中国空间科学技术

中国空间科学技术 ›› 2022, Vol. 42 ›› Issue (6): 134-139.doi: 10.16708/j.cnki.1000-758X.2022.0092

• 论文 • 上一篇    下一篇

基于BP神经网络的介质表面充电电位反演方法

张诚悦，全荣辉，张海呈   

  1. 南京航空航天大学航天学院, 南京210016
  • 出版日期:2022-12-25 发布日期:2022-11-09

Dielectric surface charging potential inversion method based on BP neural network

ZHANG Chengyue，QUAN Ronghui，ZHANG Haicheng   

  1. School of Astronautics，Nanjing University of Aeronautics and Astronautics，Nanjing 210016，China
  • Published:2022-12-25 Online:2022-11-09

PDF (PC)

128

摘要： 为了实现全面、实时的在轨卫星充放电风险分析，基于在相同环境下，不同材料表面充电的关联性，利用BP神经网络建立了一种由Kapton材料表面充电电位反演卫星其他常用介质材料表面电位的模型。以Kapton材料的表面电位以及材料厚度为输入，其他介质材料的表面电位作为模型输出，使用COMSOL建立的表面充电模型对神经网络进行训练，将反演误差降低到10%以下，并利用Kapton与Teflon材料表面充电地面试验数据验证反演模型的准确性，结果显示Teflon表面电位的反演值与实测值间的相对误差小于16%。

关键词: 表面充电, BP神经网络, 数值仿真, 反演, 卫星

Abstract: To realize the comprehensive and real-time risk analysis of in-orbit satellites＇ charge-discharge, a BP neural network for the surface potential inversion of dielectrics commonly used on satellites with the Kapton surface potential was built based on the relation between surface charging of different materials in the same environment.The Kapton surface potential and the materials thickness were taken as the inputs, while the surface potential of other dielectric materials were taken as the model outputs.By using the surface charging model established by COMSOL to train the neural network, the inversion error was reduced to less than 10%.The accuracy of the inversion model was verified by the surface-charge-experiment data of Kapton and Teflon.The results show that the relative error between the inversion value and the experimental value is less than 16%.

Key words:  , surface charging, BP neural network, numerical simulation, inversion, satellite