A.V. Ksendzuk1, A.V. Semin2, M.K. Melkumyan3

1,2 MIREA – Russian Technological University (Moscow, Russia)
3 Public Joint-Stock Company Interstate Joint-Stock Corporation Vympel (Moscow, Russia)
1–3 sinritzen@yandex.ru

This paper investigates the effectiveness of using neural network models for spacecraft coordinate prediction based on data in TLE (Two-Line Element Set) format. The focus is on comparing the accuracy of trajectory prediction with actual data, where the distance metric between the true and predicted spacecraft trajectories is used to assess quality. Two approaches – neural network prediction and SGP-4 model prediction with consideration of orbital data specificity – are considered. Spacecraft (SC) coordinate prediction is one of the key tasks in space mission control, flight safety and collision avoidance in orbit. With the development of machine learning technologies and neural network methods, it is now possible to consider alternative approaches to SC coordinate prediction. Neural network models can become an effective tool for predicting the coordinates of spacecraft. For successful planning of spacecraft (SC) observations, it is necessary to accurately predict their position and velocity vector in space. The prediction is usually based on the orbital parameters of the spacecraft, represented in TLE (Two-Line Element Set) format. This data, regularly published by NORAD on the platform www.spacetrack.org, is publicly available and is used to calculate the position of satellites. The satellite coordinates and velocity vector are traditionally calculated using the SGP4 (Simplified General Perturbations) model, which serves as the basis for the TLE data. However, studies show that the accuracy of spacecraft coordinate prediction decreases as the time interval between the date of TLE formation and the date of prediction increases. The main reason for this error is a number of simplifications adopted in the motion model and the influence of disturbances that are not fully accounted for in the SGP4 model. The use of more sophisticated physical models may reduce the errors somewhat, but the instabilities in the orbital parameters limit the possibility of significantly improving the forecast accuracy. Nevertheless, assuming that the spacecraft motion is influenced by factors whose statistical parameters are periodic in nature, it is possible to use machine learning techniques such as neural networks for more accurate prediction.

Ksendzuk A.V., Semin A.V., Melkumyan M.K. Coordinate prediction of spacecraft using neural networks. Achievements of modern radioelectronics. 2025. V. 79. № 5. P. 101–109. DOI: https://doi.org/10.18127/j20700784-202505-12 [in Russian]

1. PAO «MAK «Vy`mpel»: oficial`ny`j sajt. URL: https://macvympel.ru/projects/skkp/ (data obrashheniya 05.09.2024).
2. AO «CzNIImash»: oficial`ny`j sajt. URL: https://tsniimash.ru/science/scientific-and-technical-centers/flight-control-center-fcc/security-in-space/index.php?sphrase_id=30569 (data obrashheniya 05.09.2024).
3. Shilin V., Olejnikov I. Oblast` kontrolya – okolozemnoe prostranstvo. Problemy` i perspektivy` razvitiya sistemy` kontrolya kosmicheskogo prostranstva. Vozdushno-kosmicheskaya oborona. URL: http://www.vko.ru/koncepcii/oblast-kontrolya-okolozemnoe-prostranstvo (data obrashheniya 05.09.2024).
4. Hochreiter S., & Schmidhuber J. Long Short-Term Memory. Neural Computation. 1997. 9(8). 1735–1780. Osnovy` i primenenie LSTM dlya obrabotki vremenny`x ryadov.
5. LeCun Y., Bengio Y., & Hinton G. Deep Learning. Nature. 2015. 521(7553). 436–444. Obzor sovremenny`x dostizhenij v oblasti glubokogo obucheniya.
6. Vallado D.A. Fundamentals of Astrodynamics and Applications: Space Technology Library. 2001. Klassicheskoe rukovodstvo, opisy`vayushhee osnovy` orbital`noj mexaniki i model` SGP4.
7. Goh K., & Khan M. Satellite Tracking Using Neural Networks: A Review. Journal of Aerospace Information Systems. 2022. 19(3). 211–229. Obzor primeneniya nejronny`x setej dlya otslezhivaniya sputnikov.
8. Wang T., & Kuan D. Neural Networks for Spacecraft Orbit Determination and Prediction. Journal of Spacecraft and Rockets. 2019. 56(4). 958–967. Primenenie nejrosetej dlya opredeleniya i prognozirovaniya orbit kosmicheskix apparatov.
9. PostgreSQL: oficial`ny`j sajt. URL:  https://www.postgresql.org/docs/current/ (data obrashheniya: 05.09.2024).
10. Cubukcu, Umur & Erdogan, Ozgun & Pathak, Sumedh & Sannakkayala, Sudhakar & Slot, Marco. (2021). Citus: Distributed PostgreSQL for Data-Intensive Applications. 2490-2502. 10.1145/3448016.3457551.
11. Ivanov Yu.P. Kompleksirovanie informacionno-izmeritel`ny`x ustrojstv letatel`ny`x apparatov: Ucheb. posobie dlya vuzov / Pod red. V.A. Bodnera. L.: Mashinostroenie. 1984. 207 s.