基于稀疏自编码深度神经网络的入侵检测方法  

An Intrusion Detection Method Based on Sparse Self-Coding Deep Neural Network

作　　者：任伟[1] REN Wei(Guangdong Branch of China Telecom Co.,Ltd.,Guangzhou 510627,China)

机构地区：[1]中国电信股份有限公司广东分公司,广东广州510627

出　　处：《移动通信》2018年第8期27-32,37,共7页Mobile Communications

摘　　要：针对现有未知网络攻击检测方法采用人工或浅层机器学习方法选取特征导致检测精度较低的问题,提出一种基于稀疏自编码深度神经网络的入侵检测模型。该模型采用多层无监督神经元将高维、非线性的数据映射成低维空间,建立高维空间和低维空间的双向映射的自编码网络结构。除此之外,通过稀疏自编码获得权值和偏置初始化深度神经网络隐含层的参数,用归一化后的网络底层连接记录数据对深度神经网络进行训练并测试。实验结果表明,本文所采用的方法能够提高入侵检测模型的准确率,优于浅层机器学习选取特征的分类算法,是一种高效且可行的入侵检测模型。In view of the low detection accuracy resulting from the manual or shallow machine learning methods to select the feature for existing unknown network attack detections, an intrusion detection model based on sparse self-coding deep neural network is proposed in this paper. First, the multi-layer unsupervised neuron is used to map the highdimensional and nonlinear data into low dimensional space. The self-coding network structure of bidirectional mapping in high dimensional space and low dimensional space is established. the weights are obtained by using sparse self-coding and the parameters of hidden layer of deep neural network are initialized by using the bias. Finally, the deep neural network is trained by the recorded data of normalized network underlying-layer connections. Experimental results show that the method adopted in this paper can improve the accuracy of intrusion detection model. As a high-efficiency and feasible intrusion detection model, it is better than the method based on shallow machine learning to select feature.

关 键 词：特征提取 稀疏自编码 深度神经网络 入侵检测 

分 类 号：TN929.5[电子电信—通信与信息系统]