Research On Real-time Traffic Prediction Of IP Backbone Networks Based On Light-weighted Deep Learning Framework

With the advancement of information technology,the scale of IP backbone networks has expanded dramatically,and the network structure has become more and more complicated,thus network traffic prediction has become an important research topic.Due to the expansion of the network scale and the rapid growth of users in modern networks,network congestion or overload occurs from time to time,so real-time prediction of network traffic is of great significance for the efficient operation of the network,network resource allocation,and network planning.In recent years,the importance of network traffic prediction in practical applications has received increasing attention in the field of artificial intelligence research.The long-term and short-term memory(LSTM)neural network has become the one of methods of mainstream network traffic prediction,due to its ability to handle long-and short-range dependencies between time series.When applying the LSTM model to real-time network traffic prediction,while there are problems such as too large model parameters and high computational costs,which can easily lead to over-fitting of the model.In the task of network traffic prediction with strict waiting time,this is intolerable.On the other hand,the existing LSTM model is modeled only from the correlation characteristics of network traffic,ignoring the spatial correlation of router nodes in the network structure.Therefore,there are still some challenges in modeling network traffic in real time and accurately.This paper designs two different light-weighted IP backbone network traffic real-time prediction neural network algorithms based on deep learning technology.The main research contents and innovations are as follows:Aiming at the problems that the LSTM model is applied to real-time network traffic prediction,the model parameters are too more,the calculation cost is too high,and it is easy to cause the model to overfit.In view of the time correlation of network traffic,this method introduces a recurrent neural network based on the minimal gated unit to construct a light-weighted deep learning prediction model.The experimental results show that compared with the traditional LSTM traffic prediction method,the method proposed can obtains traffic prediction performance with less model training time in this paper that is quite even slightly better than the LSTM model,and is also superior to existing methods in terms of traffic prediction accuracy and real-time performance.For the above model,only the time correlation of network traffic is considered,and its spatial correlation is ignored.Therefore,in combination with the spatial correlation of router nodes in the network topology,this paper proposes a real-time prediction of IP backbone network traffic based on the space-time correlation STGM model.Simultaneously utilize the characteristics of time correlation and spatial correlation of network traffic,compared with the method that only considers the time correlation of network traffic,the experimental results show that the proposed prediction method has a simplified model and a certain improvement in real-time prediction performance.Finally,based on the analysis of application requirements for real-time IP traffic forecasting,this paper designs and implements a prototype system for real-time IP traffic forecasting based on the methods proposed above,which implements real-time traffic forecasting and warning functions.Through this system,users can make real-time prediction of network traffic,in order to ensure the real-time and accuracy of the predicting,and this system realize the management of network traffic warning through configuration.