Research On Machine Learning Assisted Traffic Grooming Algorithm In Optoelectronic Hybrid Networks

With the development of 5th generation mobile networks,user traffic is increasing and user mobility is strengthening.The dynamicity of services poses a great challenge on network operators to make an optimal network decision.Conventional network optimization is based on a model-driven approach,which solves complex network problems by establishing a rigorous mathematical model.However,the problem of the model-driven approach lies in the difficulty in accurately modeling for a dynamic network scenario.Data-driven approach becomes an effective method for network optimization through drawing experience and knowledge from historical data.However,in a complex network circumstance,data-driven approach has difficulty in convergence,which results in reduced network performance.Therefore,it is a key issue for operators to design the optimal network decision to meet the complex spatio-temporal characteristics of services and to realize the tight coupling of traffic demand and network condition.This paper focuses on machine learning assisted traffic grooming policy in hybrid optoelectronic networks,designes and implements intelligent optical transport network management and control platform,and research work and innovations are as follows:(1)This paper proposes a collaborative data-driven and model-driven approach for traffic grooming in hybrid optoelectronic networks to overcome the issues of model-driven approach and data-driven approach.The scheme uses the traditional auxiliary graph to model the cross-layer network traffic grooming,and applies deep reinforcement learning to modify the weight of edges in the auxiliary graph,so as to improve the auxiliary graph model.Simulation results show that the proposed scheme can make better routing and wavelength assignment decisions for services than the traditional model-driven approach,and achieve the optimization of overall network performance.The scheme can also achieve good generalization ability on different datasets.(2)This paper designs the solution of machine learning assisted intelligent optical transport network management and control platform which integrates software defined network,network function virtualization,artificial intelligence and other technologies for solving the problems of network resource optimization,management and control.Based on the solution,combined with network equipment to build a management and control platform which generates real traffic flow and simulates real network scenarios.The experimental results show that the platform achieves flexible deployment of services,completes the process of intelligent traffic grooming,and improves network wavelength resources efficiency through the management and control of optoelectronic hybrid networks resources,which verifies the feasibility and effectiveness of the proposed algorithm.