Scholarlink

<p>The prediction products of the polar motion parameters play a crucial role in precision orbit determination, high-precision positioning, timing services and other fields. In this paper, we proposed employing the Nonlinear Auto Regressive (NAR) neural network model for polar motion prediction. The network parameters are determined through the cross-validation method, leading to the establishment of the NAR neural network prediction model. The EOP 20 C04 product provided by the International Earth Rotation and Reference Systems Service (IERS) was utilized as experimental data, and 365 sets of ultra-fast forecasts spanning 1—7&#xa0;days were conducted for the polar motion parameters. Furthermore, we also discussed the optimal amount of data for NAR, it was found that the training accuracy is optimal when using a 10-year dataset. Compared with the LS + AR model, the proposed method achieves an average improvement of 5.82% and a maximum improvement of 8.24% in the X direction, and an average improvement of 9.06% and a maximum improvement of 13.65% in the Y direction. Compared with the IERS Bulletin A and other methods, the prediction accuracy in the X and Y directions has been improved to a certain extent under different forecast spans.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Ultra-Short Term Prediction of Polar Motion Based on NAR Neural Network

  • 期刊论文

  • 发表日期：

    2026年3月1日

查看全文
  • Fei Wu,
  • Lintao Tao,
  • Yufeng Dai,
  • Leyang Wang,
  • Qingchen Cao

摘要

The prediction products of the polar motion parameters play a crucial role in precision orbit determination, high-precision positioning, timing services and other fields. In this paper, we proposed employing the Nonlinear Auto Regressive (NAR) neural network model for polar motion prediction. The network parameters are determined through the cross-validation method, leading to the establishment of the NAR neural network prediction model. The EOP 20 C04 product provided by the International Earth Rotation and Reference Systems Service (IERS) was utilized as experimental data, and 365 sets of ultra-fast forecasts spanning 1—7 days were conducted for the polar motion parameters. Furthermore, we also discussed the optimal amount of data for NAR, it was found that the training accuracy is optimal when using a 10-year dataset. Compared with the LS + AR model, the proposed method achieves an average improvement of 5.82% and a maximum improvement of 8.24% in the X direction, and an average improvement of 9.06% and a maximum improvement of 13.65% in the Y direction. Compared with the IERS Bulletin A and other methods, the prediction accuracy in the X and Y directions has been improved to a certain extent under different forecast spans.