An agent based co-operative preference model

Jayousi, Rashid

Abstract

Distributed problem solving is often characterised by multiple valid solutions, a solution being considered to be valid if it meets all the constraints. When there are multiple valid solutions, the user has a choice, which can be specified in terms of preferences on different desirable aspects (i.e. resources) of the solution. In that event, the best solution is the one that meets as many preferences as possible.
In this thesis, a multi-agent approach has been used for distributed problem-solving, each autonomous agent (task agent), under the supervision of a relevant task coordinator, solves its part of the subtask, in cooperation with other task agents. To resolve contention in the preferences for the same resources, a market-based payment scheme is applied for the preferences to be bought and sold by the contending task agents through their coordinators. The best solution is achieved for a task, when further iteration does not increase its total preference value, that is a convergence is achieved.
This thesis presents a preference model that includes a preference specification strategy, a preference processing technique, and a theoretical performance model, the latter describes the quantitative behaviour of the preference model. The thesis also presents a simulation study to show that the preference model works satisfactory and according to the theoretical performance model.
For the simulation study we used the problem of distributed scheduling in the manufacturing domain. The results of the study show that our agent based strategy not only reaches convergence on the final preference value for the whole system, that value is also independent of initial order of subtask allocation. The results verify the validity of our approach handsomely.

Citation

Jayousi, R. An agent based co-operative preference model. (Thesis). Keele University. Retrieved from https://keele-repository.worktribe.com/output/1020265

• Thesis Type: Thesis
• Deposit Date: Dec 20, 2024
• Publicly Available Date: Dec 20, 2024
• Keywords: Department of Computer Science
• Public URL: https://keele-repository.worktribe.com/output/1020265
• Award Date: 2003