Thesis

Contributions towards the operationalisation of empirical bayes for discrete event simulation

Downloadable Content

Download PDF
Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2026
Thesis identifier
  • T17599
Person Identifier (Local)
  • 200883350
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • Discrete event simulation (DES) is a well-established methodology in operational research and management science, facilitating the design, analysis and improvement of complex real-world systems. In many cases the nature of the underlying system results in DES models which are large in scale, complex, and computationally expensive to run. These attributes complicate their use in practice and necessitate the careful design and analysis of DES experiments to ensure valid and efficient inference concerning DES model quantities of interest. It is envisaged that empirical Bayes (EB) methods could provide a welcome addition to the DES practitioner’s toolkit of experimentation methods. EB methods have been much used in recent years to exploit the parallel structures of the largescale data sets arising through the use of modern scientific technologies such as microarrays and FMRI. Whilst the structural similarities between problem settings lends intuitive support to the idea that EB methods may be of benefit, implementation is unlikely to be trivial. This thesis presents an investigation into the utility of adopting an EB approach to DES model experimentation. The specific contributions are discussed in more detail next. We first present a computational study examining the application of EB to DES model experimentation. This study establishes proof of concept by demonstrating that substantial efficiency gains are possible, while also identifying key practical challenges. These challenges motivate the need for greater operationalisation to support practitioners applying EB in DES contexts. The first contribution of this thesis is the design and development of a decision support tool to guide practitioners on the suitability of EB. This involves the determination of the factors of relevance in measuring EB suitability, and the development of a predictive statistic for the relative performance of EB versus traditional approaches in the DES experimentation context. Towards the second contribution, this thesis investigates adapted EB procedures, designed and developed to overcome issues and challenges identified in the DES experimentation context. This work involves the novel use of a classical weighting mechanism, allowing the data from model scenarios to be weighted according to their similarity, limiting unhelpful bias and ensuring robustness for DES practice. The final contribution of the thesis is the presentation of a numerical study demonstrating the use of the methods developed on a large-scale, industrial DES simulation model. Whilst the earlier numerical testing of the methods developed demonstrates their statistical efficiency, this study allows us to demonstrate their applicability to and practical value within a real-world DES context. Taken together, these contributions demonstrate the potential for EB to support effective and efficient DES model experimentation.
Advisor / supervisor
  • Bedford, Tim
  • Quigley, John
Resource Type
DOI
Date Created
  • 2025

Relations

Items

Thumbnail Title Date Uploaded Visibility Actions
file details PDF of Thesis T17599 2026-02-23 Public Download