an efficient fractional factorial experiment to estimate factor effects automatically, removing the need for pre-estimation. Empirical results show that FF-CSB classifies factor effects as well as CSB in fewer runs when factors are only grouped by their sign (positive or negative). In theory, the procedure can achieve more efficient run times when factors are also sorted by estimated effect after the first stage. This analysis tests the efficiency and performance characteristics of a sorted FF-CSB procedure under a variety of conditions and finds that the procedure classifies factors as well as unsorted FF-CSB with significant improvement in run times. Additionally, various model- and user-determined scenarios are tested in an initial attempt to parameterize run times against parameters known or controlled by the modeler. Further experimentation is also suggested.
Type
Thesis
Description
Series/Report No
Department
Organization
Naval Postgraduate School (U.S.)
Identifiers
NPS Report Number
Sponsors
Funder
Format
xvi, 35 p. ;
Citation
Distribution Statement
Approved for public release; distribution is unlimited.