@COMMENT This file was generated by bib2html_grace.pl <https://sourceforge.net/projects/bib2html/> version 0.94
@COMMENT written by Patrick Riley <https://sourceforge.net/users/patstg/>
@COMMENT This file was prepared using the NASA Astrophysics Data System (ADS)
@COMMENT https://ui.adsabs.harvard.edu/
@ARTICLE{2026Meas..27220973Y,
       author = {{Yang}, Zhixin and {Zheng}, Yousi and {Tang}, Feifei and {Liu}, Hui and {Wang}, Bin and {Li}, Nanjie and {Zhao}, Yongmao},
        title = "{Low earth orbit satellite short-term orbit prediction using the LSTM-Transformer neural network model}",
      journal = {Measurement},
     keywords = {Low Earth orbit (LEO) satellites, Short-term orbit prediction, Longshort-termmemory (LSTM), Transformer},
         year = 2026,
        month = may,
       volume = {272},
          eid = {120973},
        pages = {120973},
     abstract = "{The precise orbit of Low Earth Orbit (LEO) satellites is crucial for
        LEO-enhanced Global Navigation Satellite System (GNSS) precise
        positioning,and short-term orbit prediction is extremely
        necessary to compensate for the time delay caused by orbit
        determination. The traditional dynamical propagation method is
        susceptible to error accumulation, and single neural network
        models have limitations in effectively capturing temporal
        dependencies. In this study, we propose a hybrid neural network
        based on Long Short-Term Memory (LSTM) and Transformer
        architectures for LEO satellite orbit prediction, which combines
        the sequential processing capabilities of LSTM with the self-
        attention mechanism of the Transformer architecture. The orbit
        propagation errors are first calculated using traditional
        methods, and then the proposed hybrid model is employed to
        predict these errors for orbit correction. Four LEO satellites
        from different orbit altitudes, GRACE-C, SWARM-B, SENTINEL-3A,
        and SENTINEL-6A, are comprehensively evaluated to validate the
        prediction performance of the LSTM-Transformer model and
        corrected orbit accuracy. The results demonstrate that, under
        optimal length of sliding window (WL) parameter conditions, the
        prediction performance of propagation errors using the LSTM-
        Transformer model is improved by 40\%-80\% compared to the LSTM
        model. The corrected orbit accuracy by the predicted propagation
        errors is improved by 40\%-95\% and 5\%-50\% compared to the
        traditional method and the LSTM model, respectively, with no
        systematic bias present. Additionally, the LSTM-Transformer
        model also demonstrates strong generalization capabilities, with
        a 98\% consistency compared with the standard models. During
        solar activity periods, the accuracy of orbit correction using
        this hybrid prediction model has also been improved by more than
        30\% compared with the traditional method.}",
          doi = {10.1016/j.measurement.2026.120973},
       adsurl = {https://ui.adsabs.harvard.edu/abs/2026Meas..27220973Y},
      adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}