| 引用本文: | 方科,张庆振,倪昆,崔朗福.飞行时间约束下的再入制导律[J].哈尔滨工业大学学报,2019,51(10):90.DOI:10.11918/j.issn.0367-6234.201808104 |
| FANG Ke,ZHANG Qingzhen,NI Kun,CUI Langfu.Reentry guidance law with flight time constraint[J].Journal of Harbin Institute of Technology,2019,51(10):90.DOI:10.11918/j.issn.0367-6234.201808104 |
|
| 摘要: |
| 为应对现代战场的信息化与集群化发展趋势,从多高超声速飞行器饱和打击任务需求出发,针对其中的再入飞行时间约束条件进行研究,提出一套基于Deep Q-learning Network(DQN)的时间可控再入制导律. 该制导律工作流程为首先纵向轨迹预测-校正模块根据当前飞行状态和攻角-速度剖面规划出倾侧角幅值；然后在线约束强化管理模块对其进行安全限幅处理；最后倾侧角符号规划模块以调节再入飞行时间为目标,在对横向飞行状态进行马尔科夫决策过程建模的基础上,设计相应的深度神经网络进行离线训练以在线生成倾侧角符号,进而与幅值信息共同组成最终的倾侧角指令.多组仿真的对比分析结果表明:在标称环境下的多任务仿真中,时间可控再入制导律能够自主地进行倾侧角符号的在线规划,在不影响制导精度的前提下,对再入飞行时间进行调整以满足不同的任务需求；在参数拉偏的蒙特卡洛仿真中,其在保证再入飞行安全、稳定的同时,仍然能将时间误差控制在合理的范围之内.从而验证了相对于传统方法而言,本研究所设计的再入制导律在任务适应性、鲁棒性与时间可控性等方面均具有良好表现,能够有效地满足飞行时间约束下的再入任务需求. |
| 关键词: 高超声速飞行器 再入制导 预测-校正制导 深度强化学习 DQN |
| DOI：10.11918/j.issn.0367-6234.201808104 |
| 分类号:V448.235 |
| 文献标识码:A |
| 基金项目: |
|
| Reentry guidance law with flight time constraint |
|
FANG Ke,ZHANG Qingzhen,NI Kun,CUI Langfu
|
|
(School of Automation Science and Electrical Engineering, Beihang University, Beijing 100083, China)
|
| Abstract: |
| In order to cope with the development trend of informationization and clustering in modern battlefields, this paper studies the reentry flight time constraint in the mission of multi-hypersonic vehicles’ saturation attack, and proposes a time-controllable reentry guidance law based on the Deep Q-learning Network(DQN). Its workflow is mainly divided into three parts. First, according to the current flight state and angle of attack vs.velocity profile, the amplitude of bank angle is planned based on the prediction-correction module of longitudinal trajectory. Then the online constraint enhanced management module performs safety limiting processing the amplitude of bank angle. Finally, based on Markov Decision Process modeling of lateral flight state, the symbol planning module aims to adjust reentry flight time and to designs the corresponding deep neural network for offline training to generate the symbol of bank angle online, and then amplitude information is combined to form the final bank angle command. The comparative analysis of mutiple simulations shows that in multi-mission simulation under nominal environment, the time-controllable reentry guidance law can independently plan bank angle’s symbol online and adjust reentry flight time to meet different mission requirements without affecting guidance precision. In the Monte Carlo simulations with biased parameters, the time error can still be controlled within a reasonable range while ensuring safe and stable reentry flight. Therefore, compared with the traditional method, the reentry guidance law designed in this paper has good performance in terms of task adaptability, robustness, and time controllability, and it can effectively meet reentry mission requirements with the flight-time constraint. |
| Key words: hypersonic vehicle reentry guidance predictor-corrector guidance deep reinforcement learning deep Q-learning network |
