Research On Application Identification And Policy Control Method Based On Deep Learning

The Internet is characterized by openness,limited resources,and high target value.There are demands for real-time monitoring and analysis of business traffic data,and abnormal network traffic control.In recent years,researchers have proposed a method of network traffic identification based on machine learning,firstly,the classification performance of traditional machine learning algorithms depends on the quality of traffic features.Therefore,how to design a set of traffic features with high representation ability is a difficult problem currently to be solved.Secondly,if a deep learning model wants to have a better traffic recognition rate,it needs to consume a large amount of data sets and a more complex model architecture.Therefore,how to reduce computing costs and optimize the model structure is the direction of continuous work for researchers.Finally,as the current network traffic becomes more complex,abnormal traffic is on the rise year by year,and there is always a lack of an ideal solution for traffic security management,which can not only finely manage normal traffic,but also manage and control abnormal traffic reasonably.In view of the above problems,this paper conducts research on the above problems based on the relevant algorithms of machine learning.The main research contents of this paper are as follows:(1)Research on Encrypted Traffic Knowledge Recognition Method Based on GACNN.Aiming at the problems of high computational complexity of convolutional neural network and high structural complexity of high-performance models,this paper proposes a new convolutional neural network architecture to identify the type of encrypted traffic application,and at the same time applies genetic algorithm to the new model for automation Parameter adjustment greatly reduces the cost of manual parameter adjustment,and at the same time,the recognition rate of the model trained with the optimal parameters is better.Experimental results show that the evaluation of the method’s recognition rate,recall rate and other aspects can meet the actual production requirements.(2)Research on Malicious Encrypted Traffic Identification Method Based on GARandom Forest.Aiming at the problem that the current encrypted traffic feature representation ability is not strong,and the traditional machine learning algorithm has high parameter tuning costs,this paper first proposes a new set of fine-grained malicious encrypted traffic multi-fusion features with the help of feature engineering.This set of features expresses traffic from four perspectives.Information,a total of 45 dimensions.Secondly,the genetic algorithm is used to automatically optimize the parameters of the random forest algorithm,which greatly shortens the model training time.The test results show that the multi-fusion feature of malicious encrypted traffic proposed in this paper has good representation ability.Combined with the optimized random forest algorithm,it has good performance in aspects such as recognition rate,precision rate,recall rate,and F1.(3)Research on Optimization Method of Traffic Policy.Aiming at the lack of traffic management strategies,this paper optimizes traditional traffic policy control methods from four perspectives: traffic control,traffic management,priority policy,and business permission.After analysis and evaluation,it can be seen that the optimized scheme has good practical application value.