测绘学报 ›› 2026, Vol. 55 ›› Issue (10): 1732-1743.doi: 10.11947/j.AGCS.2026.20260135

• 大地测量学与导航 • 上一篇    下一篇

基于径向基插值的LS-BiGRU-Attention模型的极移短期预报

王乐洋1,2(), 严凯玲1,2, 吴飞1,2(), 阙海波3   

  1. 1.东华理工大学测绘与空间信息工程学院,江西 南昌 330013
    2.东华理工大学自然资源部环鄱阳湖区域矿山环境监测与治理重点实验室,江西 南昌 330013
    3.同济大学测绘与地理信息学院 上海 200092
  • 收稿日期:2026-04-09 修回日期:2026-10-13 出版日期:2026-11-16 发布日期:2026-11-16
  • 通讯作者: 吴飞 E-mail:wleyang@163.com;201960114@ecut.edu.cn
  • 作者简介:王乐洋(1983—),男,博士,教授,研究方向为大地测量反演和测量平差。E-mail:wleyang@163.com
  • 基金资助:
    国家自然科学基金(42574064)

Short-term prediction of polar motion using an LS-BiGRU-Attention model based on radial basis function interpolation

Leyang Wang1,2(), Kailing Yan1,2, Fei Wu1,2(), Haibo Que3   

  1. 1.School of Surveying and Geoinformation Engineering, East China University of Technology, Nanchang 330013, China
    2.Key Laboratory of Mine Environmental Monitoring and Improving around Poyang Lake of Ministry of Natural Resources, East China University of Technology, Nanchang 330013, China
    3.College of Surveying and Geo-informatics, Tongji University, Shanghai 200092, China
  • Received:2026-04-09 Revised:2026-10-13 Online:2026-11-16 Published:2026-11-16
  • Contact: Fei Wu E-mail:wleyang@163.com;201960114@ecut.edu.cn
  • About author:Wang Leyang (1983—), male, PhD, professor, majors in geodetic inversion and surveying adjustment. E-mail: wleyang@163.com
  • Supported by:
    The National Natural Science Foundation of China(42574064)

摘要:

随着深空探测、精密定轨等领域的发展,对极移数据的精度、实时性提出了更高要求,高精度的极移数据存在时间延迟,极移变化本质上呈现出复杂的非线性特征。本文引入非线性径向基函数插值法对日分辨率极移数据进行平滑重构与时间重采样,以获得适用于模型输入的6 h重采样时序。通过计算极移X分量与Y分量之间的互相关系数验证了极移X分量与Y分量之间存在的强自相关性。为充分利用相关性,本文构建了非线性的LS-BiGRU-Attention多变量组合预报模型,将重构后的X、Y分量双序列同步联合输入网络中进行动态特征提取。结果显示,在短期预报中该组合模型能有效捕捉非线性动态特征,至预报第30天时,相较于传统LS+AR模型,本文方法在PMX和PMY方向的预报精度分别提升了5.9%和12.5%。

关键词: 极移, 预报, 径向基插值, 时序重采样

Abstract:

Fields such as deep space exploration and precise orbit determination impose higher requirements on the accuracy and real-time performance of polar motion data. However, high-precision polar motion data suffer from time delays, and polar motion variations inherently exhibit complex nonlinear characteristics. Therefore, in this paper, a nonlinear radial basis function (RBF) interpolation method is introduced to smoothly reconstruct and temporally resample daily-resolution polar motion data, aiming to obtain a 6-hour resampled time series suitable for model input. By calculating the cross-correlation coefficient between the X and Y components of polar motion, the strong correlation between them is verified. To fully utilize this correlation, this paper constructs a nonlinear multivariate combined prediction model, LS-BiGRU-Attention, in which the reconstructed dual sequences of X and Y are simultaneously and jointly inputted into the network for dynamic feature extraction. The results indicate that the combined model can effectively capture non-linear dynamic features in short-term prediction. By the 30th day of prediction, compared with the traditional LS+AR model, its prediction accuracy in the PMX and PMY directions is improved by 5.9% and 12.5%, respectively.

Key words: polar motion, prediction, radial basis function interpolation, temporal resampling

中图分类号: