Research On Dynamic Resource Allocation In Optical Network Based On Graph Neural Network

With 5G commercial,5G network construction in China has been expanding its scope and industry,as a new generation of mobile communication network,the problem of 5G is not only the interpersonal communication,but also solve the people and things,and the communication between,which makes the network information exponential rise,of the business needs of the users are more complex and multivariate,Traditional optical network architecture and transmission technology can no longer meet such a huge information base,resulting in low utilization of network resources,high transmission delay and easy to cause network congestion.Elastic optical network(EON)uses orthogonal frequency division multiplexing(OFDM)to divide network bandwidth resources into smaller granularity and flexibly allocate resources according to the requirements of diversified services on the network.Routing and spectrum allocation(RSA)is a key problem for EON.How to better solve THE RSA problem to reduce network congestion,improve resource utilization,and improve the performance of optical networks is of great importance.This article mainly aims at the resource allocation problem of elastic optical network are studied,focusing on two key problems when RSA distribution,debris and high real-time performance problems,based on the figure of the neural network resource allocation optimization strategy,first to study the static scene,and then study key dynamic scene,and proving its feasibility through the experimental platform,The main work of the paper is carried out from the following aspects:1.A study on spectrum allocation scheme of optical network service based on graph neural network.For static scenarios through traditional RSA strategy,after light LuJianLi spectrum fragmentation serious lead to network resource consumption is too large,the paper first analyzes the existing fragments of the spectrum in EON evaluation method,this paper proposes a minimize network spectrum state change degree of spectrum allocation scheme,based on the business distribution spectrum is compared before and after the link,The minimum degree of change is the spectrum allocation criteria.On the other hand,a static RSA strategy based on graph neural network(GNN)is proposed.The combination of graph neural network and reinforcement learning(RL)is adopted.The graph neural network adopts message propagation neural network(MPNN)architecture.Then,RL uses ε-greedy strategy to select each move,and finally selects the action set with the highest overall reward value as the final decision.The simulation and result analysis of the algorithm and other two traditional algorithms in NSFNET network and USNET network show that the algorithm in this paper can indeed make the flow distribution more balanced,better reduce the fragmentation of spectrum,reduce the overall consumption of network resources.2.The study of dynamic RSA problem based on graph neural network.On the one hand,considering the sudden and diversified services in dynamic scenarios,a large number of spectrum fragments can not be utilized in the network during the establishment of optical paths,which leads to the decrease of the overall resource utilization rate of the network.On the other hand,considering the requirement of high real-time resource allocation in dynamic scenarios,a dynamic RSA strategy based on GNN is proposed.K were first introduced in the article the most short-circuit algorithm are selected for the requests of business K alternative paths,and then according to the proposed spectrum allocation judgment standard,spectrum allocation for K path after light LuJianLi scheme defined bonus,geri weis-corbley model through training,the reward value calculation,and finally choose the maximum value of incentives of RSA strategy for business to establish the light path.The simulation and result analysis of the proposed algorithm and other two traditional algorithms in NSFNET and USNET networks show that the proposed algorithm can effectively improve the response time of traffic and reduce the generation of spectrum fragments,so that the network can carry more traffic,and thus reduce the blocking rate of traffic.Finally,an optical network control platform based on RYU controller is built,and the proposed algorithm is used to select routes for services on the platform.Finally,the successful transmission of services proves the feasibility of the algorithm.