Research On Traffic Prediction Of Wireless Cellular Network Based On Deep Learning

With the rapid development of mobile communication technology and the rapid popularization of mobile phones,wireless cellular network has played an important role in shopping,catering,transportation,entertainment and other aspects,and the demand of users for traffic will grow rapidly.According to the forecast,the growth of global wireless cellular network traffic data will exceed five times by 2024.The continuous increase of user traffic demand poses more challenges to wireless cellular network performance.How to provide stable and efficient services for users with limited resources is the key problem to be solved by mobile operators.Wireless cellular network traffic prediction can accurately analyze user needs,reasonably allocate network resources,provide users with high-quality services,and ensure the stable operation of the network.At present,there are many wireless cellular network traffic prediction methods,such as statistical methods,machine learning methods,but these methods can't simultaneously model the spatial and temporal characteristics of wireless cellular network traffic data.Therefore,in view of the shortcomings of the existing research,this paper deeply explores the characteristics of wireless cellular network traffic data,and uses deep learning technology to conduct spatiotemporal modeling of wireless cellular network traffic,so as to improve the prediction accuracy.This paper proposes two wireless cellular network traffic prediction methods based on deep learning.One is a wireless cellular network traffic prediction method based on residual network and recurrent neural network.This method mainly uses the strong spatial feature extraction ability of residual network and the temporal feature processing ability of recurrent neural network to predict the wireless cellular network traffic.Compared with the traditional convolution neural network,residual neural network can not only preserve the space feature extraction,but also avoid the problem of gradient explosion.The recurrent neural network makes full use of its memory unit,which can retain part of the relevant information when processing the sequence data,so it can extract the time correlation from the wireless cellular network traffic data.The second one is based on the first model,adding attention module to achieve wireless cellular network traffic prediction.This method further considers the capacity of the memory unit of the recurrent neural network,and uses the feature selection of attention module to further improve the prediction accuracy by means of probability and statistics.Then,the two methods are simulated on real data sets,and the results show that,compared with other basic prediction models,the wireless cellular network prediction method based on the combination of residual network,recurrent neural network and attention has higher prediction accuracy.The addition of attention module improves the accuracy and stability of prediction.Finally,based on the research results of wireless cellular network traffic prediction,a prototype cloud service system of wireless cellular network traffic prediction is designed and implemented.It verifies the feasibility of the research method in the application level.