Research And Implementation Of Intrusion Detection Method Based On Pruning Neural Network

In 2020,driven by network poverty alleviation,the scale of Internet users in China will continue to expand,with nearly 1 billion internet users;The development trend of Internet enterprise cluster has initially taken shape,which has become the main driving force of China's economic growth;In the prevention and control of the epidemic and the resumption of work and production,the Internet also embodies the powerful power of anti-epidemic empowerment and intelligence.However,the changing security situation is not optimistic.With the rapid development of Internet technology,people have higher and higher requirements for network security.In order to avoid intentional or accidental threats to systems,services and data,and ensure availability,confidentiality and integrity.In order to obtain a more secure network environment,the traditional network protection technology is widely used,including user authentication,firewall,data encryption,etc.they have a certain defensive role,but they are all passive security technology.As an active information security protection measure,intrusion detection system effectively makes up for the defects of traditional security protection technology.Using anomaly based network intrusion detection is expected to build a more secure network environment to meet the needs of network security.The main contributions of this paper are as follows:(1)The application of neural network based on pruning depth in the field of intrusion detection.Aiming at the biggest challenge of constructing NIDS based on anomaly:constructing high-performance intrusion detection classifier model,this paper applies pruning method in deep learning model compression field to intrusion detection,and proposes an intrusion detection method based on pruning deep neural network:P-DNN.Firstly,a deep neural network with complex structure and good detection performance is trained by expanding the feature dimension.Then,through pruning operation,the weight with smaller absolute value in the deep neural network is assigned to 0,which only retains the connection with more important information in the weight and reduces the complexity of the model.Finally,the link with larger absolute value of the remaining weight is heavily trained.The KDD cup 99 data set is used to evaluate the effectiveness of the method,and good experimental results are obtained.The model constructed by P-DNN achieves the detection rate of 0.9904 for known attacks and 0.1050 for unknown attacks.Compared with the related work,the best intrusion detection performance is obtained:cost is reduced to 0.1875,ACC is increased to 0.9317.(2)The relationship between the importance of the information owned by the connection and the absolute value of the weight.In this paper,through the comparative experiments of three pruning methods,it is proved that in the deep neural network under the intrusion detection environment,the connection with larger weight absolute value has more important information than the connection with smaller weight absolute value.Through pruning operation,the most suitable pruning rate in intrusion detection environment is found,which further improves the intrusion detection performance of neural network,reduces the storage resource requirements and transmission resource requirements of the model,and increases the feasibility of application in resource constrained environment.(3)Design and implementation of intrusion detection system based on pruning neural network.This paper studies and designs the system architecture and functional modules of intrusion detection system based on pruning neural network,and designs and realizes the feature acquisition module,model detection module,alarm collection module and alarm visualization module.Finally,through the system test,the intrusion detection system based on pruning neural network is successfully constructed,and 13 functional requirements of the system are realized.