摘要
真实-虚拟-构造为近距空战对抗训练提供了有力支撑。针对课题对蓝方虚拟实体的实际决策建模需求,在对比分析深度强化学习与经典智能优化方法的基础上,从优化理论的角度对神经网络的权值空间和结构空间进行定义,提出基于智能优化的进化神经网络决策模型及其求解方法。首先,分析近距空战战术特点,战机飞行运动模型,实际决策建模需求。其次,分别设计战机关键飞行状态、动作空间、适应度函数,实现蓝方端到端感知与决策。最后,给出基于经典遗传神经网络的决策模型及求解示例。结果表明,所提方法可实现蓝方战机通过对抗数据来学习对手作战特点的功能,验证了模型及方法的有效性;同时所提方法对目前智能优化及其改进算法,以及不同结构神经网络具有通用性。
The live-virtual-constructive(LVC)provides a strong support for the close-range air combat confrontation training.Aiming at the actual decision modeling requirements of blue-side virtual entity,based on the comparative analysis of deep reinforcement learning and classical intelligent optimization methods,the weight space and structure space of neural network are defined from the perspective of optimization theory,and the evolutionary neural network decision model based on intelligent optimization and its solution method are proposed.Firstly,the tactical characteristics of close-range air combat,fighter flight motion model and actual decision modeling requirements are analyzed.Secondly,the key flight state,action space and fitness function are designed to realize the end-to-end perception and decision of the blue-side.Finally,the decision model based on classical genetic neural network and its solution example are given.The results show that the proposed method can realize the function of learning the combat characteristics of the opponent through the confrontation data,and verify the effectiveness of the model and method;at the same time,it has universality for the current intelligent optimization and its improved algorithm,as well as different structure neural networks.
作者
高昂
董志明
李亮
段莉
郭齐胜
GAO Ang;DONG Zhiming;LI Liang;DUAN Li;GUO Qisheng(Military Exercise and Training Center,Army Academy of Armored Forces,Beijing 100072,China;Unit 61516 of the PLA,Beijing 100076,China)
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2021年第6期1606-1617,共12页
Systems Engineering and Electronics
基金
军队科研计划项目(41405030302,41401020301)资助课题。
关键词
真实-虚拟-构造
近距空战
决策模型
神经网络
智能优化
智能蓝方
live-virtual-constructive(LVC)
close-range air combat
decision model
neural network
intelligent optimization
blue-side of intelligence
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;通讯作者: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;李亮(1982-),男,讲师,博士,主要研究方向为装备需求论证与试验鉴定评估;段莉(1976-),女,高级工程师,硕士,主要研究方向为信息系统;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。
