Satellite Network Traffic Scheduling Algorithm Based on Multi-Agent Reinforcement Learning

Traditional network traffic scheduling algorithms are difficult to accurately analyze and model. In the face of satellite networks with complex network structures, diverse transmission resources, and high dynamics, routing strategies lack flexibility. In order to reduce the possibility of network congestion in the multi-layer satellite network, a satellite network traffic scheduling algorithm based on multi-agent reinforcement learning is designed. By improving the experience pool and combining the priority sampling mechanism, the training speed and generalization ability of the model are improved. By comparing with the other four classic algorithms, we verified that this algorithm has a faster convergence speed and a better flow balance effect.