Research On Network Traffic Analysis Based On Deep Learning

Network traffic analysis is an important means for network administrators to manage,maintain and protect the network effectively.From the analysis,they can monitor running status,helpful to optimize resource configuration and anomaly detection.This paper studies the network traffic analysis from the perspective of deep learning and focuses on improving the analysis performance by optimizing the traffic analysis modelAt present,the network traffic analysis divides into two directions:network traffic prediction and network traffic classification.Formerly,researchers took attention to one aspect.This paper start from these two directions,propose network trafic analysis model based on deep learning.The analysis model describes the structure of network traffic analysis also avail data modeling and application.In terms of network traffic prediction,this paper proposes to use the Long-Short Term Memory(LSTM)model to predict the network traffic data.This paper also make the most of network traffic data self-similar and long-term correlation,propose a neural network combined with LSTM and Artificial Neural Networks(ANN).Experimental results show that LSTM can predict the timing sequence forces model better than current methods in the reality data set collected at home and abroad.In terms of the business classification of network traffic,this paper visualizes network traffic data and visually shows that the deep learning technology that is effective in image classification can be applied to traffic data classification.Considering that the length of network flow data is variable,the convolution neural network that supports variable length input is applied to the field of network traffic analysis for the first time.In the experiment,we constructed two task scenarios of network flow business classification and anomaly detection,through deep network structure for data classification training and testing.Experimental results show that it is superior to the previous methods in usability.