Research On Intrusion Detection Technology Based On Deep Learning

With the rapid development of modern science and technology,the technical means used by various cyberattacks are constantly updated.Today's cyberattacks are more than traditional Trojans,viruses,and botnets.There are many new types of attacks,such as APT attacks and 0day exploits.Therefore,the traditional security defense mechanisms need to improve the defensive performance in order to cope with the current complex and versatile network security environment.Intrusion detection technology is an active security defense mechanism.The research on intrusion detection technology has become one of the main research directions in the field of network security protection.Since the rise of artificial intelligence,its application in intrusion detection technology has achieved remarkable results.However,the performance of intrusion detection models is still the focus of current research,especially in the detection of unbalanced data.In order to solve these problems,we focus on feature learning and the algorithms used for classification,and propose a new intrusion detection model based on deep learning.The main work of this paper is as follows:(1).In the field of intrusion detection,the features of original data are complex,and the data after preprocessing is redundant.To solve this problem,we propose using Deep Belief Network(DBN)to learn the features of original data.Since the model structure of deep neural network will affect the performance of the model,we conducted a lot of experiments to optimize the DBN model structure,and determined the DBN model structure that is most suitable for our research.Experiments show that the abstract features learned from the DBN model are more conducive to classification than the features selected by the PCA algorithm.(2).In this paper,RBFNN is used as the classifier of intrusion detection model.For the design of the structure and parameters of the RBFNN model is very complicated,we proposed an improved hybrid learning method to train the model,which uses an improved K-Means clustering algorithm to determine the structure of the model,and uses the supervised learning method to correct the parameters of the model.Experiments show that this method can avoid the problem of determining the optimal hyper parameter K through a large number of experiments,which greatly reduces the complexity of the training model.(3).A complete intrusion detection model is proposed based on our main work.The model mainly includes data preprocessing,DBN feature learning and RBFNN classification.The data processing part transforms the original dataset into a standard dataset.The DBN feature learning model abstracts the high-dimensional features of the standard dataset into a combined feature that is more conducive to classification,and finally RBFNN is used to classify the data.Experiments show that the detection accuracy of the model in our research is better than most of traditional classification algorithms on binary classification,and the detection performance on multi-classification is better than BP neural network,especially on the R2 L category.We propose a new intrusion detection model based on DBN and RBFNN.Experiments show that the model has better detection performance in the detection of binary classification and multi-classification problems,and the model improves the accuracy of small sample category data detection.It provides a feasible solution for the detection of unbalanced data sets.