基于深度强化学习的智能仿真平台设计  

The Design of Intelligence Simulation Platform Based on DRL

作　　者：吴昭欣 李辉[1,2] 王壮 陶伟[3] 吴昊霖 侯贤乐 Wu Zhaoxin;Li Hui;Wang Zhuang;Tao Wei;Wu Haolin;Hou Xianle(Nation Key Laboratory of Fundamental Science on Synthetic Vision,Sichuan University,Chengdu 610065,China;College of Computer Science,Sichuan University,Chengdu 610065,China;China Ship Development and Design Center,Wuhan 430000,China)

机构地区：[1]四川大学视觉合成图形图像技术国防重点学科实验室,成都610065 [2]四川大学计算机学院,成都610065 [3]中国舰船研究设计中心,武汉430000

出　　处：《战术导弹技术》2020年第4期193-200,共8页Tactical Missile Technology

基　　金：教育部联合基金(6141A02011607)。

摘　　要：针对当前作战仿真中指挥决策模型建模困难的问题,提出使用深度强化学习智能体进行指挥决策的方法,设计了一种基于深度强化学习技术的智能仿真平台。通过智能接口连接传统仿真平台与智能体,使智能体具备感知仿真环境态势并依据感知结果进行指挥决策的能力。利用该平台进行了舰载机进近引导实验,结果表明,智能仿真平台具备训练指挥决策智能体的能力,智能体可以做出合理的指挥决策,引导舰载机到达最终进近点。In view of the difficulty in modeling command&control model,an intelligent simulation platform based on deep reinforcement learning(DRL)is designed,and a method that is taking advantage of DRL agent for command&control is proposed.Through combining traditional simulation platform and agent,endowing the agent with the ability to sense the simulation environment and made command decisions on the basis of the perception.The carri-er-based aircraft guidance simulation results show that the platform is qualified enough to train command&control agent,which could make appropriate decisions for the aircraft to arrive at the final approach point.

关 键 词：作战仿真 深度强化学习 人工智能 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]