Machine Learning Based Resource Allocation And Assignment Technology In Optical Networks And Its Applications

With the vigorous development of many new generation information technology such as the 5^thgeneration(5G)wireless technology,the rapid growth of network traffic has proposed ever higher requirements for optical transmission networks.The traditional WDM optical network has poor resource utilization which makes it lack of the capability to satisfy current and future network environment requirements.The orthogonal frequency division multiplexing(OFDM)based Elastic optical networks(EON)has been widely accepted as a promising solution for future optical networks due to its high flexibility and good resource utilization.How to reasonably implement routing and allocate spectrum resources for incoming traffic is the focus of academia.In this thesis,the routing and spectrum allocation(RSA)issue in EON is studied comprehensively through theoretical analysis and numerical simulation based on machine learning technology and its applications.Firstly,some fundamental concepts of EON and the routing and spectrum allocation problems are introduced briefly in this thesis,and the network traffic prediction model is analyzed in detail.Some important machine learning algorithms such as back propagation neural network and its application in traffic prediction are analyzed deeply.Focusing the shortcoming of traditional RSA algorithms which is limited to single domain information,this thesis proposes a multidimensional awareness routing algorithm combined with prediction in EON,using BPNN model and history traffic time information to forecast the coming traffic information,and then integrate the information of other two dimensions in spectrum and adjacent links to consideration when routing,the path which has the minimal degree of comprehensive resource contention is chosen over the candidate paths to establish connection,simulation results show that the proposed routing algorithm,compared with other routing algorithms which only consider one single dimension,obtains the optimization and has higher resource utilization.Finally,in view of the current spectrum allocation algorithms are mostly only consider the spectral dimension,and ignore the differences in time dimension between traffic will also affect the bearing capacity of the link to the future,this thesis puts forward a spectrum allocation algorithm based on time and frequency domain,jointly consider the metrics setted in the time domain and frequency domain when allocating spectrum,choose the location which has minimal effect on bearing ability of link to allocate spectrum on basis of the current spectrum usage,simulation results show that the proposed spectrum allocation compared to the mostly used algorithm,such as FF,and the algorithm only considers the distribution of the spectral dimension has lower blocking probability,and improve the spectrum utilization.