Research On Method Of Intrusion Detection Based On KPCA And ELM

Nowadays the rapid development of internet takes people to be more digital to connect everything,meanwhile brings a various of network security issues.Internet intrusion may not only lead to the leakage of personal privacy and enterprise information,but also threaten national security.Traditional passive defense method for network security unable to cope with complex internet intrusion at present.As a kind of positive security defensing technique,Intrusion Detection can overcome the shortcoming of passive defense of traditional way.The essence of intrusion detection is behavior classification problem.Kernel principal component analysis(KPCA)is a nonlinear principal component analysis method,which can effectively extract nonlinear characteristics from data.Extreme learning machine(ELM)is a new algorithm for single layer feed forward neural network,which has the advantages of fast training rate and strong generalization ability,and can solve the problems of classification and regression well.To avoid the problem that high-dimensional characteristics can reduce the intrusion detection performance,the paper has put forward the feature extraction method based on KPCA using particle swarm optimization(PKPCA),at the same time to build a intrusion detection method based on PKPCA and ELM,on the basis of the study of the KDD CUP 99 data sets,PSO,KPCA and ELM.Firstly,according to the truth that experimental data features include character type and number type,we use the method of numbering to transform all the characteristics of the intrusion data into numeric characters.At the same time,aiming at the problem that the difference of the characteristic values will reduce the detection performance,we normalize the experimental data.Then we adopt KPCA feature extraction method with mixed functions on the characteristics of high-dimension experimental data and KPCA feature extraction method,to solve the problem that KPCA can not simultaneously give consideration to the strong learning ability of local kernel function and the strong generalization ability of global kernel function.Then,in order to solve the problem of selecting parameters of mixed kernel function(weight coefficient c of mixed kernel,width of gaussian radial basis kernel functions and order d of polynomial kernel function),PSO is adopted to optimize the above parameters and improve the performance of KPCA feature extraction based on mixed kernel function.Finally,the intrusion data processed by PKPCA feature extraction method are detected and classified by ELM algorithm.Simulation results show that PKPCA feature extraction method can effectively reduce the dimension of intrusion data,and the intrusion detection method based on PKPCA and ELM has good detection performance.Since the ELM algorithm randomly selects the input layer weight and the hidden layerthreshold,the selected parameters can not be guaranteed to be the optimal value,and this will reduce the detection and classification performance of ELM.To solve this problem,this study proposes the classified algorithm ELM optimized by Dynamic inertia weight particle swarm(DPSO-ELM)detection and classification algorithm.Firstly,to solve the problem that PSO algorithm is prone to fall into the local optimal solution,this paper uses the dynamic inertia weighted particle swarm optimization algorithm to improve the optimization performance of PSO algorithm.Then,aiming at the problem that the random selection of input layer weight and hidden layer threshold by ELM will reduce the classification performance of ELM detection,DPSO is used to optimize the input layer weight and hidden layer threshold of ELM to improve the detection and classification performance of ELM.Finally,DPSO optimized ELM algorithm is used to detect and classify the KPCA intrusion data after feature extraction.Simulation results show that the PKPCA-DPSO-ELM intrusion detection method has better detection performance than the comparison method.