Adaptive Parallel Fusion of Convolutional Neural Network and Long Short-Term Memory Networks for Short-Term BeiDou Satellite Clock Bias Prediction

Accurate short-term prediction of satellite clock bias (SCB) is important for maintaining the continuity of real-time precise positioning and timing services provided by the BeiDou Navigation Satellite System (BDS). To address the limited adaptability of individual models to different prediction tasks and the potential propagation of upstream feature-extraction errors in cascaded architectures, an adaptive parallel fusion strategy combining a convolutional neural network (CNN) and a long short-term memory (LSTM) network was proposed. Independent CNN and LSTM branches were used to extract the local variations and temporal dependence features of the SCB series, respectively. Then, based on the fitting results of the two branches over the modeling sequence, the optimal fusion weight was determined through coarse and fine searches and subsequently applied at the output stage to combine their recursive predictions. To evaluate the adaptability and prediction performance of the proposed model, three sets of experiments were designed by varying the input time scale, prediction time scale, and onboard atomic clock type. Comparative analyses were conducted against the quadratic polynomial, grey model, CNN, LSTM, and cascaded CNN-LSTM models. The results show that the proposed model produces the lowest root mean square (RMS), mean absolute error (MAE), and Range across the three input time scales. It also performs best at different prediction time scales for all metrics except the RMS of the 6 h prediction, which is the second lowest. For the 12 h prediction, the RMS, MAE, and Range are 0.537, 0.461, and 1.078 ns, respectively. For BDS-3 medium Earth orbit satellites equipped with rubidium and passive hydrogen maser clocks, the RMS values are 0.602 and 0.330 ns, while the corresponding Range values are 1.173 and 0.722 ns, respectively. Overall, the proposed model shows good adaptability to different prediction tasks and achieves high overall prediction accuracy.

Authors

Institutions

Publication Details

Journal
Sensors
Published
2026-11-17
DOI
https://doi.org/10.3390/s26227236
Primary Topic
GNSS positioning and interference
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Adaptive Parallel Fusion of Convolutional Neural Network and Long Short-Term Memory Networks for Short-Term BeiDou Satellite Clock Bias Prediction

Zhongchen Guo, Jiajia Yuan, Chao Hu, Tingting Fang et al.
4 links
Sensors
GNSS positioning and interference
article

Adaptive Parallel Fusion of Convolutional Neural Network and Long Short-Term Memory Networks for Short-Term BeiDou Satellite Clock Bias Prediction

Zhongchen Guo, Jiajia Yuan, Chao Hu, Tingting Fang, Liying Yao, Lei Xu, Xiwen Pan, Qiang Zhao
article en

Abstract

Accurate short-term prediction of satellite clock bias (SCB) is important for maintaining the continuity of real-time precise positioning and timing services provided by the BeiDou Navigation Satellite System (BDS). To address the limited adaptability of individual models to different prediction tasks and the potential propagation of upstream feature-extraction errors in cascaded architectures, an adaptive parallel fusion strategy combining a convolutional neural network (CNN) and a long short-term memory (LSTM) network was proposed. Independent CNN and LSTM branches were used to extract the local variations and temporal dependence features of the SCB series, respectively. Then, based on the fitting results of the two branches over the modeling sequence, the optimal fusion weight was determined through coarse and fine searches and subsequently applied at the output stage to combine their recursive predictions. To evaluate the adaptability and prediction performance of the proposed model, three sets of experiments were designed by varying the input time scale, prediction time scale, and onboard atomic clock type. Comparative analyses were conducted against the quadratic polynomial, grey model, CNN, LSTM, and cascaded CNN-LSTM models. The results show that the proposed model produces the lowest root mean square (RMS), mean absolute error (MAE), and Range across the three input time scales. It also performs best at different prediction time scales for all metrics except the RMS of the 6 h prediction, which is the second lowest. For the 12 h prediction, the RMS, MAE, and Range are 0.537, 0.461, and 1.078 ns, respectively. For BDS-3 medium Earth orbit satellites equipped with rubidium and passive hydrogen maser clocks, the RMS values are 0.602 and 0.330 ns, while the corresponding Range values are 1.173 and 0.722 ns, respectively. Overall, the proposed model shows good adaptability to different prediction tasks and achieves high overall prediction accuracy.

SensorsVol. 26(22)
Anhui University of Science and Technology (CN), Wuhan University (CN), Soochow University (CN), Suzhou University (CN), Jiangxi University of Science and Technology (CN)
National Natural Science Foundation of China, Shandong University of Science and Technology, Anhui University of Science and Technology, Wuhan University, Anhui Provincial Department of Education, Scientific Research Foundation of Education Department of Anhui Province of China, Suzhou University
GNSS positioning and interference
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon