航空学报 > 2022, Vol. 43 Issue (2): 325024-325024   doi: 10.7527/S1000-6893.2021.25024

基于GRU的敌方拦截弹制导律快速辨识方法

王因翰1, 范世鹏1, 吴广2, 王江1, 何绍溟1   

  1. 1. 北京理工大学 宇航学院, 北京 100081;
    2. 北京航天自动控制研究所, 北京 100854
  • 收稿日期:2020-11-27 修回日期:2020-12-21 发布日期:2021-02-02
  • 通讯作者: 范世鹏 E-mail:fspzxm@sina.com
  • 基金资助:
    北京市科学技术委员会项目(Z181100003218013)

Fast guidance law identification approach for incoming missile based on GRU network

WANG Yinhan1, FAN Shipeng1, WU Guang2, WANG Jiang1, HE Shaoming1   

  1. 1. School of Aerospace Engineering, Beijing Institute of Technology, Beijing 100081, China;
    2. Beijing Aerospace Automatic Control Institute, Beijing 100854, China
  • Received:2020-11-27 Revised:2020-12-21 Published:2021-02-02
  • Supported by:
    Beijing Municipal Science & Technology Commission Project (Z181100003218013)

摘要: 针对拦截弹制导律辨识问题,基于门循环单元(GRU)神经网络提出了一种制导律辨识方法。首先,构建了三维空间下的拦截弹-我方飞行器相对运动模型,并假设拦截弹采用经典的比例导引(PN)制导律或增强比例导引(APN)制导律。由相对运动模型提取数据,构成训练样本集和测试样本集,样本输入为敌我双方运动学信息,标签为敌方拦截弹对应的制导律。其次,建立了包含三层隐含层的GRU网络模型,采用基于Adam算法的反向传播对网络进行训练,探究了样本时间跨度等因素对辨识准确度的影响。最后,通过数学仿真对网络参数进行了优化,并在多种条件下进行了仿真验证。结果表明,相比其他网络结构而言,所提方法采用GRU拥有良好的抗噪能力和辨识精度;同时,相比于基于卡尔曼滤波器的制导律辨识模型,所提方法能够大幅减少辨识所需时间。

关键词: 制导律辨识, 比例导引, 增强比例导引, 门循环单元网络, 神经网络

Abstract: To identify the guidance law of enemy interceptor missile, a fast guidance law identification method is proposed based on the Gate Recurrent Unit (GRU). A model for the relative motion of the interceptor missile and aircraft in the three-dimensional space is constructed, and it is assumed that the interceptor adopts the classical PN guidance law or APN guidance law. The training sample set and test sample set are extracted from the model. The input of the samples is the kinematic information of both missile and aircraft, while the label is the guidance law of the missile. A neural network with three hidden layers is established based on the GRU. The back propagation based on Adam is used to train the network. The influence of the elements including noise, sample time span, type and size of neural networks on identification accuracy is investigated by simulation, and parameters of GRU networks are optimized. Simulations are conducted under different conditions. The results show that the method proposed has better performance of anti-noise and higher precision than other types of networks. In addition, compared with the identification model based on Kalman filter, the proposed method can reduce the identification time.

Key words: guidance law identification, proportional navigation, augmented proportional navigation, GRU network, neural network

中图分类号: