Selection of simulation variance reduction techniques through a fuzzy expert system

In this thesis, the design and development of a decision support system for the selection of a variance reduction technique for discrete event simulation studies is presented. In addition, the performance of variance reduction techniques as stand alone and combined application has been investigated. The aim of this research is to mimic the process of human decision making through an expert system and also handle the ambiguity associated with representing human expert knowledge through fuzzy logic. The result is a fuzzy expert system which was subjected to three different validation tests, the main objective being to establish the reasonableness of the systems output. Although these validation tests are among the most widely accepted tests for fuzzy expert systems, the overall results were not in agreement with expectations. In addition, results from the stand alone and combined application of variance reduction techniques, demonstrated that more instances of stand alone applications performed better at reducing variance than the combined application. The design and development of a fuzzy expert system as an advisory tool to aid simulation users, constitutes a significant contribution to the selection of variance reduction technique(s), for discrete event simulation studies. This is a novelty because it demonstrates the practicalities involved in the design and development process, which can be used on similar decision making problems by discrete event simulation researchers and practitioners using their own knowledge and experience. In addition, the application of a fuzzy expert system to this particular discrete event simulation problem, demonstrates the flexibility and usability of an alternative to the existing algorithmic approach. Under current experimental conditions, a new specific class of systems, in particular the Crossdocking Distribution System has been identified, for which the application of variance reduction techniques, i.e. Antithetic Variates and Control Variates are beneficial for variance reduction