Research On Intrusion Detection Method Based On CNN-BiGRU Fusion Self-Attention Mechanism

With the increasing number and categories of network attacks in the era of big data,intrusion detection techniques are constantly updated and optimised.In response to the problems of traditional intrusion detection methods in terms of lower detection accuracy and longer detection time,an intrusion detection method based on CNN-Bi GRU fusion self-attention mechanism is proposed.Firstly,the data is balanced by the SMOTETomek integrated sampling method,which effectively alleviates the problem of highly unbalanced distribution between classes in the dataset;secondly,a CNN-Bi GRU-based network model is proposed to improve the accuracy of intrusion detection and reduce the loss of important information by calculating the importance of each feature through the self-attentiveness mechanism,and finally,the global pooling layer is output to the classifier for classification.The public dataset CIC-IDS2017 was used to conduct multiple comparison experiments in the same environment.In terms of data balancing,after SMOTETomek sampling not only can avoid attack categories with very small sample size such as Heart Bleed from being detected properly,but also the identification accuracy of Bot,Do S slowloris and FTPPatator attacks are improved by 1.88%,2.37% and 6.23% respectively,proving that the balancing process can improve the classification accuracy for a few categories.In terms of model detection performance,ablation experiments yielded an accuracy improvement of 2.83%,3.99% and 0.98%for the proposed method compared with CNN,Bi GRU and CNN-Bi GRU models without the selfattentive mechanism,respectively,demonstrating the effectiveness of the proposed method.Finally,the intrusion detection method based on CNN-Bi GRU fused with the self-attentive mechanism is verified to have better detection performance compared with other methods through comparison tests.The thesis has 32 figures,9 tables,and 54 references.