Research On Network Security Situation Awareness Based On Convolutional Neural Network

The development of modern network technology has led to a sharp increase in the amount of network data,and increased the risk of network attacks,which requires modern network security situation awareness technology to have the ability to process a large amount of data efficiently.In order to efficiently understand and predict the situation of networks with large sample size and multi-attribute characteristics,and reduce the influence of human factors on situation awareness,this thesis naturally introduces the widely used deep learning into this field,and proposes a situation awareness method based on CNN,which improves the effectiveness of situation assessment,understanding and prediction,and provides a new research idea for scholars in the field.The most important research work of this model is shown in the following four parts:(1)Build situation awareness model based on DDST-CNNIn order to maximize the network depth of the network security model,reduce the training time of the network security model and reduce the training parameters of the network security model,this thesis combines the convolution decomposition technology and the deep separability technology to build a composite structure unit.In order to make full use of the time-series relationship between data,this thesis introduces the time factor to control the data fusion degree and the number of network input channels by adjusting the timer value,so that the network model can learn the original data and fusion data at the same time,and establish an effective mapping of time series data in the construction of the model framework.Thus,based on the above structure,a situation awareness model based on DDST-CNN(Decomposition & Depth Separable & Timer-Convolution Neural Network)is built.(2)Adjust the data form and take advantage of CNNIn this thesis,the advantages of weight sharing and local perception of convolutional neural network in image processing are utilized effectively,this thesis transforms one-dimensional raw data into two-dimensional matrix,compares the data attribute eigenvalues to the image pixel values,and then fuses them into multi-dimensional data to load convolutional neural network structure,and transforms the relationship between different channels at the same location into the same attribute different time Thus,the advantages of CNN in space are effectively utilized.(3)Carry out effective model optimizationIn order to improve the effectiveness of the situation awareness model based on DDST-CNN,this thesis matches the most appropriate activation function and optimizer,and adjusts the unbalanced data of attack types.Experiments show that the above optimization method improves the accuracy of situational evaluation and prediction to a certain extent.(4)Verify the validity of DDST-CNN modelIn this thesis,KDDCup-99,AWID and UNSW?NB15 data sets are preprocessed,and the network security situation assessment and prediction are carried out with the model proposed in this thesis.In addition,KNN,random forest,LSTM,decision tree and other methods are used to do the comparative experiments of network security situation assessment and prediction,which verify the effectiveness of the proposed model.The innovative situation awareness research method based on CNN constructed by our institute has certain practical significance in modern network,and can realize efficient analysis and timely warning of network data,so as to guide network managers to make corresponding decisions.The effective application of data processing and time series has some reference value to scholars in situation awareness field.Meanwhile,the model building method of this thesis has some innovative significance in the field of deep learning.