| Network intrusion detection identifies all kinds of attacks in the network by collecting and analyzing network information data,which is an important technology to ensure the security of network boundaries.At present,the mainstream detection methods based on feature selection,data mining and neural network rely on a large amount of data and complex computing methods,and can not effectively deal with new attacks.In order to solve the above problems,this paper proposes a detection method based on the combination of improved particle swarm optimization algorithm and SVM model,which realizes the accurate correlation between feature subset and model parameters,and only needs small sample data to accurately identify traditional and new network intrusion attacks.In view of the lack of local search ability of the standard particle swarm optimization algorithm in the later iteration,this paper uses the evolutionary operation of genetic algorithm for reference to improve the particle update strategy.The improved algorithm introduces the particle update module based on GA on the basis of the original standard particle swarm optimization algorithm.The particles in the incoming module generate new particles through crossover and mutation operations,calculate the fitness difference between the new particles and the old particles,and compare with the pre-set error to determine whether to update the particles in the module.The flow after the particle outgoing module is consistent with that of the standard particle swarm optimization algorithm.The flexible switching between local search ability and global search ability of the algorithm can be realized by dynamically changing the error value in the iterative process.Experiments on classical test functions show that the improved algorithm is better than the standard particle swarm optimization algorithm in convergence accuracy and convergence speed.In view of the unstable model performance caused by manual setting of SVM model parameters,and the model performance can not be further improved due to the lack of correlation between feature selection and model parameters,this paper combines the improved particle swarm optimization algorithm with SVM model,and uses the improved particle swarm optimization algorithm to jointly select the feature subset and SVM model parameters,at the same time,in order to avoid the high dimension of data feature attributes.Irrelevant and redundant features interfere with model training.Data preprocessing module based on label coding and data dimensionality reduction module based on PCA are introduced in the algorithm flow before model training.Experiments show that under the same conditions,the accuracy of the optimized model detection is increased by 15% on average,the error detection rate is reduced by 51% on average,the missed detection rate is reduced by 65% on average,and the detection speed is improved by about 30% on average.Based on the techniques and methods of the above research,this paper designs a complete set of experimental flow,and realizes the modules of data reading,data preprocessing,data dimensionality reduction,model training and detection,on the basis of which a series of tests are carried out.The experimental results show that the model training time is shorter,the detection rate is higher,and the false detection rate and missed detection rate are lower,which fully verify the feasibility and effectiveness of the proposed method. |
