Research On Deep Learning Intrusion Detection Based On Feature Fusion

With the continuous development of the Internet,network intrusion behaviors have become more diverse,and the corresponding network security technologies need to be constantly updated and improved.Intrusion detection system is the only network security system that runs all the time,its importance is self-evident.In the existing intrusion detection,the port-based detection method is not very accurate,and the DPI-based detection method has high complexity and cannot handle encrypted traffic detection.Therefore,the intrusion detection system based on machine learning is the current mainstream research direction,but it is often limited by manual extraction of features when dealing with huge amounts of data,resulting in a low detection rate.In view of the above problems,this paper,on the basis of studying the intrusion detection system of machine learning,can automatically extract features according to deep learning to improve the characteristics of model training time,introduce the deep learning algorithm into the intrusion detection system,and propose an improved structured volume The integrated neural network MSCNN,the Bi-directional Long Short-Term Memory(Bi LSTM)and the attention mechanism are several new model algorithms that are integrated with each other.detection rate.Build an intrusion detection model based on machine learning.The main workflow is data processing,training,classification detection and performance evaluation.According to the comparison between the KDDCUP99 data set and the NSL-KDD data set,combined with the intrusion detection system and data classification,the intrusion detection benchmark data set NSL-KDD is used as the experimental intrusion data.At the same time,five machine learning algorithms including support vector machine,linear discriminant analysis,K-proximity algorithm,quantum support vector machine and quantum linear discriminant analysis are selected to construct an intrusion detection model based on machine learning and apply it to the data classification and detection of intrusion detection system.Five machine learning algorithms are trained based on the same data set,the data is preprocessed so that the machine learning algorithm can identify it,and then appropriate parameters are selected for each machine learning algorithm to complete the classification and detection of intrusion data for each algorithm.,to evaluate the data classification and detection performance of these machine learning algorithms.In order to further improve the detection rate,better perform deep learning intrusion detection modeling analysis and build a deep learning algorithm model,this paper combines the data classification and detection performance of five machine learning algorithms in intrusion detection systems,and proposes a deep learning based feature fusion.Intrusion detection model,and by using various algorithm models on the test set for data two-classification and five-classification detection,to detect the actual effect.The experimental results show that the optimal model is the new algorithm model MSCNN-Attention,which is a fusion of the attention mechanism and the convolutional neural network after changing the structure.The detection accuracy rates of the two-class and five-class data are99.7280% and 99.3740%,respectively.Compared with other model algorithms,this algorithm model greatly improves the detection rate of data classification in the intrusion detection system,improves the algorithm model in the intrusion detection system,and verifies the effectiveness of the MSCNN-Attention algorithm model.