基于多智能体协同的无人机簇群多域节能抗干扰通信

Multi-agent collaboration based UAV clusters multi-domain energy-saving anti-jamming communication

随着无人机网络向着集群化方向发展,无人机簇群通信面临恶意干扰电磁环境下频谱和能量资源不足以及环境部分可观测等问题.针对联合频谱域和功率域的无人机簇群抗干扰问题,以最小化长期传输能量损耗和跳频开销为优化目标,通过建立分布式部分可观测马尔可夫决策过程模型,构建基于多智能体协同的无人机簇群节能抗干扰通信框架.具体地,各簇头无人机作为智能体,利用长短时记忆神经网络的信息长期记忆优势,结合双深度Q学习方法,采用多智能体框架完成分布式训练,最终实现仅需各簇群本地观测信息即可完成协同多域节能抗干扰通信分布式决策.仿真结果表明,本文所提算法可适应部分可观测且未知动态变化的无人机簇群传输环境和干扰环境,相较于基准算法能更有效地降低长期传输能量损耗和跳频开销,且同时提升数据传输成功率.

With the development of unmanned aerial vehicle(UAV)clustering,the communication within UAV clusters faces a scarcity of spectrum and energy resources as well as environmental partial observability in malicious jamming electromagnetic environments.To minimize long-term system transmission energy consumption and frequency-hopping overhead,we study the anti-jamming problem in UAV clusters from joint spectrum and power domains.Then,a multi-agent collaborative-based energy-saving anti-jamming communication framework for UAV clusters is constructed after establishing a distributed partially observable Markov decision process.Particularly,each UAV cluster head as an agent uses the information long-term memory advantage of a long short-term memory neural network,combines the double deep Q network method,and adopts a multi-agent framework to complete the distributed training.Finally,a distributed decision-making solution for collaborative multi-domain energy-saving anti-jamming communication is achieved,relying on only the local observed information of each UAV cluster.Simulation results show that the proposed algorithm can adapt to the UAV clusters transmission environment and the jamming environment with partially observable and unknown dynamic changing patterns.Compared to the baselines,it reduces long-term transmission energy consumption and frequency-hopping overhead and improves the data transmission success rate more effectively.