Answers are grounded in this paper and cite their sources.
The optimisation of satellite constellation design has become increasingly critical as global reliance on satellite communication and data networks expands. Modern constellations have to balance connectivity, resilience, and energy efficiency while adapting to dynamic orbital conditions and intermittent link availability. Traditional graph-theoretic methods, such as minimum spanning trees (MST), greedy heuristics, and spectral clustering, provide some optimisation but are inherently static. They often struggle in highly dynamic satellite environments where topological changes occur frequently, reducing resilience and suboptimal resource utilization. To address these limitations, this study integrates reinforcement learning with graph-theoretic modelling to optimize constellation performance adaptively. Three reinforcement learning algorithms, Deep Q-Network (DQN), Proximal Policy Optimization (PPO), and Advantage Actor-Critic (A2C), were implemented and evaluated on a large-scale dataset of 50,000 satellite records. The proposed framework models the constellation as a weighted graph, where links are defined by latency, bandwidth, and communication success rates. Experimental results demonstrate that reinforcement learning significantly enhances resilience, with values reaching 0.98 compared to 0.61 under MST, while maintaining high connectivity (≈0.98). Among the evaluated algorithms, A2C demonstrated the most stable reward distribution, highlighting its potential for practical deployment.
No takes yet. Share an insight, caveat, or question.
Круглова et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: