航空学报 > 2023, Vol. 44 Issue (19): 328400-328400   doi: 10.7527/S1000-6893.2022.28400

基于迁移学习的角度约束时间最短制导算法

罗皓文1,2, 何绍溟1,2(), 金天宇1,2, 刘子超1,2   

  1. 1.北京理工大学 宇航学院,北京  100081
    2.北京理工大学 无人机自主控制技术北京市重点实验室,北京  100081
  • 收稿日期:2022-12-14 修回日期:2023-02-01 接受日期:2023-03-20 出版日期:2023-10-15 发布日期:2023-03-31
  • 通讯作者: 何绍溟 E-mail:shaoming.he@bit.edu.cn
  • 基金资助:
    国家重点研发计划(2022YFE0204400)

Impact-angle-constrained with time-minimum guidance algorithm based on transfer learning

Haowen LUO1,2, Shaoming HE1,2(), Tianyu JIN1,2, Zichao LIU1,2   

  1. 1.School of Astronautics,Beijing Institute of Technology,Beijing  100081,China
    2.Beijing Key Laboratory of UAV Autonomous Control Technology,Beijing Institute of Technology,Beijing  100081,China
  • Received:2022-12-14 Revised:2023-02-01 Accepted:2023-03-20 Online:2023-10-15 Published:2023-03-31
  • Contact: Shaoming HE E-mail:shaoming.he@bit.edu.cn
  • Supported by:
    National Key Research and Development Programm(2022YFE0204400)

摘要:

由于导弹所处的气动环境等外界条件复杂多变,因此在特定气动环境中表现优异的深度监督学习制导算法将无法直接作用于新环境,这给制导指令的精准预测带来了很大挑战。针对上述问题,本文以落角约束下优化飞行时间的制导问题为背景,提出了一种基于迁移学习的角度约束时间最短制导算法,该算法只需极少量的新数据,便能使制导律迅速适应不同的气动环境。在本文中,我们首先通过域对抗地训练特征提取器和域判别器,从不同气动环境的弹道数据中提取出对气动变化不敏感的关键特征;其次设计了适应于不同气动环境的偏置加速度预测器,进而实现了对导弹的精准制导;最后,大量数值仿真结果表明,在新气动环境下,基于迁移学习的角度约束时间最短制导算法仍能实现制导指令的精准预测。

关键词: 计算制导, 深度学习, 迁移学习, 落角约束, 偏置比例导引律

Abstract:

The aerodynamic environment and other external conditions of the missile are complex and changeable. The deep supervised learning guidance algorithm with excellent performance in the specific aerodynamic environment cannot be directly applied to the new environment, which brings great challenges to the accurate prediction of guidance instructions. To solve the above problem, this paper proposes a Transfer-learning-based Impact-angel Constraint with Time-minimum Guidance algorithm (TICTG) for missile guidance, which minimizes the impact time under impact angle constraints. The proposed algorithm can quickly adapt the guidance law to different aerodynamic conditions with very little new data. Firstly, we extract key features insensitive to aerodynamic changes from ballistic data in different aerodynamic environments by training the feature extractor and domain discriminator against the domain ground. Secondly, we design a bias acceleration predictor adapted to different aerodynamic conditions, so as to achieve accurate guidance of the missile. A large number of numerical simulation results show that the method proposed can achieve accurate prediction of guidance instructions in the new aerodynamic environment.

Key words: computational guidance, deep learning, transfer learning, impact angle constraint, bais proportion navigation

中图分类号: