Prediction and Performance of BDS Satellite Clock Bias Based on CNN-LSTM-Attention Model

Area of Science:

• Geodesy and Satellite Navigation
• Artificial Intelligence in Geospatial Science

Background:

• Satellite Clock Bias (SCB) significantly impacts Precise Point Positioning (PPP) accuracy.
• International GNSS Service (IGS) real-time products face disruptions, affecting positioning reliability.

Purpose of the Study:

• To develop an advanced model for accurate and stable SCB prediction.
• To mitigate the impact of network interruptions on real-time PPP.

Main Methods:

• Integration of Convolutional Neural Networks (CNNs) and Attention mechanisms with Long Short-Term Memory (LSTM) networks.
• Development of the CNN-LSTM-Attention model for feature extraction and weighted prediction.
• Validation through various forecasting horizons and dynamic PPP experiments.

Main Results:

• Significant prediction accuracy improvements across multiple forecasting horizons compared to benchmark models (LP, QP, ARIMA, BP, LSTM).
• Achieved positioning accuracy comparable to post-processed products in dynamic PPP experiments.
• Demonstrated superior accuracy and stability in SCB prediction.

Conclusions:

• The CNN-LSTM-Attention model effectively addresses SCB prediction challenges in real-time PPP.
• The model enhances positioning accuracy and stability, meeting application demands.
• This approach offers a robust solution for reliable satellite navigation.