The Research On Self-Organizing Mechanisms In Multi-Agent Systems Considering Hybrid Interaction Characteristics

The cooperation and coordination in multi-agent system is an important research subject. During the interaction process, many hybrid characteristics appear. However, previous researches seldom study how to promote the coorperation of multi-agent systems with hybrid characteristics systematically. Considering different hybrid interaction characteristics in different multi-agent systems, different self-organizing mechanisms are proposed to promote the cooperation and coordination among agents in this paper.Firstly, considering the hybrid interaction structures, the emergence model of social convention is extended to a dual-network. Each Agent plays coordination game with other agents in different network layers and no default behavior is specified. Then the influences of self-organizing mechanisms are explored. It presents that the second layer network can obviously accelerate the convergence of social convention. In addition, random neighbor rewiring and random global rewiring are effective to improve the convergence speed.Then, considering the hybrid interaction strategies, a malicious action is added to the agents in the tribute/tax model and it will hinder the cooperation of the system. To control the malicious agents and promote the cooperation, some self-organizing mechanisms are proposed. It is found that individual reputation based rewiring performs best. In addition, if there are very few fixed agents which will not turn to malicious agents, malicious agents can be nearly completely eliminated with the help of individual reputation based rewiring.Finally, considering hybrid interaction roles, a novel self-organizing mechanism aiming at improving the performance of system in a task allocation environment is introduced. All agents can switch between the two roles:executive agent (EA) and intermediary agent (IA). Some underloaded agents act as IA spontaneously. By undertaking the relation adaptation work for overloaded EAs, IA helps overloaded EAs to save the load of relation adaptation. Hence, EAs can fully utilize the load to execute tasks. Compared with the Central, K-Adapt and TBM mechanisms, Intermediary-Adapt performs better than K-Adapt and TBM, and can achieve nearly 80% of the Central mechanism’s performance.