摘要
现有无人机集群的协同决策设计所依据的信息共享缺乏对无人机之间通信能力的合理假设。针对电量、载荷和路线约束下的无人机集群多目标救援问题,结合无人机飞行路线,考虑通信能力对无人机之间信息共享的限制。首先,将问题建模成部分可观测马尔可夫决策过程;然后,利用循环神经网络提出基于深度强化学习的能够适应通信拓扑结构不断变化的分布式救援策略。仿真结果表明,所提策略相较于其他策略在通信受限的情况下具有更佳的分布式救援性能,无人机数量和无人机通信能力需要依据救援场景进行联合设置方能达到无人机集群救援性能和使用成本的最佳折中。
The current designs of the cooperative decision-making of an unmanned aerial vehicle(UAV)swarm usually adopt unreasonable assumptions on the communication ability between UAVs.Focusing on a multi-target rescue problem of a UAV swarm under constraints of energy,load and path,the limitation on the information sharing due to the communication constraints and the flight path of UAVs were taken into account.Firstly,the problem was formulated as a partially observable Markov decision process(POMDP).Then,a recurrent neural network was used to propose a deep-reinforcement-learning-based distributed rescue strategy,which is able to adapt to the changeable communication topology.Simulation results show that the proposed strategy outperforms other strategies under communication constraints,and further show that a careful joint setting of the size and communication ability of a UAV swarm is needed to achieve the best compromise between the UAV swarm rescue performance and the cost.
作者
俞汉清
林艳
贾林琼
李强
张一晋
YU Hanqing;LIN Yan;JIA Linqiong;LI Qiang;Zhang Yijin(School of Electronic and Optical Engineering,Nanjing University of Science and Technology,Nanjing 210094,China;National Mobile Communications Research Laboratory,Southeast University,Nanjing 210096,China;Peng Cheng Laboratory,Shenzhen 518000,China)
出处
《物联网学报》
2022年第3期103-112,共10页
Chinese Journal on Internet of Things
基金
国家自然科学基金资助项目(No.62071236,No.62001225)
中央高校基本科研业务费资助项目(No.30920021127)
江苏省自然科学基金资助项目(No.BK20190454)
鹏城实验室重大攻关项目(No.PCL2021A15)
东南大学移动通信国家重点实验室开放研究基金资助项目(No.2022D07)。
关键词
无人机
多目标救援
马尔可夫决策过程
分布式策略
强化学习
unmanned aerial vehicle
multi-target rescue
Markov decision process
distributed strategy
reinforcement learning
