A Machine Learning Approach to Monitor Satellite Constellation Orbits

Abstract

Keeping a satellite in the right orbit is no easy feat. It’s constantly influenced by various forces like Earth’s uneven gravity, air resistance high up in the atmosphere, and even pressure from sun-light. Over time, these factors can nudge satellites off their intended paths. Nowadays, most satel-lites depend on ground control teams to keep an eye on their orbits and plan any necessary adjust-ments. This can be a slow, expensive, and tricky process, especially as the number of satellites in space keeps increasing. This paper outlines a machine learning approach to check for deviation from the intended orbit of a given satellite con-stellation. The system continuously tracks essen-tial orbital data, such as position and velocity, to see if a satellite is veering off course. To develop and test this method, we used publicly available data from Starlink satellites and other spacecraft. Starlink satellites are particularly valuable because they frequently adjust their orbits and operate in large clusters, providing rich and dynamic data. We trained a machine learning model to predict if a satellite has left from its constellation. By com-paring this prediction with the actual position of the satellite, the system can spot when the orbit starts to drift. The findings indicate that blend-ing basic physics with machine learning can ef-fectively aid in autonomous orbit correction. This approach holds promise for cutting costs, boost-ing efficiency, and simplifying the management of a growing number of satellites in the future.