Research On Key Technologies Of Network Intrusion Detection Based On Deep Learning

As an effective defense method for network security,intrusion detection technology is one of the important components in the network security system.With the rapid development of the Internet,the number of network users and the amount of network data are increasing rapidly.Network attacks are becoming more complicated and diversified,which makes traditional intrusion detection technology not well suited to the needs of development.However,deep learning shows unique advantages in solving intrusion detection problems.Therefore,this paper combines intrusion detection technology with deep learning,which uses the excellent feature dimensionality reduction and feature learning capabilities of stack sparse auto-encoder to establish a basic framework for intrusion detection models based on deep learning.To solve the problems existing in intrusion detection,such as non-adaptive extraction of features,low algorithm robustness,low detection rate of multiple classification and rare samples,the intrusion detection algorithm is improved by adopting a variety of strategies,with taking deep learning algorithm as the basic algorithm framework.The purpose is to improve the detection accuracy and the overall performance of intrusion detection.The main research work of this paper is as follows:First,for the large-scale and high-dimensional features of intrusion data,the basic model framework of intrusion detection based on deep learning is established using the advantages of stack sparse auto-encoder which has good feature dimensionality reduction and feature learning.And the method of intrusion detection data preprocessing is improved by analyzing sparse auto-encoder learning feature,activation function characteristics and the influence of parameter changes on algorithm performance.This method achieves the purpose of minimizing errors from the level of data preprocessing.The simulation results show that the proposed algorithm can obtain the optimal low-dimensional representation of the original data in an unsupervised manner,which effectively improves the detection efficiency and further improves the detection accuracy.Secondly,from the perspective of feature adaptation,for the problem that algorithm model features cannot be extracted adaptively,the differential brainstorm optimization algorithm is used to optimize the algorithm model,and a better network framework model is obtained,which realizes adaptive feature extraction.The algorithm can maximize the performance of network feature extraction and feature learning,which improves the detection accuracy of the algorithm.On this basis,in order to improve the robustness of the model,the definition of robustness and quantitative evaluation criteria were introduced.According to the definition of robustness,noise reduction parameters are introduced into the stack sparse auto-encoder.At the same time,considering the influence of noise reduction parameters on model performance,an adaptive robust network model framework is established.Finally,two kinds of DBSO-SDAE intrusion detection algorithms are proposed for binary classification,and the simulation results of the accuracy,robustness and test timeliness are analyzed.Compared with other algorithms,the proposed algorithm has better overall performance while liberating manpower consumption and reducing unnecessary computing resource overhead.In addition,for the low detection rate problem of multi-classification and rare attacks,Attack confrontation algorithms and unbalanced data processing strategies are studied.Drawing on the idea of the JSMA attack search for the most salient features,introducing salient feature analysis to further improve the learning ability of intrusion detection algorithms.On this basis,the Borderline-SMOTE algorithm is introduced to generate minority boundary samples to increase the number of rare attacks and reduce the imbalance of the data set.A hybrid intrusion detection algorithm IASFA-SDAE based on deep learning is proposed.The algorithm is suitable for multi-classification,and the necessity and effectiveness of the algorithm are verified through simulation experiments.On the basis of the proposed binary classification algorithm,the proposed multi-classification algorithm effectively improves the feature learning ability and detection accuracy of rare attacks.The algorithm of this paper improves the overall performance of intrusion detection while improving the accuracy of intrusion detection.Part of the problems existing in the intrusion detection technology are solved,which provides a feasible modeling method for the existing network intrusion detection technology and has important reference significance.