摘要
针对异构无人集群作战过程中的任务分配问题,提出一种基于分布式演员-评论家(actor-critic)强化学习网络的任务分配方法。设计由transformer与Deep LSTM网络构成的编码模块对输入态势进行预处理编码;采用LSTM作为动作模块,并使用网络级联方式实现复杂动作解耦,对异构作战单元实现动态任务指派;使用线性感知机组成评价模块输出状态价值。仿真实验表明,提出的方法能够有效地实现异构无人集群的作战任务分配,具备良好的实用性与适用性。
Aiming at the task allocation problem in the process of heterogeneous unmanned cluster combat,a task allocation method based on distributed actor-critic reinforcement learning network is proposed.Firstly,an encoding module consisting of transformer and Deep LSTM networks is designed to preprocess and encode the input situation.Then,LSTM is used as an action module and network cascade is used to decouple the complex actions and to achieve dynamic task assignment for heterogeneous combat units.Finally,a linear perceptron is used to form an evaluation module to output the state value.The simulation deduction experiments show that the proposed method can effectively achieve the combat task allocation of heterogeneous unmanned clusters,with super practicality and applicability.
作者
杨绍卿
左源
邓宝松
YANG Shaoqing;ZUO Yuan;DENG Baosong(Defense Innovation Institute,Chinese Academy of Military Sciences,Intelligent Game and Decision Laboratory,Chinese Academy of Military Sciences,Beijing 100071,China)
出处
《火力与指挥控制》
北大核心
2025年第5期77-83,90,共8页
Fire Control & Command Control
基金
军队装备综合研究基金资助项目。
关键词
任务分配
无人集群
强化学习
推演仿真
task allocation
unmanned cluster
reinforcement learning
deduction simulation
