基于图正则化概念分解的网络入侵检测研究

Research on the Intrusion Detection of Network Based on the Graph Regularized Conception Factorization

作为一种有效的主动探测网络恶意攻击防护措施,入侵检测在变电站信息系统安全防护中得到了广泛的应用.但实际网络入侵数据类型的多样性、非负性和高维度性等特点使得现有方法存在检测率低、误报率高等不足.基于非负矩阵分解的方法在入侵检测上取得了较好的效果,却忽略了嵌入在数据局部的几何结构和标记信息.为此,本文提出一种基于图正则化约束的概念分解算法.通过将数据的几何结构和标记信息同时作为约束条件,建立了一种新的概念分解模型,并提出了迭代更新求解算法.通过在网络入侵数据集KDD99上的实验验证,其结果展示了所提算法的有效性和鲁棒性.

As an effective protection way for the intensive network attack,intrusion detection has a wide application in the protection of the substation information system.It is not effective and results in low detection accuracy for the existing methods to tackle the intrusion data because of the diversity,non-negative and high dimension.Although non-negative matrix factorization methods perform efficiently on intrusion detection,they neglect the local geometric structure and the label information.In this paper,we propose a conception factorization method based on the graph regularization.Considering the local geometric structure and label information simultaneously,we propose a novel conception factorization model and develop a new iterative update algorithm.The experimental results on the KKDD9 data set demonstrate the effectiveness and efficiency.