Major space companies are developing satellite mega-constellations to provide global Internet coverage and services. Limited battery capacity is one of the biggest obstacles on mega-constellations due to the restricted weight and volume of satellites. Massive Internet packet routing tasks pose a big challenge to the energy system in such mega constellations. Incorrect use of satellite batteries during routing phases may significantly increase the energy consumption and cause node failure quickly.Existing state-of-the-art works on energy-saving routing for satellite networks paid much attention on traffic distribution and end-to-end delay issues.However, these methodologies were using many real-time network information for optimization which is not practical in mega-constellations.Note also that these works did not consider energy efficiency and guaranteed end-to-end delay simultaneously. In this paper, we propose a novel deep reinforcement learning based energy-efficient routing protocol called DRL-ER, which avoids the battery energy imbalance of constellations and can also guarantee a required bounded end-to-end delay. In DRL-ER, satellites can learn a routing policy that will balance energy usage among satellites.Extensive simulation results show that our proposed DRL-ER protocol reduces the energy consumption of satellites in average by more than 55% compared to the current state-of-the-art work, and prolongs the lifetime of constellations significantly.
|Number of pages||12|
|Journal||IEEE Transactions on Network Science and Engineering|
|Publication status||Published - 23 Nov 2020|