基于深度学习特征提取的网络入侵检测方法  

Network intrusion detection method based on deep learning feature extraction

作　　者：宋勇 侯冰楠[1] 蔡志平 SONG Yong;HOU Bingnan;CAI Zhiping(College of Computer,National University of Defense Technology,Changsha 410073,China;Department of Engineering Technology,Hunan Vocational College for Nationalities,Yueyang 414000,Hunan China)

机构地区：[1]国防科技大学计算机学院,湖南长沙410073 [2]湖南民族职业学院工程技术系,湖南岳阳414000

出　　处：《华中科技大学学报（自然科学版）》2021年第2期115-120,共6页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61379145)。

摘　　要：针对在构建深度学习模型过程中,神经网络隐藏层的层数和每层神经元节点数主要通过人工专家的主观经验设置,深度学习模型智能化不够、适应性不强的问题,提出了一种应用于网络入侵检测的自适应、智能化的深度学习特征提取方法。该方法采用逐层贪婪训练的策略,通过改进稀疏自编码神经网络训练的方式,形成了一个自适应、智能化的特征提取神经网络。最后利用基于支持向量机的多类分类器,形成了一种基于深度学习特征提取的网络入侵检测系统。实验表明:与基于自编码网络的支持向量机入侵检测模型(AN-SVM)和基于核主成分分析与遗传算法相结合的支持向量机模型(KPCA-GA-SVM)入侵检测方法相比,准确率平均提高了5.01%,误报率平均降低了6.24%,检测时间平均降低了16%,说明了该方法优于其他类似方法。For the construction of deep learning model,the number of hidden layers and the number of neuron nodes in each layer of neural network were set by artificial expert’s subjective experience, the deep learning model was not intelligent and adaptable,so a kind of applied to network intrusion detection deep learning of adaptive and intelligent feature extraction method was put forward.A method that the strategy of greed training layer by layer was adopted,and an adaptive and intelligent feature extraction neural network by improving the training method of sparse self-coding neural network was formed.In the end,a network intrusion detection system based on deep learning feature extraction was developed by using a multi-class classifier based on support vector machine.The experiment shows that proposed method compared with support vector machine based on autoencoder network(ANSVM) and support vector machine model combining kernel principal component analysis with genetic algorithm(KPA-GA-SVM)methods,the average accuracy is increased by 5.01%,the false alarm rate is reduced by 6.24%,and the detection time is reduced by16%,it shows that this method is superior to other similar methods.

关 键 词：深度学习 稀疏自编码 抑制与激励 特征提取 逐层贪婪训练 支持向量机 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]