Predicting the Properties of Resident Space Objects in LEO Using Graph Neural Networks

Predicting the Properties of Resident Space Objects in LEO Using Graph Neural Networks

Utah State University, Logan, UT

In recent years, the number of resident space objects (RSOs) in low Earth orbit (LEO) has significantly increased. As a result, decision-making tasks and the overall understanding within the space domain have become more challenging. To make critical decisions such as collision avoidance manoeuvres, satellite owners and operators must be aware of their assets’ surroundings. Therefore, spacecraft operators must understand the characteristics of these RSOs and their spatial relationships. To address this issue, we propose a novel approach using graph neural networks (GNNs) to predict RSOs’ properties in LEO. This approach captures the complex interdependencies between RSOs by representing each object as a node in the graph and defining the edges based on the objects’ spatial proximity, quantified by the orbit altitude differences. We demonstrate the potential of using GNNs to infer missing RSOs’ properties by training the network using masked nodes. Additionally, we discuss how a given neighbourhood can assist at an operator-specific level, improving space situational awareness, and enabling a more informed decision-making process.