Scholarlink

<p>The accuracy of Satellite Clock Bias (SCB) is a crucial factor affecting the performance of precise point positioning. However, the post-processed precise clock error products provided by the International GNSS Service (IGS) often fail to meet real-time requirements, and the broadcast ephemeris transmitted by satellites does not achieve the necessary precision. To address this issue, this paper proposes an ensemble model that combines Empirical Wavelet Transform (EWT) and Long Short-Term Memory (LSTM) neural networks—referred to as the EWT-LSTM ensemble model. This integration enables the model to effectively capture key features of SCB sequences, ultimately improving prediction accuracy and stability. We compared the EWT-LSTM ensemble model with the Backpropagation Neural Network model (BP), the LSTM model, and the combined model of the Improved Sparrow Search Algorithm and Gated Recurrent Unit Neural Network (ITSSA-GRU). The analysis included all available BeiDou-3 satellites, considering various orbital and atomic clock types. The experimental results show that the EWT-LSTM ensemble model significantly improves SCB prediction accuracy. For 12-h prediction tasks, the accuracy increased by 76.67%, 69.52%, and 61.51% compared to the BP, LSTM, and ITSSA-GRU models, respectively. For 24-h prediction tasks, the accuracy increased by 93.88%, 81.45%, and 74.25%, respectively.</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

Prediction of BDS-3 satellite clock bias based on EWT-LSTM ensemble model

  • 期刊论文

  • 发表日期：

    2025年6月2日

查看全文
  • Kaihui Lv,
  • Chenglin Cai,
  • Yihao Cai,
  • Wenhui Guan,
  • Zexian Li,
  • Lingfeng Cheng,
  • Yun Fang

摘要

The accuracy of Satellite Clock Bias (SCB) is a crucial factor affecting the performance of precise point positioning. However, the post-processed precise clock error products provided by the International GNSS Service (IGS) often fail to meet real-time requirements, and the broadcast ephemeris transmitted by satellites does not achieve the necessary precision. To address this issue, this paper proposes an ensemble model that combines Empirical Wavelet Transform (EWT) and Long Short-Term Memory (LSTM) neural networks—referred to as the EWT-LSTM ensemble model. This integration enables the model to effectively capture key features of SCB sequences, ultimately improving prediction accuracy and stability. We compared the EWT-LSTM ensemble model with the Backpropagation Neural Network model (BP), the LSTM model, and the combined model of the Improved Sparrow Search Algorithm and Gated Recurrent Unit Neural Network (ITSSA-GRU). The analysis included all available BeiDou-3 satellites, considering various orbital and atomic clock types. The experimental results show that the EWT-LSTM ensemble model significantly improves SCB prediction accuracy. For 12-h prediction tasks, the accuracy increased by 76.67%, 69.52%, and 61.51% compared to the BP, LSTM, and ITSSA-GRU models, respectively. For 24-h prediction tasks, the accuracy increased by 93.88%, 81.45%, and 74.25%, respectively.