Research On Network Traffic Intrusion Detection Technology Based On Deep Learning

In recent years,with the rapid development of the Internet industry and the continuous emergence of new technologies,Internet applications have been completely integrated into the daily work and life of various industries.Artificial intelligence,industrial control information,5G,blockchain and other technologies are widely used in various industries,but the ensuing network security problems have become increasingly prominent,including backdoor deployment,Trojan horse upload,malicious access and information tampering and other harmful events are increasing,illegal intrusion has become an important threat in the field of network security.Traffic intrusion detection,as a common means in the field of network security protection,can intercept related traffic,discover the whole process of intrusion behavior and analyze it.Therefore,to effectively improve the accuracy of intrusion detection has become a key issue that researchers continue to pay attention to.In this context,deep learning,as a powerful machine learning technology,has made great contributions in many fields and has been widely used in the field of intrusion detection.The research carried out in this thesis can be summarized as follows:1)This thesis proposes a new feature extraction method based on traffic data,aiming to achieve feature extraction of behavioral traffic by this method;The data processed by the feature extraction method in this thesis can not only be directly applied in the learning algorithm model,but also effectively remove redundant data,and further improve the efficiency of data processing model.2)Improve the execution strategy of particle swarm optimization algorithm,introduce mutation mechanism of genetic algorithm into inertia weight module,and improve the global search ability of particle swarm optimization algorithm.The inertia weight function is set as the inverse proportional function,and the search range is refined continuously with the increase of the number of iterations.3)A neural network structure combining autoencoder and extreme learning machine is proposed.The core of autoencoder is feature mapping,which can effectively remove the noise existing in the original data,and play a role in dimensionality reduction of data when mapping to low-dimensional space,and effectively improve the operation efficiency of the model.Therefore,the characteristics of autoencoder are integrated into the ELM algorithm model.4)Among many current studies,the study of ELM parameter focuses on the weight and offset value.This thesis adds two parameter variables,the number of nodes in the hidden layer and the activation function,to explore a new research direction for conventional optimization.