面向异常检测的时间序列树突状细胞算法  

Dendritic cell algorithm for time series oriented anomaly detection

作　　者：田玉玲[1] 

机构地区：[1]太原理工大学计算机科学与技术学院,山西太原030024

出　　处：《西安电子科技大学学报》2014年第4期144-150,共7页Journal of Xidian University

基　　金：国家自然科学基金重点资助项目(50335030);山西省基金资助项目(2013011018-1)

摘　　要：针对树突状细胞算法中信号及参数的定义存在高度随机性,导致检测率较低的问题,提出了一种时间序列数据的异常检测树突状细胞算法.采用多维数据流相关性分析和变化点检测方法对抗原进行检测,遴选出能够反映突变状态的关键点数据作为异常活动候选解;基于变化点子空间追踪算法提取特征集,准确地获取及分类各种输入信号子空间;在算法的上下文评估中加入动态迁移阈值的概念,累积一定窗口时间内的抗原评估,有效地减少了误判率.通过仿真实验证明该算法能够利用更少的存储空间和计算资源,有效地提高异常检测的检测率与准确率,具有更高的稳定性.Aiming at the fact that the high randomness existing in definitions of signals and the antigen results in the lower detection rate used by the Dendritic Cell Algorithm,the Dendritic Cell Algorithm for anomaly detection based on time series is proposed.The underlying methodology is to perform antigen detection via the change point detection and multiple data streams correlation analysis,and the change point data reflecting the mutation status as the candidate solution of the abnormal is selected.Features are extracted based on the subspace tracker algorithm and various input signal subspaces are obtained and classified precisely. A dynamic migration threshold is incorporated into the context evaluation of the algorithm.The accumulation of the antigen assessment in a certain window time decreases the false positive rate effectively.Simulation demonstrates that the algorithm shows a better performance on the detection rate,accuracy rate and stability with less memory space and computing resource.

关 键 词：树突状细胞算法 异常检测 时间序列 信号处理 子空间追踪 变化点检测 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]