上海交通大学学报

上海交通大学学报（自然版）

• 航空、航天 • 上一篇    下一篇

基于概率群集的多战机协同空战决策算法

刘波，陈哨东，贺建良
  

  1. (中航工业洛阳电光设备研究所 光电控制技术重点实验室， 河南 洛阳 471000)
  • 收稿日期:2010-08-27 修回日期:1900-01-01 出版日期:2011-02-28 发布日期:2011-02-28

Air Combat DecisionMaking for Multifighter Coordinated Attack Based on Probability Collectives

LIU Bo,CHEN Shaodong,HE Jianliang
  

  1. (Science and Technology on Electrooptic Control Laboratory, Luoyang Institute of ElectroOptical Equipment of AVIC, Luoyang 471000, Henan， China)
  • Received:2010-08-27 Revised:1900-01-01 Online:2011-02-28 Published:2011-02-28

摘要： 为解决多战机在无中心控制条件下的自主协同空战决策问题，提出了一种基于概率群集的分布式协同算法.以导弹为Agent构建协同分布式决策模型，并基于概率群集将离散的组合优化问题映射为概率分布空间上的一个凸优化问题；通过定义Agent的贡献度扩展概率群集架构.上述2种方法分别解决了模型中存在较大局部最优值区域和由于异构Agent间的非线性协同导致系统无法准确收敛的问题.实验结果表明，所提算法与传统算法相比具有更好的鲁棒性、可扩展性，收敛速度也有一定提高.

关键词: 决策, 概率群集, 协同工作, 异构Agent, 强化学习

Abstract: A new algorithm based on probability collectives was proposed to solve the decisionmaking of multifighter to coordinated attack multitarget without centric control. Firstly, the coordinated distributed decision making model is proposed based on missile which is defined as an agent. To solve an optimization problem in which there is subspace full of local optimal value near the global optimum in the solution space, the new algorithm turns it into a convex problem in a space of probability distributions based on probability collectives. Secondly, the probability collectives framework is extended by agent contribution rate to realize accurate convergence. The simulation results show that as compared with conventional methods, the proposed method is able to converge to the global optimum with more stability, scalability and faster convergence rate.

中图分类号: