弹道学报2023，Vol.35Issue(4)：53-60,8.DOI:10.12115/j.issn.1004-499X(2023)04-007
基于改进DQN方法的滑翔制导炮弹弹道规划
Ballistic Planning of Glide-guided Projectile Based on Improved DQN Method
摘要
Abstract
In recent years,the rapid development of deep reinforcement learning has provided a new option for trajectory planning.It seeks optimal trajectory planning via deep neural networks interacting with the environment.However,there is strong randomness in the interaction with the environment,which may lead to poor training results and low learning efficiency during the reinforcement learning training process.To address this issue,a deep reinforcement learning trajectory planning method was proposed based on the Improved Deep Q-Network(IM-DQN)in this paper.A combination strategy of excellent experience preservation and exploration restriction rate was added to the traditional Deep Q-Network(DQN)algorithm to guide the training direction and prevent large-scale network collapse.A three-degree-of-freedom model of a glide guided projectile was constructed,and the trajectory planning problem was transformed into a Markov decision process,determining the state space,state transition mode,reward function,and related algorithm parameters for training.Under both unconstrained and constrained conditions,the range results obtained by this method were compared laterally with the maximum lift-to-drag ratio method and the Gaussian Pseudospectral Method(GPM)based on optimal control theory.Simulation results show that the training process of the IM-DQN method is more stable than that of the DQN method.In terms of seeking optimal range,the maximum range value obtained by the IM-DQN method surpasses the aforementioned two reference methods.The research results can provide references for future reinforcement learning trajectory planning.关键词
滑翔制导炮弹/弹道规划/深度强化学习/优秀经验保存/探索限制策略Key words
glide guided projectile/ballistic planning/deep reinforcement learning/excellent experience preservation/explore limiting strategies分类
军事科技引用本文复制引用
谢蕃葳,王旭刚..基于改进DQN方法的滑翔制导炮弹弹道规划[J].弹道学报,2023,35(4):53-60,8.基金项目
中央高校基本科研业务费专项资金资助(30919011401) （30919011401）
