• 研究探讨 •    下一篇

基于深度学习的航天器组合体惯性参数在轨智能辨识

金晨迪, 康国华*, 郭玉洁, 乔思元   

  1. 南京航空航天大学航天学院微小卫星研究中心
  • 收稿日期:2018-07-20 修回日期:2018-08-29 出版日期:2019-04-25 发布日期:2018-12-28
  • 作者简介:金晨迪(1994-),男,硕士研究生,jinchdi@163.com,研究方向为航天器姿态控制
  • 基金资助:

    空间智能控制技术重点实验室开放基金资助项目(No.ZDSYS-2017-01,No.KGJZDSYS-2018-07)

 Onorbit intelligent identification of combined spacecraft′s inertia parameter based on deep learning

JIN  Chendi, KANG  Guohua*, GUO  Yujie, QIAO  Siyuan   

  1. MSRC,College of Astronautics,Nanjing University of Aeronautics and Astronautics, Nanjing 210016,China
  • Received:2018-07-20 Revised:2018-08-29 Online:2019-04-25 Published:2018-12-28

摘要: 针对在轨服务过程形成新组合体的动力学参数未知的问题,借助深度学习在多参数寻优上的优势,提出了一种基于卷积神经网络的智能参数辨识算法,实现在外力作用下,线动量和角动量不守恒条件下的航天器组合体多参数辨识。利用卷积神经网络权值共享的特点,设计4层卷积神经网络,通过短时间内对大量特定存储形式的状态数据的训练,实现航天器组合体多参数快速高精度辨识。利用数学仿真试验对算法的可行性进行验证,结果表明:在24s内,质量与质心位置收敛;1190s内,惯量参数收敛,辨识精度在3%以内。说明所提方法在外界随机干扰力和力矩影响下能准确快速辨识出航天器组合体质量、质心位置和惯量矩阵。

关键词: 深度学习, 组合航天器, 惯性参数, 在轨辨识, 卷积神经网络

Abstract: Aiming at the problem of unknown dynamic parameters of the new assembly during the onorbit service, a parameter identification algorithm based on convolution neural network was proposed with the help of deep learning in multiparameter optimization. The algorithm realizes the identification of the combined spacecraft′s multiparameter under the condition of external force and non conservation of linear momentum and angular momentum. A 4layer convolution neural networks was designed by using the characteristic of the weight sharing of the convolution neural network. The identification of inertial parameters with high precision was achieved by plenty of training of state data in a specific form of storage in a short time. The feasibility of the convolution neural network algorithm was proved by simulation calculation. The results show that the proposed method can accurately and quickly identify the mass, centroid position and inertia matrix of combined spacecraft under the influence of external random force and moment, the identification accuracy is within 3%.

Key words: deep learning, combined spacecraft, inertia parameter, onorbit identification, Convolutional Neural Network(CNN)