Research On Network Intelligent Operation And Maintenance Technology Based On Deep Learning

Due to the rapid development of Internet related technology and mobile communication,the Internet is related to all aspects of human life,has a wide range of applications in various fields,and has brought great convenience to our life.However,many problems are also exposed in front of us.On the one hand,the network traffic brought by all kinds of mobile devices changes with time,showing a certain law of change.Therefore,when the network traffic load is low,we need to take a more scientific network resource allocation strategy to reduce unnecessary energy consumption,which requires us to be able to accurately predict the trend of network traffic.On the other hand,a large number of user traffic access will also lead to network security problems.The detection of abnormal network traffic is one of the important ways to maintain network security and protect the interests of the majority of Internet users.The network abnormal traffic detection method mainly analyzes all kinds of information in the traffic packet,extracts the effective features,and uses the algorithm model to distinguish the abnormal traffic,so as to apply different types of network attack coping strategies.In this paper,through deep learning technology,the network abnormal traffic detection and network traffic prediction are studied.The main contents of this paper are as follows:Aiming at the privacy problem of network traffic data and the poor overall performance of distributed algorithm,this paper proposes a network abnormal traffic detection and traffic prediction algorithm based on weighted federated learning.By evaluating the models of different clients,the fusion weight of each model is generated,and then the model parameters of each client are weighted and fused to get the final model.Compared with the federated learning algorithm,it achieves better performance on the network abnormal traffic detection data set and traffic prediction data set.Aiming at the problems of large amount of network abnormal traffic data,high labor cost of labeling,and over reliance on expert experience,this paper proposes a network abnormal traffic detection algorithm based on self supervised learning,which obtains effective features of abnormal traffic data through additional self supervised learning tasks,and then detects abnormal traffic through representation migration.The method achieved a good result in cicids2017 dataset.In order to solve the problem that the network traffic prediction does not pay enough attention to the time dimension,this paper proposes a multi-scale traffic prediction technology based on attention.Through the analysis of traffic series in different time dimensions,the appropriate training of loss function constraint model is designed,and a more accurate traffic prediction model is obtained.In the network traffic prediction data set provided by China Mobile,the results are better than the existing single point prediction methods.