基于LSTM-RF模型的卫星钟差预报

Satellite Clock Error Prediction Based on LSTM-RF Model

ES评分 0

DOI 10.14075/j.jgg.2025.10.346
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 安徽理工大学空间信息与测绘工程学院, 淮南, 232001

摘要
针对卫星钟差预报中单一模型非线性建模能力有限与预报精度不高的问题, 提出长短期记忆网络(LSTM)与随机森林(RF)的组合预报模型。该模型充分发挥LSTM对钟差时序特征的深度挖掘能力, 同时利用随机森林的集成学习机制对LSTM预测残差进行非线性补偿, 实现预报精度跃升。实验表明, LSTM-RF组合模型相比于单一LSTM模型, 在短、中及长期预报中GPS、BDS、Galileo及GLONASS卫星预报精度分别提升74.4%~86.8%、27.5%~32.0%、80.5%~82.7%及0~18.2%, 由此验证了LSTM-RF组合模型对于非线性钟差序列进行预报的可行性和普适性。
Abstract
To address the issues of limited nonlinear modeling capability and low prediction accuracy in single-model satellite clock bias prediction, a combined prediction model integrating long short-term memory (LSTM) neural network and random forest (RF) is proposed. This model fully leverages LSTM's ability to deeply mine the temporal characteristics of clock biases, while utilizing the ensemble learning mechanism of Random Forest to non-linearly compensate for the prediction residuals of LSTM, thereby achieving a significant improvement in prediction accuracy. Experiments demonstrate that, compared to the single LSTM model, the LSTM-RF combined model enhances the prediction accuracy for GPS, BDS, Galileo, and GLONASS satellites by 74.4% to 86.8%, 27.5% to 32.0%, 80.5% to 82.7%, and 0 to 18.2%, respectively, in short-term, medium-term, and long-term predictions. This verifies the feasibility and universality of the LSTM-RF combined model for predicting nonlinear clock bias sequences.
关键词
钟差预报; LSTM-RF; 误差补偿; 多卫星系统
KeyWord
clock error forecast; LSTM-RF; error compensation; multi-satellite system
基金项目
页码 808-815
  • 参考文献
  • 相关文献
  • 引用本文

来闯, 王中元, 胡超. 基于LSTM-RF模型的卫星钟差预报 [J]. 大地测量与地球动力学. 2026; 46; (7). 808 - 815.

  • 文献评论

相关学者

相关机构