模数驱动的智能武器装备作战建模方法  

A smart agent-based combat modeling method using model and data

作　　者：朱智 杨松[1] 王涛[1] 王维平[1] 赵月华 ZHU Zhi;YANG Song;WANG Tao;WANG Weiping;ZHAO Yuehua(College of Systems Engineering,National University of Defense Technology,Changsha 410073,China;College of Information Management,Nanjing University,Nanjing 210033,China)

机构地区：[1]国防科技大学系统工程学院,长沙410073 [2]南京大学信息管理学院,南京210033

出　　处：《系统工程理论与实践》2024年第3期1083-1096,共14页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(62003359);装备发展部预研基金(80901020104)。

摘　　要：仿真建模作为一种模型驱动的机理建模,一直是武器装备系统采办、设计和评估等工作的主要手段.随着大数据和人工智能的深入应用,数据驱动的数据建模越来越受到关注.首先从模型驱动、数据驱动和模数驱动三个方面对武器装备作战建模方法进行概述,分析单独使用仿真建模或数据建模在应对当前武器装备作战建模工作的不足,提出一种模数驱动的数字化建模方法.该方法核心是模数驱动架构的设计,在此基础上运用功能决策树表示作战行为模型,对于行为模型中的决策环节,基于深度强化学习训练和嵌入数智Agent.通过构建弹道导弹突防过程中的多目标分配场景,对比仿真实验显示,采用智能网络决策比传统基于规则脚本的目标命中率有显著提高.Simulation modeling,a classical model driven physics-based method,has always gained high priority in acquisition,design,and evaluation of various combat systems.In recent years,inspired by big data and artificial intelligence,more and more simulation modelers have paid attention to the combined use of data-driven data modeling methods.In this context,this research investigated the combat system modeling literature from model-driven,data-driven,and hybrid driven perspective,respectively.Thereafter,we proposed a model and data hybrid driven intelligent modeling approach in consideration of limitations of using simulation modeling or data modeling alone.Firstly,we designed a model and data two-wheel driven architecture.Secondly,we applied a novel behavioral modeling method,namely,the function decision tree(FDT),to represent combat behaviors properly.Thirdly,for decision points in a behavioral model,we used the deep reinforcement learning to train smart agents.As a proof of concept,we built a multi-targets assignment scenario of ballistic missile penetration,and the results revealed that the smart agent-embedded ballistic missile significantly increased the ratio of target hits when compared with the traditional rule-based behavioral model.

关 键 词：模数驱动 行为建模 效能仿真 深度强化学习 

分 类 号：E917[军事]