论文部分内容阅读
研究了入侵检测系统中海量数据分类的问题.讨论了深度信念网络(DBN)的原理,提出了基于DBN的入侵检测模型.DBN由多层无监督的限制玻尔兹曼机(RBM)网络和一层有监督的反向传播(BP)网络构成.该入侵检测模型采用一种快速、贪婪的方法对DBN网络进行预训练,利用对比分歧算法逐层训练每一个RBM网络;然后,利用有监督的BP算法对整个DBN网络进行微调,并同时对RBM网络输出的低维特征进行入侵数据分类.基于KDD CUP 1999数据集的实验结果表明,使用3层以上的DBN模型分类效果优于自组织映射和神经网络方法.因此,DBN是一种有效且适用于高维特征空间的入侵检测方法.
The problem of massive data classification in intrusion detection system is studied, the principle of deep belief network (DBN) is discussed, and intrusion detection model based on DBN is proposed.DBN consists of multi-layer unsupervised restricted Boltzmann machine (RBM) networks and A layer of supervised Back Propagation (BP) network.The intrusion detection model uses a fast and greedy method to pre-train the DBN network, and uses contrasting difference algorithm to train each RBM network layer by layer. Then, BP algorithm to fine-tune the whole DBN network.At the same time, invade data classification of low-dimensional features output by RBM network.Experimental results based on the KDD CUP 1999 dataset show that the DBN model with more than three layers is better than self-organizing map And neural network method.Therefore, DBN is an effective intrusion detection method suitable for high-dimensional feature space.