Research On Network Traffic Classification Method Based On Deep Learning Model

With the growth of the Internet and the continuous emergence of various network applications,the Internet has become a necessity of human life.At the same time,there is a huge amount of network traffic data information on the Internet,and various types of network attacks and abnormal traffic have also threatened the security of cyberspace.Network traffic classification,as a research hotspot that has always been concerned in the field of network information security,is an effective means of network protection.On the one hand,massive network applications require traffic analysis,which can reasonably allocate resources and ensure the quality of network services;On the other hand,analyzing the entire network traffic information,mining the abnormal status in the network,and taking targeted measures,which will enhance the network emergency response capability in the later period,resist network illegal attack behaviors,and quickly maintain cyberspace security And other aspects are of great value and significance.As the scale of network data grows,traditional machine learning-based classification methods are difficult to meet current needs.In recent years,deep learning methods have achieved good results in the fields of computer vision and natural language processing.Based on the basic requirements of network traffic classification,this paper studies the traffic classification method mainly based on the classic model of deep learning-convolutional neural network.The main works are as follows:1.In order to overcome the shortcomings of traditional neural networks with a single structure and insufficient feature extraction,this paper proposes a network traffic classification model based on one-dimensional CNN and improved LSTM: CNN-GRU hybrid neural network model.This paper uses one-dimensional convolutional neural network(1DCNN),which is more suitable for the one-dimensional sequence characteristics of network traffic than the commonly used two-dimensional convolutional neural network.The use of GRU model can reduce training parameters,save time cost during traffic classification,improve model training speed,and optimize the effect of traffic time series feature extraction.Experimental results show that the method achieves better results on classical data sets.By comparing with the traditional CNN model and LSTM model,the hybrid neural network model in this paper has better advantages in three aspects: running time,loss rate and accuracy.2.In order to further efficiently distribute network traffic,especially to identify the difference in importance between features of short sub-sequences in long time series,this paper introduces the attention mechanism into the CNN-GRU model.By learning the important information in different parts of the salient features of the traffic,the ability to focus on more important features in the hybrid model is improved,and the effective extraction of information and the optimization of input information in the short term are better achieved.Through experimental comparison,this method has achieved good results on classic data sets.In the comparison of multiple performance indicators such as accuracy,precision,and F1 value,the experimental results of this method are higher than other hybrid neural network models,and they have higher advantages in terms of loss rate and accuracy.