Date of Award
Doctor of Philosophy
Department of Computing
Dr. Jeanne Stynes
Prof. Dr. Udo Bleimann
The main advantage for the use of agent technology is in the capacity of agents to represent the interests of individual entities and to act autonomously on their behalf. In multi-agent systems this involves communication, coordination and negotiation.
The focus of this dissertation is to find ways to optimize multi-agent systems at different levels by taking into consideration their structure and the procedures used. As optimization depends to a large extent on the problem to be solved, group forming scenarios, in which self-interested agents compete to form and join groups, have been selected here for consideration. Optimization issues are discussed at architectural level, outlining an approach that emphasizes the strengths of agents in problem solving. Here the view of self-oriented agents is combined with the central view of the system, aiming for a good solution.
A structure using negotiation units that support multiple concurrent negotiations is presented. A desperation threshold optimizes the approach, influencing the number of agents used in multiple negotiations. Communication procedures within the organization structure of a leader/representative agent are optimized by introducing a message redirect mechanism and a timestamp mechanism is used to mitigate possible collisions. To improve performance the agent-based approach is modified by redistributing the agents that could not join a group during the agent-based approach, taking into consideration the quality factor (harmony value) of groups. In this context, an optimization point is introduced to enable the switch between the two approaches. The approach is validated using a prototypical implementation.
Stengel, Ingo, "Optimization in Multi-Agent Systems : Structures and Procedures" (2008). Theses [online].
Available at: https://sword.cit.ie/allthe/290