Research On Intrusion Detection Method Based On Self-attention And Meta-learning

With the rapid development of Internet and industry technology,how to ensure the security of Internet access equipment has become the focus of attention.In the increasingly complex network environment,intrusion detection system must be able to effectively distinguish which network traffic is normal and which is abnormal.At present,the research on these problems mainly focuses on how to use different methods to effectively identify and classify malicious traffic,which brings new problems to the related research of intrusion detection.Although many supervised or unsupervised machine learning methods have been applied to the field of intrusion detection,there are still many shortcomings.This study analyzes the problems of feature selection and few-shot class recognition of network traffic,and proposes the following design scheme of intrusion detection system:(1)In this paper,the Transformer model construction method is used to associate the sequence information between features by using positional encoding technology,and then the variant stacked encoder-decoder neural network is used to learn low-dimensional feature representations from high-dimensional raw data.In addition,the self-attention mechanism is used to realize the classification of network traffic types.Extensive experiments show that compared with other intrusion detection algorithms,the proposed framework takes into account data dimension reduction and feature selection and retention,and the accuracy of traffic classification has been significantly improved,and better performance has been achieved.(2)In order to solve the problem that the samples are difficult to obtain and the number of abnormal samples is insufficient in network detection,this study also applies the idea of meta-learning algorithm to the intrusion detection system.Specifically,the Transformer model based on self-attention mechanism and the meta-learning model are combined to construct a novel neural network called FCC-Net.Subsequently,the detection effect of the model is verified on the NSL-KDD and UNSW-NB15 datasets widely used in the field of intrusion detection.Experimental results show that compared with other algorithm models,the improved Transformer algorithm model based on meta-learning uses the least data,but the detection effect exceeds the detection accuracy and detection rate of other algorithms,and the running time of the model is also shortened.