Deep Reinforcement Learning-Based Intelligent Routing Scheme for Low Earth Orbit Satellite Communication Networks

Low earth orbit (LEO) satellite networks have garnered wide attention in satellite communications owing to their low transmission latency and wide coverage. Nevertheless, traditional routing methods are insufficient to adapt to the highly dynamic nature and load balancing requirements of satellite communication networks. Therefore, this paper proposes a distributed routing scheme for LEO satellite communications networks based on dueling double deep Q-network (D3QN-DR). By minimizing the queue occupancy rate, the transmission delay, and the remaining distance to the target, the routing scheme of each satellite is modeled as a markov decision process by monitoring the link delays and status of neighboring nodes to adaptively select the optimal path under various uncertainties. According to the simulation results, the proposed scheme can significantly mitigate node congestion and decrease the packet drop ratio.