Use this URL to cite or link to this record in EThOS:
Title: An artificial intelligence framework for experimental design and analysis in discrete event simulation
Author: Taylor, Richard Paul
ISNI:       0000 0001 3513 789X
Awarding Body: University of Warwick
Current Institution: University of Warwick
Date of Award: 1988
Availability of Full Text:
Access from EThOS:
Access from Institution:
Simulation studies cycle through the phases of formulation, programming, verification and validation, experimental design and analysis, and implementation. The work presented has been concerned with developing methods to enhance the practice and support for the experimental design and analysis phase of a study. The investigation focussed on the introduction of Artificial Intelligence (AI) techniques to this phase, where previously there existed little support. The reason for this approach was the realisation that the experimentation process in a simulation study can be broken down into a reasoning component and a control of execution component. In most studies, a user would perform both of these. The involvement of a reasoning process attracted the notion of artificial intelligence or at least the prospective use of its techniques. After a study into the current state of the art, work began by considering the development of a support system for experimental design and analysis that had human intelligence and machine control of execution. This provided a semi-structured decision-making environment in the form of a controller that requested human input. The controller was made intelligent when it was linked to a non-procedural (PROLOG) program that provided remote intelligent input from either the user or default heuristics. The intelligent controller was found to enhance simulation experimentation because it ensures that all the steps in the experimental design and analysis phase take place and receive appropriate input. The next stage was to adopt the view that simulation experimental design and analysis may be enhanced through a system that had machine intelligence but expected human control of execution. This provided the framework of an advisor that adopted a consultation expert system paradigm. Users were advised on how to perform simulation experimentation. Default reasoning strategies were implemented to provide the system with advisory capabilities in the tasks of prediction, evaluation, comparison, sensitivity analysis, transient behaviour, functional relations, optimisation. Later the controller and the advisor were linked to provide an integrated system with both machine intelligence and machine control of execution. User involvement in the experimentation process was reduced considerably as support -¿as provided in both the reasoning and control of execution aspects. Additionally, this integrated system supports facilities for refinement purposes that aim at turning the system’s knowledge into expertise. It became theoretically possible for other simulation experts to teach the system or experiment with their own rules and knowledge. The following stage considered making the knowledge of the system available to the user, thereby turning the system into a teacher and providing pedagogical support Teaching was introduced through explanation and demonstration. The explanation facility used a mixed approach: it combined a first time response explanation facility to "how" and "why" questions with a menu driven information system facility for "explain" requests or further queries. The demonstration facility offers tutorials on the use of the system and how to carry out an investigation of any of the tasks that the system can address. The final part of the research was to collect some empirical results about the performance of the system. Some experiments were performed retroactively on existing studies. The system was also linked to a data-driven simulation package 'hat permitted evaluation using some large scale industrial applications. The system’s performance was measured by its ability to perform as well as students with simulation knowledge but not necessarily expertise. The system was also found to assist the user with little or no simulation knowledge to perform as well as students with knowledge. This study represents the first practical attempts to use the expert system framework to model the processes involved in simulation experimentation. The framework described in this thesis has been implemented as a prototype advisory system called WES (Warwick Expert Simulator). The thesis concludes that the framework proposed is robust for this purpose.
Supervisor: Not available Sponsor: University of Warwick
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: T Technology (General)