Research On Algorithm Optimization Of Convolutional Neural Network In Computer Network Intrusion Detection

Currently,the widespread use of computers and networks brings a lot of convenience to people’s daily life,and with it the number of malicious activities in the network will increase significantly.Facing various intrusion attempts by large botnets,network security has become a necessary consideration nowadays.In order to improve the detection accuracy of intrusion detection techniques,deep learning models are also widely used in intrusion detection.The main work of the article is as follows:(1)Data preprocessing and comparison experiments of three convolutional neural network models.By studying the existing deep learning-based intrusion detection models,it is found that a number of convolutional neural networks have achieved good results in intrusion detection.In this paper,VGGNet(Visual Geometry Group Network),Res Net(Residual Neural Network),and Conv Ne Xt is compared and tested,and the UNSW-NB15 dataset is preprocessed with cut,cleaning,and classification,and the three models are experimented using the UNSW-NB15 dataset,and the corresponding confusion matrix is obtained,and the evaluation index of each model is calculated.The accuracy of the Conv Ne Xt model is 80.13%.(2)An improved model SE-Conv Ne Xt-FL is proposed for the application scenario of this paper.through the comparison experiments of three convolutional neural network models we selected the Conv Ne Xt model as the base model,and based on this,an optimized model SE-Conv Ne Xt is proposed by applying the SE(Squeeze-andExcitation)module.While keeping the feature channels constant,the importance of different channels is learned.Then different weights are assigned to each channel according to the importance level to learn the features of the network data in a deeper level in order to improve the generalization ability of the model.The optimization model SE-Conv Ne Xt-FL is proposed by introducing the focal loss function on the SEConv Ne Xt model.the focal loss function can effectively reduce the detection error rate of the classified samples and make the intrusion detection more effective,and the missed detection rate,i.e.,FPR(False Positive Rate),on the UNSW-NB15 dataset is reduced by 3.35%.(3)Application of distributed parallel computing to improve the processing efficiency of the SE-Conv Ne Xt-FL model for data.the accuracy of the SE-Conv Ne XtFL model has a good improvement compared to the Conv Ne Xt model,but the number of parameters is too large for processing a large number of traffic data,and the single machine detection has a high latency.To improve the efficiency of the intrusion detection algorithm based on the SE-Conv Ne Xt-FL model by implementing distributed parallel computing to improve the efficiency of the model for data processing in response to the large number of network data and deep learning parameters.The SEConv Ne Xt-FL model algorithm is improved using the Tensor Flow On Spark framework so that both the training process and the prediction process run in parallel on the Spark cluster,reducing the running time of the model and improving the running efficiency.The article focuses on the application of deep learning in intrusion detection,and analyzes and studies the Conv Ne Xt model from the perspective of improving the detection accuracy and reducing the detection time,and proposes an improved SEConv Ne Xt-FL model and the parallel design of the algorithm based on this.Finally,simulation experiments are conducted on the UNSW-NB15 dataset,and the experimental results show that the SE-Conv Ne Xt-FL model improves the detection accuracy and efficiency compared with the traditional model.