Machine Learning for LEO and MEO Satellite Orbit Prediction

Abstract

Accurate orbit prediction plays a significant role in many space geodesy applications, including space situational awareness, orbital maneuvers, and satellite navigation in real-time scenarios. The traditional orbit prediction approach includes analytical and numerical algorithms to accurately propagate the satellites’ state and associated uncertainties. However, these methods are based on limited dynamic models and may have limitations in accurately capturing the complex dynamics of satellite motion. With the rapid growth in computing power and the development of advanced machine learning (ML) and deep learning (DL) algorithms, there has been growing interest in leveraging ML algorithms to enhance orbit prediction accuracy. In this study, we investigate the potential of using different ML algorithms for the orbit prediction of low Earth orbit (LEO) satellites. We also extend our analysis to medium Earth orbit (MEO) satellites. LEO Swarm-A satellite’s precise orbit products and MEO GNSS satellites’ final ephemeris products are used. The datasets are pre-processed and used in training various ML models. The ML models are then used to estimate the position and velocity of the LEO and MEO satellites. The accuracy of both LEO Swarm-A satellite and MEO GNSS satellites’ using various ML models are compared and discussed. Overall, the standalone ML-based method holds significant promise for improving orbit prediction accuracy and reliability for SWARM-A satellite and GNSS satellites, ultimately enhancing the performance and efficiency of space-based applications and services.