Research On Intrusion Detection Algorithm Based On DBN-FOA-WELM

The scope of network security problems brought about by the rapid development of the Internet has been increasing.While people enjoy the convenience brought by the Internet,they also have to pay attention and try to solve the unpredictable harm caused by network attacks.The firewall based on packet filtering technology has long been unable to cope with various attacks appearing on the network.The intrusion detection system,as the second barrier to protect the security of computer networks,has gradually attracted the attention of scientific researchers.Through a lot of in-depth research on the existing intrusion detection algorithms and methods,it is found that the essence of the algorithm suitable for intrusion detection is to classify the network intrusion attacks.However,most of the existing studies have ignored the uneven distribution of attack types in network intrusion attacks.Therefore,this article addresses the imbalances in the types of network attacks and does the following:(1)A general introduction to the concept and classification of intrusion detection is based on a lot of literature.Then carry out a simple analysis of the four major types of attacks in the network,and then introduce the concept of an imbalanced data set,and analyze the selected experimental data set NSL from the aspects of the amount of data contained in each sample and the meaning of each dimension of data-KDD data set for the future research.(2)In-depth study of machine learning and deep learning,and a brief introduction to the parts used in this article,to pave the way for future research.(3)Aiming at the imbalance problem in network attacks,an intrusion detection algorithm based on Drosophila weighted extreme learning machine is proposed.As a feedforward neural network,weighted limit learning has the advantages of short training time and good generalization performance.For the imbalance in the NSL-KDD intrusion detection data set,the weight of a few types of attacks is increased,so that the detection rate of rare attacks in network attacks is greatly improved compared with the traditional machine learning.Using the powerful global optimization capabilities of the FOA,the input weights and bias of the hidden layer in the WELM are globally optimized to avoid the algorithm falling into the local optimal solution and realizing the classification of the NSL-KDD intrusion detection data set.(4)Deep learning has already become a hotspot in scientific research today.This paper uses the powerful feature learning capabilities of deep belief networks in deep learning algorithms to perform feature learning on each piece of data in the intrusion detection dataset NSL-KDD,and then enter the third point Classification is mentioned in the mentioned classification algorithm.Combine the advantages of effective feature extraction of deep learning algorithms and targeted classification algorithms to further improve the classification effect.(5)Under the same preconditions,a comparative experiment was designed.By comparing a large number of experimental results,the research conclusions are drawn.The comparison experiments using the data in the intrusion detection data set NSL-KDD show that the weighted limit learning machine optimized by the improved Drosophila algorithm has a certain degree of detection rate and overall classification accuracy rate for minority attacks in the data set.The increase in false alarm rate has also been reduced.On this basis,the results obtained by classifying the NSL-KDD data set through deep belief networks for feature learning have been further improved,and the detection rates for the four major categories of attacks have increased by about 20%.The report rate also dropped to 0.32%,but as the complexity of the algorithm increased,the detection time also increased slightly.