Abstract
Forming satellite networks using inter-satellite links among large-scale satellites opens up a potential way of downloading satellite Earth observation big data in near real-time. Nevertheless, the time-varying topologies of satellite networks limit the visible time among satellites and between satellites and ground stations, thus limiting the effective time in which each satellite can receive and send data. This often causes considerable delays in data offloading and downloading for satellites. Besides, limited resources of satellites and sudden arrival of Earth observation data further exacerbate the delays. To cope with these challenges, a novel Self-Organized Autonomous Optimization approach, namely SOAO, is proposed to offload Earth observation big data for large-scale resource-constrained satellites. Specifically, the gradient of each satellite is defined to characterize the relationship between its available resources and the constraints coming from limited resources. Then, to deal with time-varying topology among large-scale satellites, a strategy is designed to update the neighborhood for each satellite, which can indicate their efficient data offload directions. A bidirectional selection-based optimization strategy is proposed for each satellite to make decisions on data offloading via the interaction with neighboring satellites. Finally, the effectiveness of the proposed SOAO is verified by comparing it with those of four baseline algorithms in the context of four different cases. The empirical results demonstrate the superiority of the SOAO by shortening response time from hours to minutes.