Research On Network Anomaly Detection Method Based On Deep Denoising Autoencoder

With the rapid development of cloud computing,AI and other technologies,various new types of network attacks have emerged on the Internet.Network anomaly detection and analysis methods have attracted widespread attention in the academic community and gradually become a research hotspot.Network anomaly detection aims to distinguish between malicious and normal behaviors in the network.Traditional anomaly detection methods generally have the problems of low detection efficiency and high false positive rate.To improve the performance of traditional anomaly detection methods,this paper introduces deep denoising autoencoder technology,using the autoencoder's reconstruction error and autoencoder's data dimensionality reduction to study network anomaly detection methods.This paper first addresses the problem of uneven distribution of data categories in network intrusion data,and proposes a network anomaly detection method combining the elastic net's deep denoising autoencoder(E-DDAE).By improving the loss function of the deep denoising autoencoder,a deep denoising autoencoder incorporating the elastic network is constructed.The normal data in the training set is used to perform unsupervised training on the autoencoding network,and the reconstruction error of the normal data is used as the abnormal threshold To detect the abnormal behavior of the network through the constructed autoencoder and abnormal threshold.Experimental results show that compared with AE?SVM?K-NN anomaly detection methods,while this method guarantees a good classification accuracy rate,the recall rate and F1 value are improved,and the false alarm rate is reduced.This paper then addresses the problem of poor detection performance of traditional network anomaly detection methods when processing massive high-dimensional data,and proposes a hybrid anomaly detection method(DDAE-DNN)that fuses deep denoising autoencoder(DDAE)and deep neural network(DNN).Use DDAE to perform abstract feature extraction on preprocessed high-dimensional network data to obtain a low-dimensional representation of the original input data and reduce the degree of data redundancy;and then perform supervised learning on the DNN through low-dimensional network data to complete the network attack Anomaly detection.Experimental results show that,compared with DNN,SVM,and PCA-DNN methods,this method has higher detection rate and lower false alarm rate for network abnormal data.