Routing And Spectrum Assignment Algorithms Based On Prediction For Elastic Optical Networks

The traditional optical network with wavelength division multiplexing(WDM)is based on rigid and coarse spectral grids,so it is difficult to adapt to the dynamic and significant changes of the user bandwidth demand,resulting in the low utilization of spectrum resources.Based on optical orthogonal frequency-division multiplexing technology(OOFDM),elastic optical networks(EONs)can flexibly allocate the appropriate size of spectrum resources according to the request bandwidth of users,and they become one of the development trends of optical network in the future.EONs have more advantages than those of WDM optical networks,while the spectrum resources allocation in EONs needs to meet more stringent constraints.Under dynamic traffic conditions,setting up and tearing down connections over time are generating more and more bandwidth fragmentation,which will result in reducing spectral efficiency.Therefore,the spectrum resources allocation in EONs presents a greater challenge.Routing and spectrum assignment(RSA)plays a key role in optimizing the utilization of network resources,and it has become one of the important topics in EONs researching.This thesis focuses on studying the RSA problem in EONs,the main contents and innovations are as follows:Firstly,the majority of the traditional researches for RSA only consider current network status to design algorithms,and few studies focus on the impact of future network status changes on the algorithm performance.However,RSA algorithms could definitely obtain better performance if we predict traffic variations on links and reserve spectrum resources for future arrival connections in advance.This thesis thus proposes a heuristic algorithm,entitled Minimum Comprehensive Weight with Prediction(MCWP),which uses Back Propagation Neural Network to predict the time information of each future arrival connection with holding-time awareness,and simultaneously considers the time overlay relationship among the assigned connections,the future arrival connections and the connection to be assigned.On this basis,the MCWP algorithm calculates the holding-time coincidence,the spectrum utilization and hops of each candidate path in the K shortest candidate paths,and takes the weighted sum of these three parameters as the comprehensive weight.Finally,the candidate path with the minimum comprehensive weight is selected to assign the connection.Compared with existing heuristic RSA algorithms,simulation results show that the MCWP can effectively reduce the blocking ratio and improve the utilization of spectrum resource.The results have been published on the International Conference on Information System and Artificial Intelligence 2016(ISAI 2016)indexed by EI.Secondly,the traditional research of RSA problem uses the Poisson model as the traffic source model,but existing studies have proved that the actual flow characteristics of the network exhibits long range dependence and bursty.Traditional Poisson model with short range dependence fails to describe the actual traffic properties accurately,which implies that the RSA performance has the little credibility under the Poisson model.In this thesis,we compare the different performance of the same RSA algorithm with the Poisson traffic model and the self-similar traffic model,and evaluate the proposed MCWP algorithm under the self-similar traffic to increase its feasibility in practical application.The results have been published on the International Conference on Optical Communications and Networks 2016(ICOCN 2016)indexed by EI.Lastly,the heuristic algorithm is limited to the K shortest candidate paths set,and finds the optimal path in the local solution space,which results in falling into the local optimal solution easily.To improve the global search ability of heuristic algorithms,this thesis proposes Hybrid Ant Colony Optimization with Prediction(HACOP)algorithm,which uses ant agents to search optimal solution in the global solution space,and combines the guiding principle of artificial bee colony to optimize the pheromone update mechanism at the same time.HACOP algorithm distinguishes the pheromone concentration contribution of different ants,strengthens the pheromone contribution of leading ants while weakens the pheromone contribution of misleading ants to avoid the local optimum.Compared with the existing RSA algorithm and the proposed heuristic MCWP algorithm,simulation results show that the HACOP algorithm can effectively improve the RSA algorithm performance.The results are intended to be submitted to the Journal of Optical Fiber Technology.