Research On Network Security Situation Awareness Based On Selective Convolution Network

With the rapid development of network application technology and the rapid rise of network popularity,network security has gradually become the focus of attention.This not only requires the network security situation awareness technology to accurately identify network attacks and judge the current network security situation in the complex network environment,but also needs to accurately judge the development trend of the future network security situation based on historical data.The traditional network security awareness situational awareness model has a large number of parameters and a complex model.When processing network data with a large amount of data and many features,it has low timeliness,low accuracy and poor robustness.In view of the above problems,the convolutional neural network(CNN)has the advantages of weight sharing and small number of parameters to carry out research on network security situation assessment and network security situation prediction respectively.The specific work is as follows:(1)A situation assessment model based on Selective Kernel Network(SKNet)was proposed.In order to solve the problems of large amount of network security data and data imbalance and effectively extract the correlation between features,the convolutional structural unit of the selective kernel network is firstly improved and replaced by the convolutional kernel of the traditional convolutional neural network,so as to improve the model training speed and the feature quality of extraction.Then,a quadratic classification combination model is proposed,which takes the improved selective kernel network as the basic model and is divided into two models: attack recognition and attack classification.Network security data is first identified and then classified to reduce the impact of data imbalance.Finally,a situation quantification index and situation value calculation method based on the influence of network attack are proposed.Through the verification of NSL-KDD dataset and UNSW-NB15 dataset,it is shown that the proposed improved selective kernel network quadratic classification combination model has higher attack recognition accuracy,and the situation assessment results have higher fitting degree to the real value.(2)A Network security situation prediction model based on Selective Kernel Temporal Convolutional Network(SKTCN)is proposed.In order to reduce the influence of complex and changeable network security situation and the delay of prediction results,the idea of selective kernel network and temporal convolutional network(TCN)were integrated first,and two kinds of selective kernel residual blocks were built.Then selective kernel residuals are used to construct a selective kernel temporal convolutional network model.Finally,the proposed model is verified on UNSW-NB15 and AWID data sets.Experiments show that one of the selective kernel temporal convolutional network models proposed in this thesis has smaller errors and better fitting degree than other models.