Research And Implementation Of Intrusion Detection Model Based On Deep Learning

The network data in the current network environment presents a larger,more complex and multi-dimensional characteristics than in the past,the traditional machine learning methods are faced with complex high-dimensional data need to manually extract a large number of features,the feature extraction process is complex and computationally large,which is not conducive to the current intrusion detection real-time and accuracy requirements.Deep learning has a good advantage in dealing with complex data,and it can automatically extract better representation features from the data.Based on this,this paper applies deep convolution neural network and deep self-coding network to intrusion detection,hoping to provide a new research direction for the research of intrusion detection.In the intrusion detection method based on deep convolution neural network,a method of converting one-dimensional data into two-dimensional "image data" is proposed firstly,and then a deep convolution neural network model is designed for the image after conversion,which uses two layers of convolution layer and pool layer to reduce the dimension of the image,The Relu function is introduced as a new nonlinear activation instead of the Sigmoid or Tanh function commonly used in traditional neural networks to speed up the convergence of the network,and a Dropout method is introduced in the model to prevent the network model from over-fitting phenomenon.Finally,the converted image is trained and identified by the completed deep convolution neural network model.In the intrusion detection method based on deep self-coding network,a deep auto-encoder network model is constructed by overlaying several auto-encoder networks,and the network feature data is used as the input of the model,which enablethe model to intelligently extract the distribution rules of network data layer by layer,thereby obtaining a new low-dimensional feature data set.Then,the low-dimensional data is classified and identified by BP algorithm.In this paper,the regularization correction is added to the auto-encoder network to prevent the trained auto-encoder network from directly copying the input information and affecting the training effect,moreover,noise is added to the input data,and the reconstruction error of the original data and output data is learned to achieve the purpose of denoising,so that the learned new feature data is more robust.By using KDD CUP99 dataset verification,it is shown-this method is very suitable for the Igh-dimensional data,which effectively reduces the training time and test time,and is very consistent with the real-time Requirements of the current network intrusion detection,and the prediction accuracy and false alarm rate of the Experime Ntal results have a better improvement than the current common intrusion detection methods.