Research On Network Intrusion Detection Technology Based On Deep Learning

With the rapid development of Internet technology,the amount of network data is also increasing,and more types of network attacks have been generated.Faced with the large-scale network traffic characteristic information,traditional machine learning-based intrusion detection systems have problems such as low detection rate,poor real-time performance,and low detection rate of multi-class rare samples,while deep learning algorithms are unique in solving intrusion detection problems.Therefore,this article combines intrusion detection technology with deep learning algorithms.In response to the above problems,the paper mainly uses a combination of multiple algorithms to improve the intrusion detection algorithm based on deep learning,aiming to improve the detection rate while improving the overall performance of intrusion detection.The specific research work is as follows:(1)In view of the low detection rate of the intrusion detection model,from the perspective of the spatial and temporal characteristics of network traffic and the different importance of network traffic,a network intrusion detection algorithm based on space-time features and attention mechanism is proposed.This algorithm greatly improves the detection rate of network traffic.The simulation results show that the two-class detection rate of the algorithm on the NSL-KDD data set is 99.6%,and the five-class detection rate is 92.69%.Compared with other algorithms,this algorithm not only has a better detection rate when solving network abnormal problems,but also reduces computational resource overhead.(2)Aiming at the problem of low real-time detection in the intrusion detection model,from the point of view that network traffic is a one-dimensional sequence feature,a network intrusion detection algorithm based on one-dimensional space-time features is proposed.This algorithm uses a one-dimensional convolutional neural network and a bidirectional long-short-term memory neural network to learn the one-dimensional sequence features of the data at high speed,and then classifies them after obtaining a more comprehensive space-time feature.This algorithm fully considers the internal structure of network traffic.Simulation results show that compared with other algorithms,this algorithm not only has a better detection rate when solving network abnormal problems,but also has better real-time detection in the test phase.(3)In order to solve the problem of multi-classification and low detection rate of rare attacks in intrusion detection technology,a deep learning hybrid intrusion detection algorithm model based on multi-classification and data imbalance is proposed from the point of view of data imbalance.In this model,Borderline-SMOTE algorithm is introduced to generate minority boundary samples,so as to further improve the hybrid intrusion detection algorithm based on deep learning.The simulation results show that the five-classification detection rates of the two improved algorithms for NSL-KDD data sets are 97.12%and 98.95%,respectively.Thus,the correctness and effectiveness of the algorithm are verified.In conclusion,the research work of this paper provides a more appropriate solution to the key problems of current intrusion detection technology,which not only improves the detection rate of intrusion detection algorithm,but also improves the overall performance of intrusion detection.Finally,it provides a new modeling idea for the research of network intrusion detection,and has important reference value.