Protocoles de cooperation dans les systemes multiagents: Une approche basee sur les relations de dependances (French text)

The combined advance of telecommunication technologies and organizations structures implies the emergence of new information structures. In fact, virtual enterprises and virtual marketplaces are significant examples. These new structures are modeled as using intelligent software agents who interact, cooperate and compete in an open multiagent world.; This thesis presents a co-operation model for multiagent systems based on the utility theory, the dependency relations and the case based reasoning. The model allows an agent to reason on its peers, in particular to determine its relations and situations of dependence and to evaluate the profit relating to a coalition. We consider the integration of a co-operation mechanism, as an essential component in the design of autonomous agents, which are evolving in an open multiagent system. The concept of opening is fundamental for the applications considered in this thesis: virtual enterprises and virtual market-places. By open we mean that agents may enter or leave the agency at any moment.; As in such systems, the organization of the agents can't be conceived during the design phase, the cooperation is based on the dynamic formation of coalition. As we don't suppose that agents are benevolent, our model integrates a negotiation strategy using utility functions and risk evaluation.; To validate the model we have implemented prototypes on the MIAMI and LALO platforms. Mainly, the virtual enterprise building and the virtual market-place MIAMAP have been used for experimental evaluations.; The experimental results show that the model makes a balance between the need of the agents' individual rationality and the need for social coherence in the overall system. In fact, the most obvious consequence for the use of this model, is its capacity to improve the agents' social commitment.; This thesis brings contribution to three inter-linked domains in distributed artificial intelligence: agents' interaction, automated negotiation and conflict resolution, and coalition formation.; From the interaction modeling point of view, the model which we proposed allows to take into account the concept of influence and goals adoption, things absent in the approaches based exclusively on utility.; In regards to the negotiation and resolution of conflicts, we have used a combination of a negotiation strategy based on risk evaluation and the agents' relations of dependency. This combination has led to two significant consequences: on one hand the widening of the set of possible agreements between the agents and on the other hand the emergence of a balance between the individual performance of the agents and the global performance of the system.; With regard to the coalition formation, our model introduces two concepts: the persuasion and the social commitment. The persuasion concept allows an agent to convince by argumentation its partners to adopt its goals and the concept of social commitment allows making a balance between individual rationality and social coherence in the system.