Research On Tidal Flow Prediction And Resource Allocation In Optical Network

With the rise of a series of new technologies such as 5G,big data,and artificial intelligence,as well as the popularization of smart mobile terminals,and the iteration of software for social,entertainment,and office scenarios,the overall scale and traffic demand of the mobile Internet continue to surge.Network hardware facilities and technical service capabilities put forward higher requirements.In the metropolitan area network,because users access the Internet in different locations in activities,life,and office scenarios,and the access locations will change over time,the link load will also change periodically over time.This phenomenon is called the tidal flow effect.If the optical network tidal traffic can be explored,the load changes in the subsequent period can be predicted,and flexible and dynamic resource allocation can be realized to cope with large-scale traffic aggregation and service request access in a short period of time.This paper conducts research on the phenomenon of tidal flow in optical networks.The main research contents are as follows:(1)Based on the investigation of tidal phenomena,the subject built a dynamic simulation scene that conforms to the temporal and spatial variation of MAN traffic access.By designing the network topology,simulating business changes,and formulating routing rules,the link traffic data is output,which restores the characterization of the tidal effect of the link load dimension to a large extent.(2)Focusing on the tidal scene of the metropolitan area network,the project proposes the ARIMA-LSTM traffic prediction model.The model can capture the linear and nonlinear characteristics of time series,and control the prediction error within a small range.Experiments show that the mean absolute error of the ARIMA-LSTM model is 2.533,the root mean square error is 4.510,and the mean absolute percentage error is 0.199.Compared with the single ARIMA model and LSTM model,it has obvious advantages in prediction accuracy.(3)Aiming at the problems of uneven load,low resource utilization,and business congestion caused by traditional routing and spectrum allocation strategies in tidal scenarios,the project innovatively proposed a traffic-aware strategy TA-KSP-DQN that integrates reinforcement learning.This strategy comprehensively considers the path load and business requirements,and uses the feature extraction ability of neural network and the decision-making ability of reinforcement learning to allocate resources.Compared with the traditional routing and spectrum allocation strategy SPFF,the TA-KSP-DQN strategy effectively reduces the variance of link traffic load at tidal time.In the resource allocation of the time dimension(unit:day),the service blocking rate is reduced by 2.05%.Resource utilization increased by 19.18%.