文章摘要
引用本文:阎栋,苏航,朱军.基于DQN的反舰导弹火力分配方法研究[J].导航定位与授时,2019,6(5):18-24 本文二维码信息
二维码(扫一下试试看!)
基于DQN的反舰导弹火力分配方法研究
Research on Fire Distribution Method of Anti-ship Missile Based on DQN
  
DOI:
中文关键词:  反舰导弹  目标分配  深度Q值网络
English Keywords:Anti-ship missile  Fire distribution  DQN
基金项目:国防基础科研计划 (JCKY2017204B064)
作者单位
阎栋 清华大学计算机科学与技术系北京 100086 
苏航 清华大学计算机科学与技术系北京 100086 
朱军 清华大学计算机科学与技术系北京 100086 
摘要点击次数: 26
全文下载次数: 28
中文摘要:
      反舰导弹作为海上作战的主战武器,由于其精度高、射程远、威力大等特性长期以来一直被当作舰艇编队的主要防御对象。针对反舰导弹打击舰艇编队的火力分配问题,我们提出了一种基于深度Q值网络求解反舰导弹火力分配策略的算法。不同于现有的基于领域知识的方法,深度Q值网络无需依赖任何先验信息,就能够通过与模拟器的交互自动求解最佳的攻击策略。该算法使用深度神经网络拟合Q值函数,解决了传统强化学习中的状态空间过大无法遍历的问题。实验结果表明,在各种不同的舰队防御配置下,深度Q值网络求解得到的攻击策略均获得了最佳的毁伤效果。
English Summary:
      As the main weapon in the sea battle, the anti-ship missile has long been regarded as the main defense object of the warship formation due to its high precision, long range and large power. Aiming at the problem of firepower distribution of anti-ship missiles against ship formations, an algorithm based on deep Q-value network for solving the firepower allocation strategy of anti-ship missiles is proposed. Unlike existing domain-based methods, deep Q-value networks can automatically solve the best attack strategy by interacting with the simulator without relying on any a priori information. The algorithm uses the deep neural network to fit the Q-value function to solve the problem that the state space in traditional reinforcement learning cannot be traversed too much. The experimental results show that under various flock defense configurations, the attack strategy obtained by the deep Q-value network has obtained the best damage effect.
查看全文  查看/发表评论  下载PDF阅读器