系统工程与电子技术 ›› 2021, Vol. 43 ›› Issue (12): 3643-3651.doi: 10.12305/j.issn.1001-506X.2021.12.27

• 系统工程 • 上一篇    下一篇

基于EAS+MADRL的多无人车体系效能评估方法研究

高昂1, 郭齐胜1,*, 董志明1, 杨绍卿2   

  1. 1. 陆军装甲兵学院演训中心, 北京 100072
    2. 国防科技创新研究院, 北京 100071
  • 收稿日期:2020-11-23 出版日期:2021-11-24 发布日期:2021-11-30
  • 通讯作者: 郭齐胜
  • 作者简介:高昂(1988—), 男, 博士研究生, 主要研究方向为装备作战与保障仿真、多智能体深度强化学习|郭齐胜(1962—), 男, 教授, 博士研究生导师, 博士, 主要研究方向为装备论证与试验|董志明(1977—)男, 教授, 硕士研究生导师, 博士, 主要方向为作战试验与仿真|杨绍卿(1991—), 男, 助理研究员, 博士, 主要研究方向为无人智能作战战法理论与仿真研究
  • 基金资助:
    军队科研计划项目(41405030302);军队科研计划项目(41401020301)

Research on efficiency evaluation method of multi unmanned ground vehicle system based on EAS+MADRL

Ang GAO1, Qisheng GUO1,*, Zhiming DONG1, Shaoqing YANG2   

  1. 1. Military Exercise and Training Center, Army Academy of Armored Forces, Beijing 100072, China
    2. Academy of Military Sciences PLA China, Beijing 100071, China
  • Received:2020-11-23 Online:2021-11-24 Published:2021-11-30
  • Contact: Qisheng GUO

摘要:

无人作战开始步入现代战争舞台, 多无人车(multi unmanned ground vehicle, MUGV) 协同作战将成为未来陆上作战的主要样式。体系效能评估是装备论证和战法研究的核心问题, 针对MUGV体系效能评估问题, 建立了一套以自主学习算法为基础的探索性仿真分析方法。将MUGV对抗过程建模为零和随机博弈(zero sum stochastic game, ZSG)模型, 通过使用多智能体深度强化学习类方法(multi agent deep reinforcement learning, MADRL)探索在不同对方无人车规模条件下, ZSG模型的纳什均衡解, 分析纳什均衡条件下参战双方胜率, 作战时长等约束, 完成MUGV体系作战效能评估, 并在最后给出了MUGV体系效能评估应用示例, 从而建立了更可信、可用的体系效能评估方法。

关键词: 多无人车, 体系效能评估, 多智能体深度强化学习, 探索性分析仿真

Abstract:

Unmanned combat has stepped into the stage of modern war, and multi unmanned ground vehicle (MUGV) cooperative combat will become the main style of land combat in the future. The evaluation of system effectiveness is the core problem of equipment demonstration and combat method research. A set of exploratory simulation analysis method based on multi agent deep reinforcement learning is established to evaluate the combat effectiveness of MUGV system. The MUGV confrontation process is modeled as a zero sum stochastic game (ZSG) model. By using multi-agent deep reinforcement learning MADRL to explore the Nash equilibrium solution of ZSG model under the condition of different scale of enemy unmanned ground vehicle, and analyzing the constraints such as the winning rate and the duration of combat under the condition of Nash equilibrium, the combat effectiveness evaluation of MUGV system is completed. Finally, the application example of operational effectiveness evaluation of MUGV system is given. Thus, a more reliable and available system performance evaluation method is established.

Key words: multi unmanned ground vehicle, system effectiveness evaluation, multi agent deep reinforcement learning (MADRL), exploratory analysis simulation

中图分类号: