Research On Generation Adversarial Network Intrusion Detection Based On Deep Learning

With recent advances in network technology,network intrusion incidents targeting important information infrastructure have repeatedly occurred,seriously affecting China’s national security and socioeconomic stability.Network intrusion detection is an important part of network security.It refers to the determination of network intrusion activities through the analysis of network traffic,with a view to preventing attacks,intrusions,interference,destruction and illegal use of networks,ensuring the stable and reliable operation of networks and safeguarding the completeness,confidentiality and usability.Supervised deep learning has been widely used in the field of network intrusion detection.The method requires massive labeled network data samples in the training stage.However,data labeling entails high labor costs and continuous advances in cyber attack approaches mean that labeled network data need to be frequently updated.To address the above problems,the following study is conducted:1.The semi-supervised generative adversarial network(GAN)is used in network intrusion detection in order to reduce the use of labeled data,and its objective function is optimized based on the cross entropy function,followed by proposing the network intrusion detection model based on the cross entropy semi-supervised generative adversarial network(CE-SGAN).In the CE-SGAN model,the objective function that discriminates networks allows the discriminator network to accurately identify the class of pseudo data generated by the generative network,and ensures the class of pseudo data is the same as that of the label matrix;the objective function of the generative network ensures that the pseudo data it generates include all classes of the data,thereby preventing the data of a single attack type from being used as the object of training while overlooking detection of other attack types.The experimental results show that the recall rate of the CE-SGAN model is 80.69 and the detection accuracy rate is 86.04%.2.To improve the detection capacity of the model,a semi-supervised generative adversarial network based on conditional random fields(CRF-SGAN)network intrusion detection model is proposed.Applying conditional random fields(CRF)not only helps better discriminate network data characteristics but also helps optimize and decode global parameters in a serialized manner.The experimental results show that the accuracy rate of the CRF-SGAN model is 90.17,representing an increase of 4.13% compared with that of the CE-SGAN model.