基于多智能体深度强化学习的体系任务分配方法  

Task Assignment Method of Operation System of Systems Based on Multi-agent Deep Reinforcement Learning

作　　者：林萌龙 陈涛[1] 任棒棒 张萌萌[1] 陈洪辉[1] LIN Menglong;CHEN Tao;REN Bangbang;ZHANG Mengmeng;CHEN Honghui(Science and Technology on Information Systems Engineering Laboratory,National University of Defense Technology,ChangshaHunan 410073,China)

机构地区：[1]国防科技大学信息系统工程重点实验室,湖南长沙410073

出　　处：《指挥与控制学报》2023年第1期93-102,共10页Journal of Command and Control

摘　　要：为了应对在未来复杂的战场环境下,由于通信受限等原因导致的集中式决策模式难以实施的情况,提出了一个基于多智能体深度强化学习的分布式作战体系任务分配算法,该算法为各作战单元均设计一个独立的策略网络,并采用集中式训练、分布式执行的方法对智能体的策略网络进行训练,结果显示,经过学习训练后的各作战单元具备一定的自主协同能力,即使在没有中心指挥控制节点协调的情况下,依然能够独立地实现作战任务的高效分配.To cope with the situation that the centralized decision-making is difficult to be implemented due to impeded communication and other reasons in complex battlefield environment in the future,a distributed task assignment algorithm of operation system of systems(SoS)is proposed based on multi-agent deep reinforcement learning technology.This algorithm designs an independent policy network for each combat unit,and the centralized training and decentralized execution methods are utilized to train the policy network.The results show that each combat unit entities has certain self-coordination ability after being trained,and the combat units can still allocate the SoS tasks independently and effectively even in the absence of central command and control(C2)nodes.

关 键 词：多智能体系统 深度强化学习 体系架构 体系设计 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] E91[自动化与计算机技术—控制科学与工程]