Research On Intrusion Detection Algorithm Based On Convolutional Neural Network

With the rapid development of the Internet,the scope of 5G,artificial intelligence,meta-universe and other intelligent technologies has gradually expanded,while stepping up the integration of industry and emerging technologies in various fields,it has also brought huge hidden dangers to network security,criminals frequently invade countless network equipment,once the invasion is successful,it may cause irreparable losses,and network security problems have become particularly important.Traditional firewall technology can no longer cope with new types of attacks in the current network environment,intrusion detection system as another barrier of network security system,how to improve the ability of intrusion detection system to identify and defend against attacks is the focus of research in the field of network security.In recent years,a variety of deep learning algorithms have been applied to intrusion detection systems,providing new ideas for intrusion detection,this thesis uses deep learning methods to study intrusion detection,the core work is as follows:(1)To address the problem of data imbalance between network traffic data categories and the problem of feature extraction due to the difficulty of traditional machine learning,resulting in inaccurate feature extraction results,this thesis proposes and implements an intrusion detection model based on ST-PCA-CNN,firstly,using Smote-Tomek hybrid sampling technique to effectively solve the data imbalance problem,so that the network traffic data is in a relatively balanced state Then,we use the principal component analysis to downscale the data,transform the one-dimensional network traffic data into two-dimensional grayscale images,and then use the advantages of convolutional neural network to process two-dimensional images,extract the features of the data by convolutional neural network,establish a convolutional neural network model based on hybrid sampling,and then use the Softmax classifier to conduct classification experiments,and finally in the NSL-KDD dataset The effectiveness of the model was verified,and the results showed that the recall rates of U2 R and R2 L for a few categories were 84.32% and 83.69%,respectively,and the overall accuracy and F1 scores of the model reached 99.15% and 99.01%,respectively.(2)To address the problem that traditional machine learning intrusion detection methods cannot effectively extract both temporal and spatial features of network traffic data,we study and propose an intrusion detection model based on multi-channel convolutional neural network with bi-directional gated cyclic unit fusion and spatio-temporal attention mechanism,using multi-channel convolutional neural network to extract the implied spatial features of network traffic data,using CA module and TSA module to realize multi-channel The model can obtain more accurate network traffic features by using multi-channel convolutional neural network to extract the spatial features implied by network traffic data,using CA module and TSA module to realize the spatio-temporal attention mechanism(STAM)with multi-channel feature enhancement and fusion,and using bi-directional gated recurrent unit to extract the temporal features of network traffic.Finally,the NSL-KDD dataset is used to verify the effectiveness of the model,and the experimental results show that the model has a binary classification accuracy of 99.56% and an F1 score of 99.39%.This method has good results in improving the accuracy of intrusion detection.Taking the intrusion detection system as the research object and combining the deep learning algorithm,this thesis proposes a new intrusion detection scheme from two directions,which improves the performance of the intrusion detection system.