Research Of Particle Swarm Optimization Algorithm In Intrusion Detection

The continuous development of Internet technology has made people's lives more convenient.Network attack technologies have affected our normal lives,such as various viruses,vulnerabilities,affecting the steady development of society.Intrusion detection is an effective network security defense technology by collecting and analyzing network information data to identify intrusion behaviors in computer network systems.For Back Propagation(BP)neural networks and Deep Belief Networks(DBN),the number of parameters is large,and it is easy to fall into local extreme values during training.which reduces the detection accuracy of intrusion detection models on large-scale intrusion data and speed.Firstly,under the background of large number of particles and high dimensionality,the aggregation characteristics of the particle swarm search process are accurately analyzed by introducing the information entropy model.The particle swarm search process is optimized in sections,and a particle swarm optimization algorithm based on entropy model(EPSO)is proposed.Then,the EPSO algorithm is combined with the BP neural network.The particles are composed of the weights and thresholds of the BP neural network.The EPSO algorithm is used to optimize the weights and thresholds.The optimal particles obtained by the optimization of the EPSO algorithm are decoded into the weights and thresholds of the BP neural network,and the weights and thresholds are further optimized locally during the training phase of the BP neural network.Thereby,it improves the classification effect of the BP neural network.Finally,aiming at the shortcomings of simple structure of BP neural network and limited ability to learn the features of intrusion detection data,a DBN model consisting of multi-layer Restricted Boltzmann Machines(RBM)and BP neural network is introduced.In the pre-training process,multi-layer RBM is used to achieve the optimal low-dimensional representation of the high-dimensional features of intrusion detection data.The weights and thresholds of the DBN model are encoded as particles,and the EPSO algorithm is used for global optimization.The optimal particles are decoded into DBN parameters,and the BP neural network is used to fine-tune the weights to achieve the purpose of optimizing the DBN model.The experimental results of particle swarm optimization algorithm show that:in the five standard test functions,EPSO algorithm has better solution accuracy and convergence speed than four classic particle swarm optimization algorithms including classic particle swarm optimization and adaptive inertia weight particle swarm optimization.And reduce a large number of invalid iterations of the algorithm.The results of EPSO-BP algorithm intrusion detection experiments show that:in terms of entropy value,fitness value and mean square error,EPSO-BP algorithm has better solution accuracy and convergence speed than PSO-BP algorithm,and iterative optimization times are more Less;At the same time,compared with PSO-SVM(Support Vector Machine),the accuracy of EPSO-BP algorithm is improved by 3.34%and 1.90%.The intrusion detection experiment results of EPSO-DBN algorithm show that:when the network structure of the DBN model is set to 122-110-80-50-30-10,the accuracy of the DBN detection model is the best.In the training stage of intrusion detection model,the EPSO-DBN model has better training error and training time than the PSO-DBN model;meanwhile,the EPSO-DBN model has a higher detection rate on the test data set.