网络结构化多Agent系统中基于动态综合关系网的任务分配

Research of Task Allocation Based on Dynamic Integrated Relationship Network for Networked Multiagent Systems

大规模网络结构化多Agent系统(Networked Multiagent Systems,NMAS)在分配任务时,需要考虑Agent的能力、Agent在底层物理网络中的位置和彼此之间的实际通信耗费、Agent在交互网络中的位置和彼此之间的交互距离,因此,任务分配实现难度相对较大.已有的任务分配模型没有有效考虑NMAS底层物理网络拓扑结构和社会组织结构的作用,对NMAS的任务分配协商过程缺乏深入研究,系统中横向通信密集,且Agent社会关系资源不能充分利用.针对此类问题,本文以最小化系统的任务执行时间和通信开销为目标,研究了分布式的基于动态综合关系网的任务分配模型(Dynamic Integrated Relationship Netw ork M odel,DIRNM).模型包括三个部分:基于物理能力和社会关系资源提供力的任务承包方选择策略、基于动态综合关系网的任务分配协商过程和基于直接社会关系资源的任务再分配机制.对比实验结果表明,所提出的DIRNM在降低任务执行总时间和通信开销、提高任务分配成功率方面具有显著优越性,对网络结构动态变化具有鲁棒性.

When the large-scale networked nultiagent systems（ NMAS） allocating tasks,the capacities of the agent,the position in the underlying physical network and the interactive network of the agent,and the actual communication cost and interaction distance between the agents should be considered,so the implementation of task allocation is relatively difficult. Existing task allocation models of NMAS do not consider the impact of the topology structure of underlying physical network and social organization structure validly,and lack in-depth research of task allocation negotiation process of NMAS. These may lead to the intensive horizontal communication and the insufficient use of agents＇ social relationship resources. Aiming at such problems,this paper takes minimum total task execution time and communication overhead as the dual optimization objectives,and studies the task allocation model based on the dynamic integrated relationship network（ DIRNM）. The model includes three parts： the selection strategy of task contractor based on physical ability and social relation resources,the negotiation process of task allocation based on dynamic integrated relationship network and the task reallocation strategy based on direct social relation resources. The comparative experimental results show that the model has significant effects on reducing the total task execution time and systematic communication cost,and improving the allocation success rate. In addition,it is robust to dynamic network structure.