Research On Intrusion Detection Model And Algorithm Based On Artificial Neural Network

With the increasing complexity of network attacks and the frequent occurrence of network attacks,the traditional signature databases and intrusion detection methods through clustering gradually show some drawbacks,such as low detection accuracy,difficult data feature extraction,and low data processing efficiency.Based on this,this paper applies the neural network model to the field of intrusion detection,mainly from the optimization of neural network model parameters and the struction of neural network model and other aspects to conduct in-depth research,establish an efficient intrusion detection model,improve the accuracy of intrusion detection.The main work of this paper is as follows:(1)Aiming at the problem that BP neural network randomly initialized parameters in the intrusion detection process can easily cause the model to fall into local optimality,an intrusion detection model(IGWO-BP)that optimizes BP neural network by improved grey wolf algorithm is proposed.In order to improve the optimization ability of the grey wolf algorithm,we propose an improved grey wolf algorithm with a chaotic map initialization population,a nonlinear convergence factor strategy and a dynamic weight strategy to optimize the initial weights and thresholds of the BP neural network.The IGWO-BP model is applied to the network-based intrusion detection dataset.The experimental results show that the IGWO-BP model has achieved better detection results on the NSL-KDD and UNSW-NB15 datasets,and the performance is also greatly improved compared with the existing models.(2)Aiming at the potential defects of regularized extreme learning machine(RELM)due to random initialization parameters,a network intrusion detection algorithm based on beetle swarm optimization and improved regularized extreme learning machine(BSO-IRELM)is proposed.The output weight matrix of RELM is solved by the LU decomposition method,thereby shortening the training time of RELM.At the same time,the beetle swarm optimization(BSO)algorithm is designed to jointly optimize the weight and threshold of RELM.In order to avoid the BSO algorithm falling into local optimum,tent mapping reverse learning,Levy flight group learning and dynamic mutation strategy are introduced to improve the optimization performance.Finally,BSO-IRELM is applied to the network intrusion detection dataset NSLKDD.The simulation results show that BSO-IRELM algorithm has obvious advantages in each evaluation index compared with the existing models.(3)Aiming at the shortcomings of the shallow neural network model,we introduce deep learning into the field of intrusion detection,and propose an intrusion detection model based on integrated deep learning.Aiming at network intrusion detection,this paper proposes an integrated deep intrusion detection model based on SDAE-ELM on the basis of stacked denoising Auto Encoder(SDAE)in order to overcome the shortcomings of SDAE model,such as long training time and low classification accuracy,and realize timely response to intrusion behavior.For host intrusion detection,by constructing a deep learning framework of deep-level structure DBN and Softmax classifiers,an integrated deep intrusion detection model based on DBN-Softmax is designed,which improves the detection ability of host intrusion data to a certain extent.At the same time,we use the mini-batch gradient descent method to train and optimize the SDAE-ELM and DBN-Softmax models to improve the training efficiency and detection performance of the model.The experimental results on the KDD Cup99,NSL-KDD,UNSW-NB15,CIDDS-001 and ADFA-LD datasets show that SDAE-ELM and DBN-Softmax have superior intrusion detection effects compared with other machine learning models.(4)Aiming at the problems of long training time and high resource consumption of deep neural network,we introduce broad learning system(BLS)into intrusion detection and propose a residual errors sparse broad learning system(RES-BLS),to solve the intrusion detection task of model falling into local optimum and node redundancy.The model uses SVD to solve the output weight matrix,uses residual learning to adjust network errors,and enhances nodes through sparse pruning.Finally,the RES-BLS model is applied to the KDD Cup99,NSL-KDD,UNSW-NB15,and ADFA-LD datasets.The simulation results show that the RES-BLS model has better detection capabilities than the existing models.