An Intelligent Routing Algorithm for LEO Satellites Based on Deep Reinforcement Learning
Peiliang Zuo, Chen Wang, Ze Kun Yao, Shaolong Hou, Hua Jiang
Abstract
LEO satellite network (LEO-SN) plays an important role in the sixth generation mobile networks (6G) and the vision of space-air-ground integrated network (SAGIN). Unfortunately, the traditional routing decision methods for LEO-SN based on the central mode are facing more and more difficult challenges due to the complex orbits, large quantity, diverse loads, different storage spaces and communication capabilities of the satellite nodes. This paper proposes an intelligent decentralized routing algorithm for satellite nodes in dynamic LEO-SN. The nodes can select the return link adaptively according to the spatial position, mutual distance, queuing delay and available bandwidth of the surrounding satellite nodes on the basis of the model obtained through training of the Deep Q Networks (DQN). Simulation and analysis show that the proposed intelligent routing method possesses well convergence characteristics and generalization ability, and has better latency performance compared with a variety of routing methods.