항공우주시스템공학회지 (Journal of Aerospace System Engineering)

  • 제17권3호
  • /
  • Pages.57-67
  • /
  • 2023
  • /
  • 1976-6300(pISSN)
  • /
  • 2508-7150(eISSN)
항공우주시스템공학회 (The Society for Aerospace System Engineering)
권호 표지
DOI QR코드

DOI QR Code

인공 신경망을 이용한 저궤도 위성의 궤도 예측 알고리즘 개발

Algorithm for Orbit Prediction of LEO satellites using Neural Network

  • 김지훈 (연세대학교 천문우주학과 우주비행제어연구실) ;
  • 박상영 (연세대학교 천문우주학과 우주비행제어연구실) ;
  • 전상미 (레이다 연구소, LIG 넥스원)
  • Jee Hoon Kim (Astrodynamics and Control Laboratory, Dept. of Astronomy, Yonsei University) ;
  • Sang-Young Park (Astrodynamics and Control Laboratory, Dept. of Astronomy, Yonsei University) ;
  • Sangmi Chon (Radar R&D, LIG Nex1)
  • 투고 : 2023.01.06
  • 심사 : 2023.04.28
  • 발행 : 2023.06.30
PDF 다운로드
⟨ 이전 논문 다음 논문 ⟩

초록

본 연구는 기계학습을 통해 저궤도 위성의 궤도를 예측하는 새로운 알고리즘을 개발하였다. TLE 데이터와 SGP4 알고리즘으로 시계열 데이터를 생성하고, 이심률에 따른 특이점 문제를 해결하기 위해 수정분점 궤도 요소를 기계학습 요소로 설정하였다. 데이터 분할, 데이터 스케일링 등의 전처리 과정을 거치고, LSTM 레이어를 포함하는 모델과 최소 제곱법으로 선형 회귀를 하는 학습 모델로 학습하였다. 세가지 위성에 해당 알고리즘을 적용하여 약 8 일간 8.2 km 위치 오차 이내로 궤도를 예측할 수 있었다. 개발된 알고리즘은 기계학습으로 미래의 위성의 궤도를 SPG4 전파기의 정밀도 내에서 예측할 수 있다.

This study presents an algorithm for the orbit prediction of Low Earth Orbit (LEO) satellites using a machine learning. Utilizing the Two-Line Elements (TLE), time series data are generated through the Simplified General Perturbations 4 (SGP4) algorithm. Learning parameters are set-up using the modified equinoctial orbital elements. Data pre-processing was conducted, including data decomposition and data scaling. The learning model contains a Long Short-Term Memory (LSTM) layer and an ordinary least squares linear regression. As a result, orbit prediction of three satellites is performed within 8.2 km position error for about 8 days. This means that the future states of the satellite can be predicted within the accuracy of the SGP4 propagation.

키워드

과제정보

이 논문은 국방과학연구소 재원으로 LIG넥스원의 "디지털배열 표적신호 획득처리기술개발" 과제 지원을 받아 수행되었음.

참고문헌

  1. S. P. Shuster, "A Survey and Performance Analysis of Orbit Propagators for LEO, GEO, and Highly Elliptical Orbits," pp. 3-15, 2017.
  2. David A. Vallado, "Fundamentals of Astrodynamics and Applications", 2nd Ed., Springer, pp. 619-636, 1997.
  3. F. Hoots and R. Roehrich. "SPACETRACK REPORT NO. 3 Models for Propagation of NORAD Element Sets," pp. 10-20, December 1988.
  4. H. Peng, and X. Bai, "Artificial Neural Network-Based Machine Learning Approach to Improve Orbit Prediction Accuracy," Journal of Spacecraft and Rockets, pp. 1248-1260, June 2018.
  5. Hao Ren et al. "Research on Satellite Orbit Prediction Based on Neural Network Algorithm," Proc. of the 2019 3rd High Performance Computing and Cluster Technologies Conference, pp. 267-273, June 2019.
  6. H. Yang, J. Zhu, and J. Zhang, "The Research of Low Earth Orbit Prediction of Satellite Based on Deep Neural Network," DEStech Transactions on Computer Science and Engineering, pp. 291-299, March 2018.
  7. L. Fang, and N. Nagarajan, "Neural Network Based Orbit Propagation for Small Satellite Missions," 18th Annual AIAA/ASU Conference on Small Satellites, 2004.
  8. D. Wei, and C. Zhao. "An Accuracy Analysis of the SGP4/SDP4 Model," Chinese Astronomy and Astrophysics, vol. 34, pp. 69-76, 2010. https://doi.org/10.1016/j.chinastron.2009.12.009
  9. R. Wand, J. Liu, and Q.M. Zhang. "Propagation error analysis of TLE data," Advances in Space Research, vol. 43, pp. 1065-1069, 2009. https://doi.org/10.1016/j.asr.2008.11.017
  10. J. Mason, "Development of a MATLAB/STK TLE Accuracy Assessment Tool, in support of the NASA Ames Space Traffic Management Project," arXiv: Space Physics, 2013.
  11. R. Rojas, "Neural Networks - A Systematic Introduction", Springer-Verlag, Berlin, New-York, pp. 151-184, 1996.
  12. A. B. Arrieta et al. "Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges toward Responsible AI," Information Fusion, pp. 82-125, 2020.
  13. M. J. H. Walker, B. Ireland, and J. Owens, "A set of modified equinoctial orbit elements," Celestial mechanics vol. 36, pp. 409-419, 1985. https://doi.org/10.1007/BF01227493
  14. Cross Validation manual,  https://medium.com/@soumyachess1496/crossvalidation-in-time-series-566ae4981ce4
  15. R. J. Hyndman and G. Athanasopoulos, "Forecasting: principles and practice", 2nd Ed, OTexts: Melbourne, Australia, pp. 197-236, 2018.
  16. Time Series Decomposition manual,  https://towardsdatascience.com/different-types-of-timeseries-decomposition-396c09f92693
  17. Data Scaling manual,  https://machinelearningmastery.com/how-to-improveneural-network-stability-and-modeling-performance-withdata-scaling/
  18. Hyperparameter and Optimization manual,  https://towardsdatascience.com/understandinghyperparameters-and-its-optimisation-techniquesf0debba07568
  19. Overfitting and Underfitting manual,  https://machinelearningmastery.com/diagnose-overfittingunderfitting-lstm-models/
  20. Early Stopping manual,  https://towardsdatascience.com/a-practical-introductionto-early-stopping-in-machine-learning-550ac88bc8fd
  21. J. H. Kim, "Orbit Prediction of LEO satellites using Machine Learning," MS Thesis, Dept. of Astronomy, Yonsei University, Feb 2022.