Research On Network Intrusion Detection Methods Based On Deep Learning

The Internet has become an indispensable part of people’s daily life.Although Internet has brought about convenient life,the problem of network security has become increasingly prominent,which has also caused serious distress and anxiety to people.Frequent cyber security incidents in recent years have caused huge economic losses and irreparable disasters to countries,enterprises and individuals.Detection is an effective network security defense technology.It mainly collects and analyzes network data to identify intrusion behaviors.Traditional intrusion detection methods are mostly based on methods such as statistics and rule matching.With the advent of the big data era,the detection rate and efficiency of traditional algorithms are very low in the face of massive and complex intrusion data.In view of the diversity of network attacks,how to improve the performance of intrusion detection is a major problem.The research on network intrusion detection algorithms in this thesis has important theoretical significance and practical application value.The main work of this article is as follows:1)The traditional multi-layer perceptron intrusion detection model is improved.This model is designed with three network structures.By adjusting the number of neurons in the hidden layer,it matches the two subsequent multi-layer perceptron models.Verify the optimal network structure of the model;the model introduces Dropout and Batch Normalization,which significantly improves the accuracy of the model detection;for the feature dimension of the dataset is too high,the principal component analysis method is used to reduce the feature dimensions to make it consistent with the model’s data input;Build two network models to improve the convergence speed of the model.Experiments results show that the designed network model has faster convergence speed than other algorithm models,greatly improving the training accuracy rate,up to 99.98%,the test accuracy rate can reach 92%,and the model detection efficiency is also higher.2)Aiming at the problem of vanishing gradients in imbalanced data sets,an intrusion detection model based on convolutional neural networks was used.The model uses oversampling technology for the data level of the data set to reduce the amount of data,and the cross-entropy loss function is used at the algorithm level;the Adam optimizer is added for the model convergence speed to achieve the rapid gradient decline;for the overall model The dropout layer is added to the network architecture to make the model performance more stable.The experimental results show that the designed network model has a good feature learning ability.Using KDDcup99 dataset detection,the accuracy rate of intrusion detection test reaches 94%,which further improves the detection accuracy rate.Experiments on the KDDcup99 dataset show that compared with traditional intrusion detection models,the two models based on multi-layer perceptron and convolutional neural network not only improve the accuracy of different categories of classification,but also improve the detection accuracy and Detection efficiency.It also has better detection rates against U2 L and R2 L attacks.At the same time,the two models converge faster than other algorithms.