Research On Network Intrusion Detection Algorithm Based On Deep Learning

In recent years,the boom of the Internet has brought great convenience to people’s production and life.Mobile payment,online shopping,cloud office and instant messaging have become an integral part of our life.At the same time,network security problems are becoming more and more common.User data leakage events occur frequently,servers are subjected to Dos attacks,and the types of network attacks are more diversified,even threatening national security.As a defense method,intrusion detection system can detect abnormal access in the network,which has attracted more and more attention.Traditional network intrusion detection mainly relies on manual feature extraction,such as feature selection through analysis.This method has poor universality and cannot deal with modern large-scale network data traffic.In addition,due to imbalanced class,the detection rate of the model for minority class is not high,which greatly affects the performance of the model.This dissertation will make use of the advantage that deep learning can automatically learn high-dimensional feature information from large-scale data to study intrusion detection algorithm.The main research contents of this dissertation are as follows:(1)An intrusion detection model based on Gated Recurrent Unit(GRU)with attention is proposed.The model combines a variety of different neural network structures to automatically extract different spatial scale features from the data,which improves the detection effect.The model is divided into three parts.The first part is one-dimensional convolutional neural network,which is used to construct feature vectors to replace the embedding layer in the traditional structure.The second part is gated neural network,which comprehensively learns the context information according to the characteristics of gated neural network,and finally uses the attention mechanism to comprehensively learn all the feature information.Experimental verification is carried out on the CSE-CIC-IDS2018 data set,and the experimental results are discussed and analyzed in detail.(2)The method based on attention mechanism solves the problem of manual feature extraction and low accuracy to some extent,but it performs poorly in the detection of minority classes of unbalanced data.Aiming at the problem of class imbalance,this dissertation proposes a deep learning model,which can effectively identify intrusions in the network.The model is divided into pareto over sample(PAOS)data imbalance processing module and deep residual convolutional neural network(DRCNN)intrusion detection module.The PAOS module uses smote oversampling algorithm to increase only a few samples,and controls the number of samples to reduce the noise introduced by oversampling data.The DRCNN module is divided into three parts.Firstly,the residual network structure is used to extract the features of standardized data,then the convolutional neural network is used to analyze and summarize the features,and finally the full connection structure is used to distinguish and classify a variety of labels.Experiments were conducted on two data sets.On the UNSW-NB15 data set,the PAOS-DRCNN model achieved an accuracy of 97.82%,a recall of 97.82%,a precision of 98.15%,a F1-Score of 97.95%.On the CICIDS2017 data set,the model achieved an accuracy of 99.86%,a recall of 99.86%,a precision of 99.88%,a F1-Score of 99.86%.The experimental results show that the PAOS-DRCNN model can effectively identify a variety of different attacks.