| With the increasing maturity of Internet technology,network security issues are becoming more and more prominent.The development of some new network business technologies has made network security more challenging,and users’ requirements for network security are getting higher and higher.Some existing traditional network intrusion detection technologies still have certain defects and shortcomings in the face of rapidly changing attack methods,so new technologies need to be proposed to improve the original intrusion detection technology,so that its overall performance can be improved,so that the intrusion detection system can more comprehensively and effectively protect the security of computer networks.Artificial immunity is the current research hotspot in the field of intrusion detection,and the negation selection algorithm(NSA),as one of the classical algorithms in artificial immunity algorithm,has been widely used and studied in the field of intrusion detection because its main principle process is more similar to that of intrusion detection technology.The main research contents of this paper are as follows.Since the negative selection algorithm suffers from the problem of randomly generated distributions of detectors that are not uniformly distributed in the non-self space.While the genetic algorithm has the advantages of high coverage and low consumption,the genetic algorithm is used to replace the detector generation step of the original negative selection,and the optimized balanced subset generated by the genetic algorithm is used as the immature detector,and then the final mature detector is obtained after the autosomal tolerance step of the negative selection algorithm,and then the next detection operation is performed.Principal component analysis(PCA)is used for dimensionality reduction of the dataset during data preprocessing,and the model of this paper is tested using the NSL-KDD dataset and verified by several evaluation indexes such as accuracy,detection rate and false alarm rate,and the results show that the accuracy of the improved negative selection algorithm for intrusion detection behavior is much higher than that of the original negative selection algorithm,and the false alarm rate is also lower,and compared with various other models also It also has obvious advantages over various other models.Research on convolutional neural networks(CNN)in intrusion detection has been a hot topic in recent years,with the advantage of being able to handle high-dimensional data better.The LeNet-5 structure is a classical convolutional neural network structure,and the intrusion detection model based on LeNet-5 has good recognition and detection effects and high classification accuracy for large-scale data.The main research of this paper is as follows.To address the problem that the intrusion detection model based on LeNet-5 has a low recall rate for small sample data,an improved intrusion detection model of LeNet-5 is proposed,the network model structure is adjusted in the experiment,and the SMOTE algorithm is used to solve the problem of unbalanced data set and improve the data content of small samples for the problem that the NSL-KDD data set is unevenly classified and small samples are difficult to be detected.The validation is performed using several evaluation metrics of accuracy and recall.The experiments show that the improved model has significantly improved the classification effectiveness over the original model in intrusion detection data.In particular,there is a significant increase in accuracy and recall for a few classes of data sets,and the overall performance of the model has been improved. |
