Research And Implementation Of Network Flow Feature Fusion System Based On Hierarchical Representation Learning

There are more and more means of network attack as the IT getting improved,and the frequency is higher and higher.With people’s attention to network security,it is particularly important to propose more effective methods to collect network traffic data,analyze and detect.At present,deep learning is mainly used in the field of vision and natural language processing,and has achieved outstanding research results,and some network security algorithms based on deep learning have also achieved preliminary results.Among them,hierarchical representation learning has the effect of better fusion of multisource network flow information and feature extraction.Therefore,the research and application of deep learning to solve the problem of target tracking is of great significance for the development of network security technology.The main research difficulty in the network flow information fusion system is to propose an algorithm to make the most of the information in the network flow.For these complex interference factors,there are still many problems to be solved.This paper mainly studies how to improve the representation ability of the model to network traffic,and how to solve the problem of network noise identification to improve the adaptability of the model to interference factors.The main research contents are as follows:(1)The network flow fusion algorithm based on hierarchical representation learning is studied and implemented.Under the machine learning,so as to solve the problems of single feature,incomplete information,local over fitting and complex embedded configuration of graph structure learning,this paper proposes a multi-source information fusion algorithm based on hierarchical representation learning network.This method defines a new network information fusion model,which preserves the spatial characteristics of network traffic as much as possible.Based on the idea of hierarchical representation learning,it gradually learns the characteristics of the whole model from coarse to fine granularity.Experiments show that the fusion method can effectively add the existing DDo S attack detection method based on machine learning.In the future work,we will continue to study the usability of the method in network early warning,situation awareness and other fields.(2)The depth residual learning tracking model based on feature fusion is studied and implemented.The model uses the deep learning network to extract the depth features of the target,uses the idea of hierarchical representation learning to construct the network layer to quickly learn the features,and calculates the target position according to the data of next to article.Three tuples are designed to store network flow information,extract features of different dimensions of the target,and train the corresponding residual network at the same time to improve the accuracy of the tracking model.The fusion strategy of response graph generated by multi-dimensional features adopts linear fusion,and sets different weight values according to the importance of features.Experiments show that the model has been improved in various indicators,the robustness of the model is better,and it has better adaptability for complex network environment.By adjusting the feature fusion strategy,the ability of the model to deal with occlusion is improved.(3)The network flow information fusion system based on the above algorithm is studied and implemented.Finally,the system provides the user’s own sampling function,uses the appropriate model to carry on the training,transforms the complex network flow data into the triple storage,and fuses the HRM model to mature its training.Users can use the software to detect the target network flow,and accurately identify the DDo S attacks,worm attacks and other network security abnormal states,which can obtain a great practical value.