Research Of Intrusion Detection Method Based On DBN And ELM

With the rapid development of the Internet,mankind has entered an era of intelligence.The network connection makes people and people,people and things,and things have a closer relationship,but at the same time,it also causes many network security problems.As the network is hacked,personal privacy may be violated,important corporate information may be leaked,and more seriously,it may pose a threat to national security.Traditional network defense methods can only passively perform security defense against intrusion,but they are powerless to deal with the current complex and diverse network security issues.Intrusion detection technology is a technology that actively protects current network security.This technology improves on the passive defense shortcomings of traditional network security protection technology.In the research of intrusion detection method,the essence is to transform it into the problem of classifying intrusion behavior.The main work of this article is as follows:1)The thesis applies Deep Belief Network(DBN)optimized by ALO(Ant Lion Optimizer)algorithm to feature extraction of intrusion detection,a DBN optimized by ALO algorithm,using data sets for experiments,and obtaining This method can effectively improve the effect of system feature extraction.2)The paper proposes an improved deep learning algorithm combining DBN and Extreme Learning Machine(ELM).Each set of neurons is voted by using the set idea to determine the final decision result.The deep confidence network algorithm is a common deep learning model and has good performance.However,the training of this algorithm is divided into two parts,first pre-training and then fine-tuning.These two parts require a relatively long time.The ELM algorithm is a single hidden layer neural network,which has the advantages of fast model training and excellent generalization performance.However,studies have shown that for ELM algorithm to obtain better prediction results,it is usually necessary to increase the number of neurons in the hidden layer,which leads to the training of the model and the required memory will also become larger.Through the study of the two algorithms,due to the insufficient utilization of the middle layer information of the algorithm model,an improved-DBN-ELM algorithm that draws on the idea of sets is proposed,and DBN,ELM and DBN-ELM,improved-DBN-ELM algorithm are respectively Simulation application in the intrusion detection data set KDDCUP99,the results show that the improved-DBN-ELM algorithm can effectively improve the accuracy of intrusion detection.This paper applies the optimized Deep Belief Network(DBN)to the feature extraction of intrusion detection,proposes an improved DBN optimized by classification and partitioning,and applies it to the data set for experiments.This method can effectively prevent the system from overfitting and learning local optimal solutions.