Research On Intrusion Detection Method Based On Deep Belief Network

In recent years,with the popularization and wide application of network,network security has attracted people’s strong attention,and even has risen to the national strategy.Traditional intrusion detection technology has many problems,such as too single intrusion detection method,detection performance can not meet the actual needs,poor adaptive ability and so on.With the increasing popularity of the network,the increasing probability of network attacks and the frequent occurrence of vulnerabilities,the existing intrusion prevention methods can not very well meet the current network security needs.With the improvement of computer computing ability and the updating of machine learning technology,DBN has been widely used in computer vision,speech recognition and natural language processing because of its advantages of non-linear network structure and the inherent characteristics of extracting data.In order to reduce the emergence of current network security problems and improve its security,this paper especially integrates deep confidence network algorithm with intrusion detection.It mainly includes the following three aspects:(1)Because intrusion detection needs high adaptive ability,this paper proposes an improved intrusion detection algorithm based on deep confidence network.Firstly,the problems of traditional simulated annealing training algorithm are analyzed,and an improved simulated annealing algorithm is proposed.Then,in order to overcome the instability caused by fixed empirical learning rate,an adaptive learning rate model with variable weight is proposed.Finally,an improved intrusion detection algorithm based on deep confidence network is proposed.(2)Aiming at the intrusion behavior of high-dimensional network data and the difficulty of extracting features,an intrusion detection algorithm based on deep confidence network and deep sparse LSSVM(DBN-DLSSVM)is proposed.The algorithm uses DBN to extract intrusion detection data features,and takes the extracted features as input data of deep sparse LSSVM.The intrusion behavior and normal behavior in intrusion detection are identified by deep sparse LSSVM training.(3)In the experiment process,for ADBN algorithm,the number of RBM iterations,the number of nodes in the first hidden layer and the last hidden layer are determined by selecting different parameters to repeat the experiment.DBN algorithm and ADBN algorithm are compared,and the performance of intrusion detection algorithm is analyzed from three aspects: classification accuracy,false alarm rate and detection time;deep sparse LSSVM,PCA-LS-SVM and DBN-LSSVM are compared,and the performance of intrusion detection algorithm is analyzed from three aspects: detection rate,false alarm rate and detection time.