航空学报 > 2017, Vol. 38 Issue (12): 321222-321222   doi: 10.7527/S1000-6893.2017.321222

一种多智能体协同信息一致性算法

陈旿, 张鑫, 金鑫, 李泽宏, 洪亮   

  1. 西北工业大学 自动化学院, 西安 710072
  • 收稿日期:2017-03-07 修回日期:2017-05-12 出版日期:2017-12-15 发布日期:2017-05-12
  • 通讯作者: 陈旿 E-mail:chenwu@nwpu.edu.cn
  • 基金资助:
    陕西省重点研发计划(2017GY-069)

A cooperative information consensus algorithm for multi-agent system

CHEN Wu, ZHANG Xin, JIN Xin, LI Zehong, HONG Liang   

  1. School of Automation, Northwestern Polytechnical University, Xi'an 710072, China
  • Received:2017-03-07 Revised:2017-05-12 Online:2017-12-15 Published:2017-05-12
  • Supported by:
    Major Research Project of Shaanxi Province (2017GY-069)

摘要: 以无人机(UAVs)/导弹集群为代表的多智能体协同作战在未来战场中占有重要地位。协同信息的共享和一致性是多智能体系统完成协同、编队、集结、同步等协同任务的关键基础和前提。首先,基于邻居系统和团势能建立了信息一致性模型,将多智能体的协同信息的偏差映射为团势能。然后,通过以并行能量最小化求解马尔可夫随机场最大后验概率的方法实现了分布式无中心条件下的协同信息一致性。与传统的一致性算法相比,所提出的算法引入了虚拟基准的概念。当无外部基准输入时,基于平均场方法,通过邻居之间的协同信息交互建立虚拟基准;当存在领航节点或虚拟领航节点时,将领航节点协同信息的状态及状态导数作为虚拟基准。仿真结果表明:所得出的算法具有对网络规模不敏感、快速收敛、高鲁棒性的优点;对有/无基准输入的情况可采用相同的算法,体现了算法具有较好的适应性。

关键词: 多智能体, 一致性, 马尔可夫随机场, 平均场, 虚拟基准

Abstract: Multi-agent cooperative operation plays an important role in the cyberspace war, and the main application lies in the field of multiple Unmanned Aerial Vehicles (UAVs)/multi-missile collaborative cluster. Sharing collaborative information and consistency are the foundation and prerequisite for the multi-agent to complete collaborative tasks such as coordination, formation, flocking and synchronization. A consensus information model is established based on the neighbor system and the cluster potential, and the bias of the cooperative information is mapped to the cluster potential energy. By using the minimization of parallel energy to solve the maximum a posteriori probability of the Markov random field, cooperative information reaches a consensus with distributed and non center condition. Different from the traditional consensus algorithm, the algorithm proposed introduces the concept of virtual reference. A virtual reference is established by cooperative information interaction among the neighbors by using the mean field theory with no external reference input. When the pilot node or the virtual pilot node exists, the state and its derivative of the pilot node cooperative information are used as the virtual reference. Simulation results show that the proposed algorithm has the advantages of insensitivity to network scale, fast convergence and high robustness. The algorithm can be also used in the presence/absence of reference input, meaning that the algorithm has great adaptability.

Key words: multi-agent, consensus, Markov random field, mean field, virtual reference

中图分类号: