Congestion Control and Routing Optimization for LEO Satellite Networks Based on Deep Reinforcement Learning

This paper presents a routing optimization method for LEO networks using DRL. The experimental results demonstrate that the enhanced DQN-based algorithm is effective in optimizing load balancing and alleviating routing congestion. These findings provide strong support for the performance optimization and routing decisions of LEO networks. It can be observed that the average link utilization trained by the MADQN-MQ algorithm is lower, indicating better load conditions. Compared to the Dijkstra and MADQN algorithms, it has a certain advantage in addressing the load balancing issue in the network.