Research On SDN Intelligent Routing Optimization Based On Deep Reinforcement Learning

The traditional network is tightly coupled with the control plane and the forwarding plane,which makes the system upgrade cost greater,and the network operation and and the network operation and maintenance workload is relatively large.and the network operation and maintenance workload is relatively large.The control function is centralized by the software-defined network controller.In this new type of network structure,how the controller intelligently schedules dynamic network traffic has become one of the important issues in the network field.Deep reinforcement learning,which combines deep learning awareness and reinforcement learning decision-making ability,can accurately and timely sense network traffic changes on the one hand,and on the other hand,it can incentivize good feedback and discourage negative feedback,and its good performance is being used in more and more fields.attention.Deep reinforcement learning is divided into two categories: based on value function and based on policy gradient.These two kinds of deep reinforcement learning have certain work in the intelligent research of SDN controller.This paper analyzes the limitations of non-intelligent routing under SDN,which is single and unable to learn the previous wrong experience,so as to effectively avoid recurrence,and discusses the problems of existing non-intelligent routing mechanisms under SDN.In response to this problem,the current case of the combination of SDN routing mechanism and artificial intelligence is introduced,and combined with the above research and discussion,an SDN intelligent routing optimization method based on deep reinforcement learning is proposed.This article completes the following improvements:1.Researched the current related work of intelligent routing based on Deep Q Network(DQN),noting that many models use the same neural network architecture when training neural networks,and only change the network parameters to change for training,this method has problems such as insufficient network throughput during actual use.Therefore,this paper proposes a neural network with different structures for different destination nodes to effectively solve the problem of dynamic routing control in the network.Simulation results show that the method has good convergence and stability,and effectively improves the network throughput.2.Due to the problem that DQN cannot optimize on continuous space and state,Deep Deterministic Policy Gradient(DDPG)is used to continuously optimize the scheduling of traffic.Based on DDPG,the paper uses small-scale adjustments on the offline training line to improve agent convergence and ease resource overhead.Simulation results show that the DDPG-based SDN routing optimization method has good convergence and effectiveness,and DQN-based Compared with the intelligent routing method,this method has great practical significance and value for the operation and maintenance management of the SDN network.