EUVIMEDEUVIMED entdecken
Crossref

Hybrid Neural Orbit Propagation Under Sparse Space-Based Observations

Jiaping Xiao, Lu Bai, Mir Feroskhan

IEEE Transactions on Aerospace and Electronic Systems · 2026 · 62 · 9199-9212

DOI: 10.1109/taes.2026.3680858 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Aerospace and Electronic Systems, 2026. Verfasst von Jiaping Xiao und 2 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Gefördert von: Ministry of Education - Singapore, MOE Academic Research Fund Tier 2.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate long-term orbit propagation (OP) is a core capability for space situational awareness (SSA), underpinning conjunction assessment, collision avoidance, and catalog maintenance. High-precision orbit propagation (HPOP) models can accurately integrate the motion equations without relying on two-line elements, using only initial conditions. In operational space-based SSA, however, initial states are usually obtained from sparse angles-only measurements, and the environment is uncertain, especially in atmospheric density and solar radiation pressure. These factors cause the propagated trajectories to diverge over time, even when using sophisticated numerical propagators. This article proposes a hybrid physics-based residual learning framework that integrates an HPOP model with a residual-correction module based on long short-term memory (LSTM) networks. The LSTM operates on sliding windows of propagated states and learns to predict systematic residual errors that accumulate over long horizons due to model deficiencies and orbit determination (OD) biases. The corrected state is obtained by adding the learned residuals to the HPOP prediction. We derive the HPOP–LSTM hybrid neural OP formulation, describe the residual-learning architecture, and present numerical experiments in low Earth orbit, medium Earth orbit, and geostationary orbit scenarios across various observation perspectives. Results show that the proposed hybrid model substantially reduces propagation errors from 400 to 1 km over 24 h across various orbit types, particularly in the presence of OD errors typical of sparse angles-only tracking.

Förderung

  • Ministry of Education - Singapore
  • MOE Academic Research Fund Tier 2

Zitieren

APA

Xiao, J., Bai, L., & Feroskhan, M. (2026). Hybrid Neural Orbit Propagation Under Sparse Space-Based Observations. IEEE Transactions on Aerospace and Electronic Systems, 62, 9199-9212. https://doi.org/10.1109/taes.2026.3680858

Vancouver

Xiao J, Bai L, Feroskhan M. Hybrid Neural Orbit Propagation Under Sparse Space-Based Observations. IEEE Transactions on Aerospace and Electronic Systems. 2026;62:9199-9212. doi:10.1109/taes.2026.3680858

BibTeX
@article{Xiao2026,
  author = {Xiao, Jiaping and Bai, Lu and Feroskhan, Mir},
  title = {Hybrid Neural Orbit Propagation Under Sparse Space-Based Observations},
  journal = {IEEE Transactions on Aerospace and Electronic Systems},
  year = {2026},
  volume = {62},
  pages = {9199--9212},
  doi = {10.1109/taes.2026.3680858}
}
Kostenlos starten →