On Analysis and Optimization of Low Earth Orbit Satellite Constellations

Abstract

This thesis presents a comprehensive framework for analyzing, modeling, and optimizing large-scale Low Earth orbit (LEO) satellite constellations, utilizing Starlink as a case study. By reframing constellation behavior as a dynamic programming problem, the research effectively decomposes high-dimensional data, derived from historical Two-Line Elements (TLE), into Reduced Orbital Parameter Vectors (ROPVs). These vectors are embedded into low-dimensional submanifolds using Uniform Manifold Approximation and Projection (UMAP) and clustered using Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) to identify distinct “satellite families.” Within each family, critical physical parameters are inferred and validated amidst atmospheric uncertainties, informed by dynamic thermospheric models and solar–geomagnetic indices, enhanced by deep learning-based solar radio flux forecasting. Furthermore, a high-fidelty orbital simulation is integrated with physics-informed reduced order modeling techniques to parameterize the low-thrust performance of krypton Hall-effect thrusters, enabling satellite constellation multi-objective optimization for phasing, station-keeping, and propellant allocation.