Research On Network Traffic Intrusion Detection Based On Deep Learning

The rapid development of the Internet and network communication technology has brought great changes to people's lives and production.While people enjoy the convenience brought by the network,network security issues have become increasingly serious,which has brought great challenges to the normal operation of the network environment and privacy protection.Anomaly detection of network traffic has become a key method to solve network security problems.This thesis introduces deep learning into the field of intrusion detection.First,a learning algorithm framework based on deep adaptive features is proposed for traffic anomaly detection.The algorithm can determine the number of layers of the depth model and the number of neurons in each layer according to the characteristic dimension of network traffic.Secondly,the feature parameter is set reasonably in the algorithm,and the advantages of transfer learning and the powerful learning ability of the deep model for the feature are used to enable the deep neural model to further learn the original multi-dimensional data to obtain new abstract features.This thesis also combines the advantages of deep learning and traditional shallow machine learning,and uses the traditional machine learning classifier for the final intrusion detection classification.Secondly,this thesis proposes a deep asymmetric convolutional encoder for network intrusion detection.This model combines the advantages of autoencoder and convolutional autoencoder.By retaining the encoder part of the convolutional autoencoder,a deep asymmetric structure is formed by stacking multiple hidden layers.The deep convolutional asymmetric encoder can be used for feature extraction and combined with random forest for further intrusion detection classification.In this thesis,three network intrusion detection datasets include NSL-KDD,UNSWNB15 and CICIDS2017,are used for the evaluation of intrusion detection experiments.The experimental results show that the deep adaptive learning algorithm sets reasonable parameters ,which can effectively improve the detection performance of different machine learning classifiers and reduce the detection time.It has certain robustness.In addition,the deep asymmetric convolutional encoder proposed in this thesis combined with the random forest intrusion detection model was tested on the NSL-KDD and KDD99 datasets.Experimental results show that the model can effectively improve the intrusion detection performance compared with the existing model structure,and has certain stability and accuracy in multi-classification and small sample detection.