Quantum Machine Learning for Satellite Trajectory Prediction: A 3-Axis Specialized Quantum Regressor
Year: 2026
Type: conference-paper
Abstract: The task of accurate prediction of satellite trajectories is a critical facade deployed in many broader systems such as collision avoidance systems, orbital maintenance systems, and broader space situational awareness. In this paper, we will attempt to deploy the emerging field of hybrid quantum classical computation and quantum machine learning principles in the task of 24-hour-ahead prediction of satellites’ position and velocity in Earth-centered inertial (ECI) coordinates: This will require prediction on six variables: x, y, and z coordinates and x, y, and z velocities. We use real TLE data from Celestrak for Sentinel-1A, with which we generate 720-hour trajectories via SGP4 propagation to be a dataset for target features regression. Our framework extracts 33 features per axis, combining statistical measures, orbital mechanics invariants, and frequency-domain features from 72-hour sliding windows. Three-axis specialized quantum regressors employing IQP encoding are trained on 6 qubits with 3 variational layers. Each axis regressor predicts both position and velocity for that axis, achieving R² scores of 0.994-0.999 on test data. A parallel quantum circuit architecture is also evaluated, partitioning features across multiple circuits. Benchmark comparisons against classical baselines show that our quantum approach is competitive with, and in several cases exceeds, published classical performance for 24-hour satellite state prediction, with significant model size advantages.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W7225018121