基于流形学习的网络数据流异常检测

发布时间：2018-06-10 19:41

本文选题：网络数据流 + 异常检测　；参考：《济南大学学报(自然科学版)》2017年02期

【摘要】：将非线性流形学习应用于网络数据流的降维过程,基于局部保持投影(LPP)算法基本思想,提出基于类别信息的监督判别LPP(SDLPP)算法;与传统线性降维算法和传统流形学习算法的结果进行对比,以验证算法的准确性与稳定性;建立基于SDLPP算法的网络数据流异常检测系统实施模型。结果表明:SDLPP算法通过多目标优化,在保证局部保持投影同时实现类间距离最大与类内距离最小,在挖掘低维特征空间嵌入的同时提高了分类效果;非线性的流形学习算法能有效挖掘高维数据中的低维流形,保证了维数约减过程中的非线性结构;SDLPP算法能够生成显式投影映射,泛化性较好,时间复杂度低,更加适合网络数据流实时监测系统,并可应用于实际的网络数据流入侵检测模型。
[Abstract]:The nonlinear manifold learning is applied to the dimensionality reduction process of network data flow. Based on the basic idea of local maintenance projection (LPP) algorithm, a supervised discriminant LPP (SDLPP) algorithm based on category information is proposed. The algorithm is compared with the traditional linear dimension reduction algorithm and the traditional manifold learning algorithm to verify the accuracy and stability of the algorithm, and a SDL based on the algorithm is established. The model of network data flow anomaly detection system of PP algorithm is implemented. The results show that the SDLPP algorithm improves the classification effect at the same time, and improves the classification effect while mining the low dimension feature space. The nonlinear manifold learning algorithm can effectively excavate the high dimension through the multi-objective optimization. The low dimensional manifolds in the data guarantee the nonlinear structure of the dimensionality reduction process. The SDLPP algorithm can generate explicit projection mapping, good generalization, low time complexity, and more suitable for network data flow real-time monitoring system, and can be applied to the actual network data flow intrusion detection model.
【作者单位】：中国人民解放军理工大学指挥信息系统学院;
【基金】：国家自然科学基金项目(61473149)
【分类号】：TP393.08;TP181

【参考文献】