QoS-Based MADDPG-Enhanced Load Balancing Scheme for Multi-LEO Satellites Network.

With the rapid development of wireless communication and the rise of the requirement for seamless coverage, the deployment of Low Earth orbit (LEO) satellites for providing communication services to terrestrial devices is receiving a great deal of attention from academia and industry. However, within the wider coverage area of LEO, a diversity of devices may be covered. The limited resources of a single satellite and its characteristics of high-speed movement relative to the ground may lead to the failure to meet the demands of diverse services of ground devices, causing troubles for the stable connection and efficient information transmission of communication networks. To ensure the full utilization of resources and stable connection of devices, this paper considers three performance indicators for the multi-LEO satellite communication scenario, namely, communication transmission rate, satellite load, and remaining service time, and constructs a multi-satellite load balancing problem, which is aimed at maximizing the weighted sum of multiple indicators. In this paper, we transform the above optimization problem into Markov decision process (MDP) and solved by the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm. The simulation results show that the proposed strategy can improve the multi-satellite load balancing performance significantly and has faster convergence compared with traditional Deep Q Network-based (DQN-based) strategy and random selection load balancing scheme.