A Spectrum Allocation Algorithm For Elastic Optical Networks Based On Machine Learning Assisted Graph Coloring Model

With the rapid development of technologies such as big data,cloud computing,and 5G,the types and quantities of communication services have changed,which is requiring higher system capacity and bandwidth transmission interfaces.Compared with traditional wavelength division multiplexing(WDM)optical networks that require fixed grid wavelengths as the minimum granularity for service bandwidth allocation,elastic optical networks(EON)can adapt to service requests with different granularity in a dynamic,complex and diverse service environment to achieve higher spectral utilization and higher flexibility.Routing and spectrum allocation(RSA)is the most critical issue in EON,because the dynamic allocation and release of spectrum will lead to fragmentation of spectrum,which will affect the allocation of subsequent services and the utilization of network spectrum resources.This article focuses on the spectrum allocation problem in EON.Combining the graph coloring theory in graph theory,a method is proposed to establish a graph coloring model,and then transform the spectrum allocation problem into a coloring problem.Two coloring algorithms are designed,and machine learning models are used to simplify the algorithm,thus completing the spectrum allocation of EON.This thesis first introduces the RSA problem of EON,and then sorts out the relevant concepts of graph theory and graph coloring models.The goal of coloring in graph theory is to use the minimum number of colors to complete the coloring of the graph,which is consistent with the goal of using the minimum number of resources in spectrum allocation in elastic optical networks.Thus,the spectrum allocation problem can be transformed into a graph coloring problem with weights.Subsequently,this thesis designed and proposed two improved coloring algorithms,the chain search method and the greedy algorithm.These two algorithms not only utilize the relevant knowledge of graph theory,but also take into account the relevant concepts of elastic optical networks.By coloring the graph model using these two algorithms,the spectrum allocation problem in EON is completed.Theoretical analysis and simulation results show that the improved algorithm can better optimize spectral resources and reduce average link blocking rate compared to a degree maximum coloring algorithm.In addition,for the complex algorithms and iterative processes in the graph model,this thesis considers using a neural network model and using machine learning to assist in the graph coloring model to simplify the algorithm.By building a three-layer deep neural network and training the data using the graph model,the neural network model can also perform spectrum allocation according to the logic of the graph model.Simulation was conducted on NSFNET and USNET networks,and comparative experiments showed that the optimized model of the neural network can greatly reduce the time complexity of the algorithm with slightly reduced accuracy,This is of great significance for the practical application of the graph coloring model.