基于AE-BNDNN模型的入侵检测方法

Network Intrusion Detection Method Based on AE-BNDNN Model

基于深度学习的网络入侵检测系统中大量的冗余数据特征会加大模型的训练时间并降低训练效果,针对此问题,提出了AE-BNDNN入侵检测模型.首先利用自编码器网络(Auto-Encoder,AE)对入侵检测数据进行特征降维,去除冗余特征,而后在深度神经网络隐藏层添加批量规范化层,作为训练入侵检测数据特征降维后的分类器,最后采用多层网格搜索算法对AEBNDNN模型参数进行自动优化,寻找模型的最优参数.在NSL-KDD数据集上的实验结果表明,采用多层网格搜索算法优化的AE-BNDNN模型取得了较高的分类准确率和训练速度.

An issue in network intrusion detection systems is the large scale of features. The redundant features will increase the training time of the deep learning model and reduce the training effect. To solve this problem,an AE-BNDNN model is proposed. Firstly,the algorithm uses the Auto-Encoder (AE) to reduce the dimensionality of the netw ork intrusion data,and to remove the redundant features. Then a batch normalization layer is added to the hidden layer of the deep neural network. DNN with BN layer is the classifier of network intrusion data with dimensions reduction. The multi-layer grid search method is used to automatically optimize the AEBNDNN model to find the optimal parameters. Experiments on the NSL-KDD dataset show that the AE-BNDNN model optimized by the multi-layer grid search method achieves higher classification accuracy and training speed.