Research On Adaptive Dynamic Bandwidth Allocation Algorithm Based On Neural Network Prediction In 50G-PON

The rapid development of the Internet industry in today's world have become an important part of the global economic development.Optical fiber communication has a high communication speed and plays an important role in Internet communication.In fiber optic communications,the "last mile" optical access network technology,which connects users to the network center,has been extensively studied.Passive Optical Network(PON)is the most important optical access network technology due to its low cost and stable usage.In view of the point-to-multipoint structure of PON network,bandwidth multiplexing exists in the upstream bandwidth,so it is necessary to allocate the bandwidth,ensure the network packet delay and improve the network bandwidth utilization rate on the basis of avoiding transmission conflicts,so as to provide users with good network services.Based on the accurate traffic prediction of 50G-PON optical access network,this paper studies the adaptive network resource scheduling scheme for different services under different network loads.Compared with the traditional scheme,the paper studies how to guarantee the Qo S requirements of different priority services and realize the demand of improving network performance.Finally,it can achieve better performance in the parameters of average packet delay,throughput,bandwidth utilization and so on.Specific research contents include:1.Understand the technical development history of PON network,start with different PON networks customized by different organizations,and understand some technical indicators of PON link layer and physical layer;Understand the basic process of data communication in PON network,and then draw IEEE's description of 50G-PON network.Based on the understanding of PON network structure,some bandwidth allocation algorithms are studied and analyzed.2,for the prediction of network traffic,first of all,we need to understand the basic characteristics of network traffic,understand the characteristics of self-similarity and long-term correlation of network traffic,and use the ON/OFF model in OPNET network simulation software to generate network traffic data.Then,LSTM neural network is used to model and predict the network traffic,and BP neural network and RNN neural network are compared.At the same time,the activation function is studied to compare the influence of different activation functions on the prediction effect.3.Based on the characteristics of dual-wavelength composition of uplink bandwidth of50G-PON network,the multi-wavelength dynamic bandwidth allocation algorithm is studied.The advantages and disadvantages of various multi-wavelength dynamic bandwidth allocation algorithms are analyzed,and the neural network is used to predict the network traffic,so as to know the actual changes of the network in advance,and the multi-wavelength dynamic bandwidth allocation algorithm based on the prediction is proposed.Aiming at the problem that the traditional polling algorithm does not consider the network service quality,this paper divides the different services in the network into high priority services and low priority services according to the requirements of network latency,and provides different bandwidth allocation algorithms for different network services.Finally,from the network latency,packet loss rate and bandwidth utilization have achieved good results.