Comparisons via Stochastic Simulation, with Applications to Manufacturing and Services

点击翻译按钮获取中文摘要

Nelson 9622065 A great many manufacturing and service systems are so complex that the only analysis tool is discrete-event-stochastic simulation. Although simulation is a powerful tool, it suffers from the shortcoming that it is an experimental (not analytical) analysis and requires proper statistical analysis of its output. Since here, the experimenter has complete control over the experimental environment, much of classical statistics, with its emphasis on independent observations, is not applicable, as it often behooves the experimenter to purposely introduce correlation as a variance reduction technique. This research concerns development of procedures to screen a large number of systems, to find the best system, to compare alternative systems to a known standard and to compare alternative systems to a default. There are many classical ranking and selection procedures, but again, these assume the usual independence assumptions that are not always present in discrete-event, stochastic simulation. Most of the classical procedures also assume normality of output data, and are developed to measure population means only. More often than not, output data from stochastic simulations are not normal, and often, measures other than mean responses, are of interest (e.g., the probability of an item spending more than t months in production). Statistical efficiency, functional relationships among systems to be compared, non-standard performance measures, basic theory and implementation in the form of incorporation into existing popular commercial simulation software will be considered in the research. With the advent of easy-to-use simulation software, it is relatively easy to model a rather complex system via discrete-event-stochastic system simulation. However, most packages have no really adequate statistical analysis software built-in. What this research will accomplish is not only the development of appropriate and efficient statistical procedures, but also procedures which can be inco rporated into software packages and will be transparent to the users, providing significant advancement for this very important area for studying complex manufacturing and service systems.