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

In recent years,with the rapid development of all kinds of new Internet services,the business volume keeps increasing,and the existing network system is difficult to bear the explosive growth of business volume.The WDM networks that used to be widely deployed can no longer meet the business needs of today.Elastic optical network is considered as a powerful technology to solve the above problems and has gained wide attention.There are two key problems in elastic optical networks:first,resource allocation of routing and spectrum,referred to as RSA problem;Second,spectrum fragmentation as business comes in and goes out.How to allocate and arrange network resources reasonably is the key to improve resource utilization and optical network performance.In this paper,the problem of resource allocation in elastic optical networks is studied and a resource allocation algorithm based on multi-feature extraction is proposed.Aiming at spectrum defragmentation in optical networks,a spectrum defragmentation algorithm based on multi-feature extraction after service sorting is proposed.The performance of the algorithm is verified by simulation.The research contents and innovations of this paper are as follows:Research on dynamic RSA problem based on multi-feature extraction.Traditional network resource allocation algorithms have high complexity and low efficiency.When a large number of dynamic services enter an elastic optical network,the network performance deteriorates significantly after resources are allocated to services.This paper presents a dynamic RSA strategy based on multi-feature extraction in the dynamic scenario of multi-type business and massive business,taking the intensity of network spectrum occupation as the objective function.This strategy in combination with elastic optical spectrum characteristics and method of machine learning algorithms of feature extraction,first determine the objective function produces the allocation of resources needed by neural network data sets,the second link between adjacent network is extracted through different convolution kernels spectrum characteristics of the complex network spectrum table separation step by step,the neural network output spectrum,Use residual graph method and shortest path method to get RSA route.Simulation results show that compared with the traditional path-first algorithm,the network resource allocation algorithm based on multi-feature extraction proposed in this paper is significantly improved in terms of allocation rate and spectrum fragmentation.Research on spectrum defragmentation based on multi-feature extraction.In order to solve the problem of spectrum fragmentation in dynamic network,this paper proposes a spectrum arrangement strategy based on multi-feature extraction.First to prioritize network existing business,and then according to the sorting result judgment when spectrum finishing business needs change,determined by specific objective function spectrum position occupied business changes,to produce more feature extraction network data sets,using the frequency spectrum characteristics of the training data set more extract spectrum organizing network.In dynamic optical network,the results of spectrum arrangement are determined by using the trained multi-feature extraction network.Simulation results show that the proposed algorithm can increase the carrying capacity of the network and reduce blocking rate.