基于注意力机制的无人机集群动态分配  

UAV Swarm Dynamic Allocation Based on Attention Mechanism

作　　者：张家宝 陈勇[1] 薛文军 秦嘉[1] 李元恒 ZHANG Jiabao;CHEN Yong;XUE Wenjun;QIN Jia;LI Yuanheng(North Automatic Control Technology Institute,Taiyuan 030006,China)

机构地区：[1]北方自动控制技术研究所,太原030006

出　　处：《火力与指挥控制》2025年第2期86-92,99,共8页Fire Control & Command Control

摘　　要：在俄乌战争和巴以冲突中,人工智能技术支持的无人机群被频繁用于感知和打击,这种人工智能生成的超快速杀伤链在当前复杂城市作战中展现出巨大的战争潜力。面向多临机目标的无人机集群动态分配技术存在“系统响应慢、资源调度力弱、协作效率低”等问题。构建基于注意力机制的无人机集群动态分配决策模型,提升求解效率,缩短指挥决策阶段耗费时间,且决策模型具有较强的泛化性;运用强化学习(带基线的REINFORCE算法)训练,简化训练流程。In the Russian-Ukrainian war and the Israeli-Palestinian conflict,AI-supported UAV swarm are frequently used for sensing and striking,and this AI-generated ultra-fast kill chain shows great war potential in the current complex urban warfare.At present,the UAV cluster dynamic allocation technology has some problems,such as slow system response,weak resource scheduling and low cooperation efficiency.The dynamic allocation decision model of UAV cluster based on attention mechanism is constructed to improve the solving efficiency,shorten the time consuming in the command and decision stage,and the decision model has strong generalization.Use reinforcement learning(REINFORCE algorithm with baseline)training to simplify the training process.

关 键 词：无人机集群 动态分配 注意力机制 带基线的REINFORCE算法 

分 类 号：V279[航空宇航科学与技术—飞行器设计]