Use this URL to cite or link to this record in EThOS:
Title: Empirically derived methods for analysing simulation model output
Author: Mejia, Alicia
Awarding Body: London School of Economics and Political Science (University of London)
Current Institution: London School of Economics and Political Science (University of London)
Date of Award: 1992
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Often in simulation procedures are not proposed unless they are supported by a strong mathematical background. As will be shown in this thesis, this approach does not always give good results when the procedures are applied to complex simulation models, especially on output analysis. For this reason we have used an empirical rather than a theoretical approach for dealing with some of the output problems of simulation. The research carried out has dealt mainly with queuing networks. The first problem we address is that of the identification of possible unstable queues. We also deal with the problem of the identification of queues that may require a long simulation run length to reach the steady state. The method of replications is used for the estimation of terminating and sometimes of steady state parameters. In this thesis we study the relationship that exists between the number of replications used in the simulation and the simulation run length required for the parameter being estimated to reach the steady state. We also study the influence of the random number streams on the values of the mean estimates as a function of the number of replications. One of the most commonly discussed problems related to the estimation of steady state parameters is that of the initialisation bias problem. Two methods are proposed in this thesis to deal with this problem. In one of the methods we propose an effective procedure that can be used for the estimation of the number of initial observations that are to be deleted. The second method, is based on a basic forecasting technique called weighted averages and does not require the elimination of any of the initial observations. Another topic that has been studied in this thesis is the batch means method which is employed for the estimation of steady state parameters based on a single but very long simulation run. We show how a new sampling method called Descriptive Sampling is well suited for the estimation of steady state parameters with the batch means method. We also show how some of the procedures proposed in the literature for use in the batch means method do not work well in simulation models for which no analytical answer exists. The thesis demonstrates that empirically derived methods can be practically effective and could form future theoretical research.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available