融合改进LS与分量相关性的极移预报模型

A Polar Motion Prediction Model Based on Improved LS and Component Correlation Fusion

ES评分 0

DOI 10.14075/j.jgg.2025.11.376
刊名
Journal of Geodesy and Geodynamics
年,卷(期) 2026, 46(7)
作者
作者单位

1 中国矿业大学环境与测绘学院, 徐州, 221116;
2 广西精密导航技术与应用重点实验室, 桂林, 541004;
3 徐州市国有资产投资经营集团有限公司, 徐州, 221000

摘要
为探索纯数据驱动的极移(PM)高精度自主预报, 提出一种粒子群耦合遗忘因子递归最小二乘和多元自回归(PSO-FFRLS+MAR)模型。该模型对历史数据进行动态加权, 以跟踪系统缓慢变化, 并对PMX与PMY残差协同波动进行联合建模。实验表明, 该模型在1~90 d预报期内, 精度和稳定性均优于主流模型, 对PMY的提升尤为显著(相较于经典LS+AR模型, 其MAE降低35.4%), 且逐天检验证明了该提升的统计显著性。PMY精度提升更多, 从物理机制上可能源于模型更有效地捕捉到了主导该方向的强季节性大气激发。本文研究为地球定向参数的中短期自主预报提供了一种高性能数据驱动方法, 可通过误差分析关联地球角动量激发函数来理解模型物理机制。
Abstract
To explore high-accuracy autonomous polar motion (PM) prediction using purely data-driven methods, we propose a hybrid model named PSO-FFRLS+MAR (forgetting factor recursive least squares optimized by particle swarm optimization and multivariate autoregressive). This model dynamically weights historical data to track slow system variations and jointly models the co-oscillations of PMX and PMY residuals. Experiments show that during the 1 to 90 days prediction horizon, the proposed model outperforms mainstream models in both accuracy and stability, with particularly significant improvements for PMY (compared with the classical LS+AR model, a 35.4% reduction in MAE). Day-by-day validation confirms the statistical significance of these improvements. The greater enhancement in PMY accuracy may be physically attributed to the model's enhanced ability to capture the strong seasonal atmospheric excitation that dominates this component. This study provides a high-performance data-driven method for medium- to short-term autonomous prediction of Earth orientation parameters and, through error analysis, links the model's physical mechanisms with geophysical angular momentum excitation functions.
关键词
极移; 遗忘因子递归最小二乘法; 多元自回归; PMX-PMY相关性; 中短期预报
KeyWord
polar motion(PM); forgetting factor recursive least squares(FFRLS); multivariate autoregressive(MAR); PMX-PMY correlation; medium- to short-term prediction
基金项目
页码 834-841
  • 参考文献
  • 相关文献
  • 引用本文

要亚鑫, 郝飞宇, 程彤, 王潜心. 融合改进LS与分量相关性的极移预报模型 [J]. 大地测量与地球动力学. 2026; 46; (7). 834 - 841.

  • 文献评论

相关学者

相关机构