Research On Computer Network Routing Optimization Algorithm Based On Deep Reinforcement Learning

With the development of 5G,blockchain,artificial intelligence and edge computing.The future requires networks that are more efficient,reliable,secure,open and flexible.Routing,as the core of network transmission,plays an important role in the performance of the whole network.Traditional routing protocol algorithms such as RIP and OSPF have a series of problems such as slow convergence and high average latency in the face of exponential growth of network traffic and different service requirements.In order to solve these problems,through the use of deep reinforcement learning ability,network routing can realize self-learning and optimization,so as to better adapt to the complex network environment and changing network traffic requirements.In this context,this paper analyzes the bottleneck encountered in the development process of traditional routing algorithm,adopts deep reinforcement learning algorithm as the route optimization method,and optimizes it.The main work and contribution of this paper are as follows:1.We propose a centralized routing control mechanism,design an intelligent routing architecture with three-layer logic structure,and introduce a deep reinforcement learning algorithm into the routing control layer to make up for the traditional routing algorithm’s inability to learn from the past experience.Based on network performance indicators,routing policies are dynamically generated on demand to allocate network resources more reasonably.2.For continuous state space optimization problem,a routing optimization algorithm DDPGOR algorithm is designed based on the above control mechanism.And optimize it.In view of the shortcomings of DDPGOR algorithm,a better algorithm model TD3,an asynchronous strategy algorithm,is used to design TD3 OR algorithm to optimize the routing.Simulation results show that both DDPGOR and TD3 OR can effectively improve convergence and stability,significantly reduce end-to-end delay,and improve throughput and link utilization compared with traditional routing algorithms.At the same time,TD3 OR algorithm,as an optimization algorithm of DDPGOR algorithm,has better performance indexes than DDPGOR algorithm.3.Periodic traffic occupies a major part of the network.In order to effectively cope with the possible periodic traffic in the future,a RDPGOR algorithm is designed.In this algorithm,long short-term neural networks are introduced into the depth deterministic strategy gradient algorithm,and attention mechanism is added to predict network traffic,so as to formulate routing policies for periodic traffic.At the same time,in order to further optimize the model,a Bi-RDPGOR algorithm is designed to replace the fully connected network in the original strategy network with the bidirectional long shortterm neural network.Simulation results show that the two algorithms have lower network delay under periodic traffic characteristics.