| With the continuous innovation,development and application of Internet technology,the impact of the network on people's work and life is increasingly deepened.Although it greatly enriches people's social life,it has also brought some security problems that cannot be ignored.It is not only related to the interests of individuals or enterprises,but also related to national security.Intrusion detection plays an extremely important role in maintaining network security.With the wide application of 5G,IPv6 and other technologies,the ways of network intrusion are becoming more and more complex and diverse.The traditional intrusion detection technology is no longer enough to deal with massive,complex and unbalanced intrusion data.As one of the current research hotspots,machine learning has been applied in the field of intrusion detection.Therefore,after a full understanding of intrusion detection and machine learning,this thesis proposes a network intrusion detection model based on BiLSTM and DNN,which integrates attention mechanism,in order to explore a new intrusion detection method.The main work of this thesis includes:(1)Detailed analysis was made on the background,significance and current situation of intrusion detection research.Aiming at the problem of imbalanced NSL-KDD experimental data set,improvement was proposed from two levels of data and algorithm,and the model was optimized by using hybrid sampling technology and Focal Loss function.(2)In view of the problem that the existing models lack of intrusion data features and considerations of correlation between before and after,Bidirectional Long ShortTerm Memory network is utilized to extract the relevance of features,and deep neural network is utilized to extract the characteristics of the deeper,and batch normalization and Relu activation function are used to alleviate the problem of overfitting and vanishing gradient.(3)Aiming at the problem that the model lacks consideration of the importance of features,it is proposed to introduce attention mechanism into the model to solve the problem.(4)Design and complete the comparative experiment of five categories and two categories.The experimental results show that the BiLSTM and DNN intrusion detection model proposed in this thesis,which integrates attention mechanism,has better detection effect and better generalization ability on the experimental data set NSL-KDD. |
