Design And Implementation Of Network Intrusion Detection Algorithm Based On Convolutional Neural Network

In recent years,science and technology have developed rapidly,communications,big data,and cloud computing have become increasingly mature,and network technology has become popular in many aspects such as people's livelihood,economy,and politics.The Internet provides convenience for people's lives.Countless network devices,applications,and explosive growth of network information have brought huge hidden dangers to network security.In the face of massive Internet data and constantly evolving network attacks,traditional network security technologies can no longer effectively cope with the current severe network security situation.Therefore,it is of great significance to study intrusion detection technologies with active defense functions.Deep learning methods have powerful feature learning capabilities,which are helpful for feature extraction of complex data in intrusion detection.This paper analyzes intrusion detection methods and deep learning knowledge,and focuses on the research of intrusion detection methods based on convolutional neural networks.The main work of this paper is as follows:First of all,this paper explores the basic principles and analysis methods of network intrusion detection,and analyzes the two deep learning network models involved in the research.At the same time,the related evaluation indicators of network intrusion detection methods are studied.Secondly,in view of the serious imbalance in the proportion of each category of network intrusion data,which leads to the problem of low detection accuracy of minority categories,this paper designs and implements a deep hierarchical network intrusion detection method combined with hybrid sampling.This hybrid sampling method combines one-side selection and Synthetic Minority Over-Sampling Technique to construct a balanced training set.Taking into account the complexity of network data features,convolutional neural network and Bi-directional long shortterm memory are combined to build a deep hierarchical network model.Through the deep hierarchical network model,the spatial and temporal features of the data can be fully learned.Experiments show that this method improves the classification accuracy of the model.Finally,a network intrusion detection method based on group convolution snapshot ensemble is discussed in detail.Ensemble learning often achieves better results than single learner,but at the same time,it also has higher training costs.Therefore,this paper introduces the method of snapshot ensemble,and uses group convolution instead of normal convolution to build the base learner.Group convolution can reduce model parameters and complexity.Snapshot ensemble uses the cyclic cosine annealing learning rate to obtain multiple snapshot models without increasing training costs.Experimental results shows that this method can obtain a model with high generalization ability.