Research On Machine Learning Based Routing And Spectrum Assignment Algorithm For Elastic Optical Networks

With the rapid popularization of new technologies such as the Internet of Things,Cloud Computing,5G,and Big Data,many new data services such as mobile internet and high-definition video transmission are growing rapidly.Due to the advantages of low loss,high flexibility and high efficiency,Elastic Optical Network(EON)has become an important topic in the field of optical networks.Routing and spectrum allocation(RSA)is the key technique in EON.Taking into account the difference of different links,the thesis proposes a RSA algorithm based on path leisure degree,and discusses the dynamic relationship between the algorithm and network load.Traffic prediction is applied in the proposed RSA algorithm to improve and optimize the network performance furtherly.Firstly,this thesis introduces the principles and key technologies of EON,discussing major machine learning methods and their applications in optical networks,and focuses on traffic prediction problems and their impact on RSA issues,respectively.Aiming at the RSA problem in EON,this thesis studies the traditional k-shortest-path(KSP)algorithm,and raises the concept of the path integrity degree,including the current integrity degree(CID)and residual integrity degree(RID).The linear weighted sum of the CID and RID is defined as the path leisure degree.Then this thesis combines path leisure degree and traditional KSP algorithm to propose an improved RSA algorithm,which is defined as PLD algorithm.Finally,through the simulation in NSFNET,USNET and CERNET networks,the results indicate that PLD algorithm can effectively reduce the bandwidth blocking probability(BBP)and improve the spectrum resource utilization(SRU),compared with the traditional KSP-FF and KSP-RF algorithms.In addition,the influence on the different values of weight coefficient are also discussed for the trend of BBP and SRU.In addition,this thesis combines the traffic prediction technology with RSA algorithm based on path leisure degree,and then reseaches the adaptive routing and spectrum allocation problems based on traffic prediction.The Elman Neural Network(ENN)is used to predict the traffic in the actual network,and the predicted results are in well agree with the actual data.Based on studying the dynamic trends of weight coefficient with the network load,this chapter combines the predicted results and PLD algorithm to establish an optimization model.This chapter adopts the heuristic genetic algorithm to solve the optimization model,including weight coefficients as optimization parameters and traffic congestion and frequency slots consumption as objective functions.The results show that weight coefficient a is negatively related to the network load and weight coefficient ? is positively related to the network load.During the low-high load,adopting the a and ? that colud change adaptively with the network load can reduce the network congestion and improve the network performance for the most part.The research results reveal the dynamic characteristics that the weight coefficients a and ? should change with the network load.