Research On Intrusion Detection Technology Based On Deep Learning

In the information age of Internet nowadays,the scale of network is becoming larger and larger,and the information capacity is exploding.It becomes more and more important for us about the network security.Intrusion detection technology which is seen as the traditional technology of defensing attacks is the key to guarantee network security.However,in recent years,network topology is becoming more and more complex and the network data has been constantly updated and the means of intrusion have been endless.The previous intrusion detection technology can not guarantee the existing network security,so it is an important research direction to use deep learning technology to solve the problem of intrusion detection accuracy.Deep Belief Network(DBN)performs well among them,which can automatically learn abstract features,and it is conducive to classification.In order to improve the detection rate,reduce the false alarm rate further and even improve the detection rate of small sample data,this paper proposes a new intrusion detection algorithm based on deep belief network.The main research work of this paper is as follows:(1)Through the research of intrusion detection system and the technology of deep learning,the intrusion detection algorithm of shallow structure including the Support Vector Machine(Support Vector Machine,SVM)algorithm,the algorithm of neural network(Back Propagation and the three algorithms of deep structure which are automatic encoder(Autoencoder,AE),Deep Belief networks and Convolutional Neural networks(Convolutional Neural Network,CNN)are analyzed and compared respectively.It has laid a solid foundation for the proposed scheme in this paper.(2)In order to improve detection rate and reduce false alarm rate,an adaptive intrusion detection algorithm based on deep belief network has proposed.Network parameters is obtained by training the autoencoder,the network parameters are considered as initialization parameter of DBN,avoid the problem of random initialization of parameters.Change the learning rate by the error of the gradient vector through training,add momentum term to accelerate the convergence speed of the algorithm,and small batch gradient descent method is used to adjust parameters,and finally the extracted characteristics are sent to the support vector machine(SVM)classification which is optimized by particle swarm optimized algorithm.This algorithm is compared with the traditional deep belief network algorithm,SVM algorithm and BP algorithm,the algorithm proposed has good experimental results.(3)To solve the problem of unbalanced data distribution of network attack and low detection rate of small samples,an adaptive intrusion detection algorithm which improved based on technology of SMOTE has proposed.Through the improved SMOTE algorithm,the small sample data is expanded,and then the expanded data is input into the denoising automatic encoder for denoising.After the training of DBN network,adjust the network parameters through the adaptive learning rate gradually,and the classification is completed by using the SVM classifier.Experiments show that this algorithm can improve the detection rate of small samples effectively.