Specific experimental configuration.

With the advent of the big data era, the threat of network security is becoming increasingly severe. In order to cope with complex network attacks and ensure network security, a network intrusion detection model is constructed relying on deep learning technology. In order to extract and analyze network intrusion features, this study uses variational auto-encoders to extract and reduce the dimensionality of the invaded network traffic, and combines the advantages of extreme gradient boosting to perform classification tasks. Finally, a network intrusion detection model for network security is constructed by combining the gated recurrent unit. The results showed the area under the curve of the research model reached 97.48% and 95.24% in the KDD99 dataset and OODS dataset, respectively. In the confusion matrix experiment, the model achieved classification accuracy greater than 0.91 for different attack traffic samples in both the training and testing sets. When the sample sizes were 10000 and 40000, the shortest time and longest feature extraction time of the model were 0.030s and 0.112s, respectively. In summary, the constructed model on the basis of improved variational auto-encoder for network security has high accuracy in network intrusion detection.

随着大数据时代的到来，网络安全威胁日益严峻。为应对复杂网络攻击、保障网络安全，本研究依托深度学习技术构建网络入侵检测模型。为提取并分析网络入侵特征，本研究采用变分自编码器（Variational Auto-Encoder）对入侵网络流量进行特征提取与降维，并结合极限梯度提升（Extreme Gradient Boosting，XGBoost）的优势完成分类任务。最终，结合门控循环单元（Gated Recurrent Unit，GRU）构建了面向网络安全的网络入侵检测模型。实验结果表明，所提模型在KDD99数据集与OODS数据集上的曲线下面积（Area Under Curve，AUC）分别达到97.48%与95.24%。在混淆矩阵实验中，无论训练集还是测试集，该模型对各类攻击流量样本的分类准确率均高于0.91。当样本量分别为10000与40000时，该模型的最短与最长特征提取耗时分别为0.030秒与0.112秒。综上，基于改进变分自编码器构建的网络安全模型在网络入侵检测任务中具备较高的检测精度。