Deep-reinforced Optical Path Configuartion Scheme Of Nonlinear Elastic Optical Networks

The explosive growth of emerging applications(e.g.,cloud computing)and the widespred adoption of new networking paradigms(e.g.,the Internet of Things)are demanding a new network infrastructure that can support dynamic,high-capacity and quality-of-transmission(Qo T)-guaranteed end-to-end services,which traditional optical wave division multiplexing(WDM)netwroks are increasingly unable to meet.Compared with traditional fixed grid WDM networks,elastic optical networks(EONs)can flexibly establish super channels by sorting out a series of finer grained subcarriers,and can adjust the modulation format according to transmission quality requirements.Therefore,in recent years,EON has been considered as one of the most promising network technologies for the next generation of backbone networks.At the same time,the flexible resource allocation mechanism of EON makes the corresponding service configuration design more complicated.In order to improve the utilization of EON spectrum resources,it is necessary to conduct in-depth research on routing,modulation format and spectral resource allocation.Recently,our research group has done indepth research on dynamic resource algorithms of elasto-optical networks and proposed some resource allocation schemes.In this paper,an in-depth investigation on the problem of elastomer optical network resource allocation is carried out,and the domestic and foreign researches on the problem of elastomer optical network resource allocation in recent years are reviewed.On the basis of the existing research,the research on the resource allocation algorithm of nonlinear elasto-optical network based on deep reinforcement learning is carried out.In this paper,an optical path configuration scheme for nonlinear EONs based on deep reinforcement learning is proposed.When the service request arrives,the neural network can perceive the state of the nonlinear elastic optical network,and the agent will select an action according to the state to try to establish an optical path to serve the request,and the descendants of the action implementation will get the feedback of the previous action.Through the interaction with the environment,parameters of the deep reinforcement learning strategy network can be trained,so as to learn the correct optical path configuration scheme of the nonlinear elastic optical network.We set up a state space model,which mainly includes the information of business request and the state of network topology.Depth in order to facilitate the reinforcement learning agent for learning,we are not simply be elastic optical network in each link of each frequency gap state into consideration,but consider the article will be related to the current business request K alternate path based on the above M different modulation format of J a fast spectral information as a state space representation model of network topology information.Different from the scheme that uses the transmission physical distance limit to determine the modulation format,we introduce the Gaussian noise model to calculate the SNR of the service request,so as to select the modulation format that meets the SNR limit conditions.When the business request arrives,we will prepare K·M·J schemes for it,and calculate the SNR of the connection in each scheme,and then compare it with the threshold value of the modulation format.It should be noted that,unlike static optical path configuration,in dynamic optical path configuration,the impact of future business requests on current business requests also needs to be considered.Through simulation,this paper verifies the correctness of the optical path configuration algorithm of nonlinear elastic optical network proposed above based on deep reinforcement learning,and compares the influences of different neural network parameters and different numbers of K and J on the optical path configuration algorithm.At the same time,this paper also compares the blocking rate between the proposed algorithm and the benchmark algorithm.According to the simulation results,the requested blocking rate is reduced by about 46.67%compared with the benchmark algorithm.