Network intrusion detection technology is one of the methods to ensure network security.In the existing network intrusion detection methods,the machine learning algorithm-based method is most effective,which is,using machine learning algorithm as a classifier to classify related data.By this way,the intrusion behavior would be detected.As the classifier,the least squares support vector machine has the advantages of high detection precision and good generalization ability,and compared with the standard support vector machine,the computational efficiency is also improved to some extent due to the simplification of the problem to be solved in the algorithm.However,since its classification performance is highly influenced by its own parameter selection,how to select the appropriate parameters becomes the key to improve the detection accuracy of this method.There are mainly two categories methods for model parameter selection.One is based on the choice of artificial posterior experience.This method has strong subjectivity due to the different experience of researchers on the target problem.Thus,selected parameters in this way have a potential impact on the classification results;the other is based on the intelligent optimization algorithm to optimize the parameters of the classifier,but due to some shortcomings of the intelligent optimization algorithm,the accuracy of the parameter optimization is not good,thereby will also affect the classification accuracy of the classifier,so there are still some improvements in this method.In this paper,based on the existing theory,an improved comprehensive learning particle swarm optimization algorithm is proposed based on the opposite learning theory.Using the improved algorithm to optimizing least-squares support vector machine's parameters,an improved network intrusion detection model was proposed.Subsequently,the improved comprehensive learning particle swarm algorithm has disadvantages of a large number of iterations,and the convergence time is long.In order to solve this problem,based on existing parallelization model,an improved model was proposed.The parallel hybrid particle swarm optimization algorithm mainly consists of two parts: global population and sub-population,each particle in global population is also treat as a subpopulation and evolves independently.The standard particle swarm optimization algorithm is used in the global population,and using the improved comprehensive learning particle swarm algorithm with a fusion of opposite learning mechanism in the sub-population.Thus,an improved parallel hybrid particle swarm optimization algorithm is proposed,which can effectively reduce running time for parameter optimization.In addition,for the least squares support vector machine,it is necessary to construct multiple independent binary classifiers in dealing with multi-classification problems.Combined with the improved parallel hybrid particle swarm optimization algorithm,the network intrusion detection model is parallelized and improved,too.Both detection efficiency and accuracy have been improved,the ability of real-time network intrude detecting of this model improved as well.The experimental part of this paper is based on the KDDCUP99 dataset.Firstly,the apparently meaningless feature dimension is manually screened,and then the contribution rate of each feature is calculated by principal component analysis algorithm.The main component is selected by screening the features with cumulative contribution rate greater than 98%.Extract to achieve the purpose of feature selection.The parallel verification experiment of the model is mainly implemented by means of the Spark – The distributed computing framework.The comparison of related experiments proves that the proposed model has a good effect. |