基于EC-MAVEN算法的多智能体多约束作战任务分配方法  

Multi-Agent and Multi-Constraint Combat Task Allocation Method Based on EC-MAVEN Algorithm

作　　者：钟孙健 张德平[1] ZHONG Sunjian;ZHANG Deping(School of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics,Nanjing 210000,China)

机构地区：[1]南京航空航天大学计算机科学与技术学院,南京210000

出　　处：《火力与指挥控制》2024年第9期58-63,共6页Fire Control & Command Control

基　　金：国防基础科研基金(JCKY2020605C003);国防基础科研基金资助项目(JCKY2022605C006)。

摘　　要：为解决现有启发式算法在面对多约束作战任务分配时,存在收敛速度慢,探索效率低等问题。基于作战仿真环境提出改进的多智能体变分探索方法(EC-MAVEN),对具有多约束的作战任务分配问题进行建模,提高复杂任务中智能体的探索效率,并使用情景控制提高良好样本的利用率,加快训练速度,有利于跳出局部最优。通过作战打击任务分配环境进行仿真实例验证,结果表明,EC-MAVEN算法可减少陷入局部次优,缩短训练时间,提高任务成功率。To solve the problems of slow convergence and low exploration efficiency of existing heu⁃ristic algorithms in the face of multi-constraint combat task allocation.Based on the combat simulation environment,an improved multi-intelligent agent variational exploration method(EC-MAVEN)is pro⁃posed to model the combat task assignment problems with multiple constraints,and to improve the explora⁃tion efficiency of intelligent agents in complex tasks,and the control of scenes is used to improve the utili⁃zation rate of good samples and to speed up the training speed,and to facilitate the jumping out of the local optimum.The simulation example is verified by the combat strike task allocation environment,and the results show that the EC-MAVEN algorithm can reduce falling into local sub-optimality,and can shorten the training time,and to improve the mission success rate.

关 键 词：多智能体 强化学习 任务分配 协作任务 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]