Coordination mechanisms for self-interested multi-agent systems

A multi-agent system (MAS) is a collection of agents that interact with each other. Multi-agent systems (MASs) can be classi¯ed as self-interested MASs and cooperative MASs according to the features of agent goals. Coordination is one of the major issues of MAS research. It plays a central role in MASs to ensure agents achieve interactions properly. Today, the remarkable growth of MAS applications brings higher requirements and more challenges to agent coordinations. Many complex applications require a MAS to include various agents to work together under an open and dynamic environment. Toward some challenges in current agent coordination research, this thesis deeply investigates agent coordination problems in self-interested MASs, and proposes three coordination mechanisms based on three different methodologies. Firstly, this thesis investigates the use of Coloured Petri Net techniques in agent coordinations. As one of the best modelling tools, Coloured Petri Nets can express a great range of concurrent systems and interactions in graphical representations and well-de¯ned semantics, and allow formal analysis and transformations. In the thesis, a coordination mechanism that uses Coloured Petri Nets to model inter- action protocols of agents is proposed and implemented. This mechanism allows interaction protocols to be separated from hard-coded agents. The separation of agents and interaction protocols makes it possible to allow agents to evaluate and select protocols before they operate interactions. Furthermore, a Coloured Petri Net based approach that allows agents to propose, exchange and evaluate inter-action protocols is also introduced in this thesis. By citing Coloured Petri-Net techniques in agent coordinations, agents have more °exible and rational interactions. Secondly, this thesis proposes a °exible team formation mechanism for self-interested agents. In this mechanism, agents can evaluate the performance and importance of other agents in the system, and select team members with reasonable terms and costs according to the evaluation result automatically. Comparing with some traditional team formation mechanisms, the flexible team formation mechanism makes agent team compositions more reasonable, and avoids some potential benefit conflicts among self-interested team members (more suitable for self-interested agents). Finally, a knowledge level coordination mechanism that uses of ontologies to describe and manage MAS knowledge is introduced in this thesis. By using ontologies, MAS knowledge can be described and organised in hierarchies, and the dynamic features of independent knowledge sources are captured.