Research On Network Intrusion Detection Technology Based On Deep Learning

At present,the forms of Internet threats are complex and diverse,the cyber attacks also show a development trend of "globalization","normalization" and "continuous escalation",and the security accidents will still break out frequently.The network security has become a common challenge facing the global community.How to improve the speed of defense action,timely early warning,detection,prevention and control of network threats,and finally carry out efficient security detection has become the focus of public concern.The intrusion detection is a key basic technology in the security field.It can maintain the security of the Internet and computer in real time by detecting intrusion behavior.The traditional intrusion detection methods have some disadvantages,such as low detection accuracy and difficult to detect unknown attacks.In order to solve the problems of low detection accuracy and difficult to detect unknown attacks,this thesis attempts to apply deep learning technology to intrusion detection to detect network attacks.The work of this thesis is mainly reflected in the following aspects:1.Aiming at the insufficient detection effect of minority types and unknown network attacks,the improved gray wolf deep belief network algorithm(BN-CNN)based on recursive feature increase is proposed.Firstly,resampling technology is applied to increase the number of minority types of data sets,deep confidence network is used to extract network intrusion features,and features are selected based on recursive feature increase method.After the DBN model is built,The BGWO algorithm is introduced to optimize the number of nodes in the hidden layer of DBN,The detection effect of a few types and unknown attacks is further improved.2.Aiming at the low detection accuracy of traditional intrusion detection methods,a deep belief network algorithm based on improved gray wolf(BGWO-DBN)is proposed.When dealing with massive data sets,the number of hidden layer nodes of DBN is globally optimized by relying on the powerful search skills of the BGWO algorithm,and the dynamic weighting strategy and convergence factor based on cosine law change are introduced in the network parameter update stage to improve the effect of intrusion detection.3.In view of the insufficient detection effect of a few types and unknown network attacks,an improved gray wolf deep belief network algorithm(RFI-BGWO-DBN)based on recursive feature addition is proposed.Firstly,the resampling technology is applied to increase the number of a few types of data sets,the deep confidence network is used to extract network intrusion features,features are selected based on recursive feature addition method,and the BGWO algorithm is introduced after the DBN model is built,Optimize the number of DBN hidden layer nodes to further improve the detection effect of a few types and unknown attacks.