Research on Intelligent Routing Technology Based on Improved DQN

With the rise of artificial intelligence, intelligent routing technology has become a research hotspot in the current academic circles. In view of the problems of poor load balancing ability of traditional routing algorithms and difficulty in guaranteeing quality of service (QoS), this paper proposes an intelligent routing algorithm DQN-Route based on deep reinforcement learning (DRL). The algorithm adopts the deep Q network (DQN) as the training framework, and introduces the convolutional neural network (CNN) as the neural network. Based on the algorithm advantages of DQN processing continuous state space, as well as the local perception and parameter sharing capabilities of CNN, we can input the dynamically changing network state into the neural network for training. After the algorithm training converges, the action value output by the neural network is used as the network link weight to realize the dynamic adjustment of the routing strategy. Finally, the DQN-Route routing algorithm is compared with the OSPF, ECMP and Q-Learning routing algorithms respectively. The results show that the DQN-Route has better convergence, and compared with the Q-Learning routing algorithm, DQN-Route reduces the delay by 14.13%, increases the throughput by 11.34%, and reduces the packet loss rate by 9.17%.