Social Network Evolution of Multiagent Groups in Task Allocation

The social network and group structure can improve the effectiveness of tasks and the management of multiagent systems (MASs). However, in complex task environments, the fixed network and social relation are not necessarily the optimal design. Therefore, agents should be able to adjust social relations in a distributed and self-organized manner to adapt to different task demands. To solve this problem, a hybrid social evolution mechanism based on Q-learning is proposed in this article. The evolution of social relations is realized from the perspective of individual, group, and global situations to improve the success rate and comprehensive effectiveness of task allocation in the system while reducing the communication cost during task execution. In addition, the hybrid evolution mechanism considers the balance of interests between the global and the individual, which can guarantee the rationality of task benefit distribution while improving the global effectiveness. The experiments show that compared with the single-form evolution method, the hybrid mechanism proposed in this article has more advantages in the success rate, communication cost, and comprehensive effectiveness of task allocation. Moreover, the rationality of benefit distribution is also improved, and a strong and harmonious social system is realized.