Quantifying input uncertainty in an assemble-to-order system simulation with correlated input variables of mixed types
Date
2014Source Title
Proceedings of the 2014 Winter Simulation Conference, WSC 2014
Print ISSN
0891-7736
Publisher
IEEE
Pages
2124 - 2135
Language
English
Type
Conference PaperItem Usage Stats
224
views
views
278
downloads
downloads
Abstract
We consider an assemble-to-order production system where the product demands and the time since the last customer arrival are not independent. The simulation of this system requires a multivariate input model that generates random input vectors with correlated discrete and continuous components. In this paper, we capture the dependence between input variables in an undirected graphical model and decouple the statistical estimation of the univariate input distributions and the underlying dependence measure into separate problems. The estimation errors due to finiteness of the real-world data introduce the so-called input uncertainty in the simulation output. We propose a method that accounts for input uncertainty by sampling the univariate empirical distribution functions via bootstrapping and by maintaining a posterior distribution of the precision matrix that corresponds to the dependence structure of the graphical model. The method improves the coverages of the confidence intervals for the expected profit the per period.
Keywords
Graphic methodsConfidence interval
Dependence measures
Dependence structures
Empirical distribution functions
Input distributions
Posterior distributions
Simulation outputs
Statistical estimation
Distribution functions