Flexible Kooperation zwischen autonomen Agenten in dynamischen Umgebungen

This work proposes a scheme for the formation of holonic agents in a multiagentsystem acting in a dynamic environment. This model captures cooperative holonic multiagent-systems in which each agent has incomplete, vague and erroneous information about the other agents and about its dynamic and uncertain world. Holons are temporarily formed to ful ll complex tasks, that a single agent cannot handle alone. To nd assistance the head of a holon continuously tries to improve its holonic structure. Therefore it builds a set of hypothetical holons, rating them and if a higher ranked holon structure than the current one is found, it starts with the (re-)negotiation of the holon. Also if the holon is for some reason not longer capable to achieve its goals, re-negotiation starts. The scheme describes in particular, how an agent learns the properties of other agents. Based on this knowledge the agents are then able to build hypothetical holons by simulations. In the following these hypothetical holons are realised by bilateral negotiations with the agents of that pre-computed holonstructure. For describing this scheme and the algorithms based on this scheme, I rst present a formalism to map a multiagent-system onto a vector space, such that it is possible to give a formal description of the entities of a multiagenten-system and of the functions of the proposed scheme. Finally I present four algorithms to form dynamically holons based on that scheme. These algorithms consist of a simple randomised agent selection to a complex knowledge-based selection of relevant agents to solve the given tasks. These algorithms are in the end discussed and evaluated in detail.