Satellite Client Selection Algorithm Based on LHP for Constellation Federated Learning

Satellite constellation systems are pivotal in facilitating global communications and weather forecasting, among other applications. However, the extended communication durations between satellite clients and the scarcity of computing resources necessitate the strategic selection of clients to utilize constellation resources more efficiently in federated learning scenarios. From the perspective of collaboration, this paper transforms the traditional satellite communication system into collaborative modeling between satellite subsystems and proposes a utility model for satellite clients based on the Least Hop Path (LHP). Furthermore, it designs a satellite client selection algorithm based on the LHP utility model. Experimental results indicate that, compared to the random selection algorithm, the proposed algorithm reduces the number of hops by 38.8%. The proposed algorithm can significantly decrease the computational resources required by the constellation system and enhance the performance of constellation federated learning.