Research On Convolutional Neural Network Intrusion Detection Model Based On Networktraffic Feature Map

With the development of the Internet,the problem of network security is becoming more and more serious.How to effectively detect the network attack behavior is becoming more and more urgent.Therefore,intrusion detection technology came into being.Applying traditional machine learning methods to intrusion detection systems has been the direction of researchers.However,the machine learning methods have a strong feature dependence,which poses a new problem for researchers.In recent years,the rise of deep learning provides a new direction for network intrusion detection.Therefore,based on the research of Convolutional Neural Networks(CNN)intrusion detection system,this paper proposes a convolutional neural network intrusion detection model based on feature map algorithm.The main work and innovation of the thesis has the following four points:1.This paper expounds the concepts of deep learning and intrusion detection system and the research status at home and abroad.The advantages of the combination of deep learning and intrusion detection system compared with the traditional intrusion detection model are discussed.This paper analyzes how to solve the problems such as weak generalization ability and slow training of the existing intrusion detection system when the CNN is applied to the intrusion detection system in deep learning.2.This paper introduces the design architecture of intrusion detection system based on CNN.NSL-KDD dataset was studied and normalized and standardized.The intrusion detection model of CNN based on Feature graph was proposed,and four sets of comparative experiments were conducted including Random Forest(RF),Support Vector Machine(SVM),single-layer convolutional neural network(CNN-1),and Feature map-CNN(F-CNN)based on Feature graph.It is verified that the performance of CNN intrusion detection model is better than traditional machine learning intrusion detection model.Moreover,the feature graph-based CNN intrusion detection model proposed in this paper is superior to the single-layer CNN intrusion detection model in recognition performance.3.This paper introduces the design architecture of the convolutional neural network intrusion detection system based on feature graph.Due to the problem of sample imbalance in KDDTrain+data set,this paper proposes a data and equalization collection model to solve the problem of sample imbalance.In terms of model improvement,this paper proposes Subspace Weighting co-clustering algorithm and its improved algorithm--Feature Weight Matrix algorithm,and designs two intrusion detection models: Subspace Weighting Co-Clustering-CNN(SWCC-CNN)and Feature Weight Matrix-CNN(FWM-CNN).Compared with other improved models based on CNNnetwork,the two models proposed in this paper have better recognition performance.Compared with the model without feature mapping algorithm,the two feature mapping algorithm models proposed in this paper can significantly improve the recognition performance of attack categories.Moreover,FWM-CNN algorithm is less complex than SWCC-CNN algorithm and easier to be implemented in engineering.4.Through the functional analysis of the verification platform,an experimental verification platform based on network traffic identification is designed.Through the analysis of the difficulties in platform construction,the design method of the whole platform is elaborated in detail.In order to verify the effectiveness of the platform,kalilinux attack server was used to simulate the network attack on the target drone in the platform,and the captured attack traffic was processed and passed into the identification module for identification and verification.Finally,it was proved that the verification platform could identify the network attack.