Density clustering based fast and stable satellite selection for LEO navigation

The rapid deployment of mega low Earth orbit constellations provides unprecedented opportunities for global navigation, but the proliferation of visible satellites creates a critical bottleneck for real-time satellite selection. Existing methods struggle to balance computational complexity with geometric performance: exhaustive search fails to meet real-time requirements, while heuristic algorithms lack theoretical guarantees and temporal stability. This study proposes a density-clustered fast stable selection algorithm. The algorithm identifies high-density regions in celestial sphere distributions, employs hybrid geometric optimization to approximate optimal solutions within constrained search spaces, and reduces frequent handovers through the Forward Stable Strategy. Eight experiments demonstrate that the algorithm achieves near-optimal geometric precision within millisecond-level computation time. Continuous tracking experiments reveal significantly reduced satellite handover frequency, while the algorithm maintains robust performance across diverse scenarios including temporal stability, different latitudes, environmental occlusion, and parameter variations. Ablation studies confirm that the synergy between density clustering initialization and geometric optimization is critical to algorithm effectiveness. This study provides a solution that balances computational efficiency with geometric performance for real-time satellite selection in LEO navigation systems.