基于冗余点压缩的趋势异常序列检测  

Abnormal Series Detection Based on Trend Analysis with Point Compression

作　　者：李斌[1] 刘瑞琴[1] 刘学军[1] 

机构地区：[1]南京工业大学电子与信息工程学院,南京210009

出　　处：《传感技术学报》2014年第3期401-408,共8页Chinese Journal of Sensors and Actuators

基　　金：国家自然科学基金项目(61073197);江苏省科技支撑计划项目(SBE201077457);国家质检公益性科研专项项目(201210022)

摘　　要：异常序列作为时间序列的一种特殊模式有着极其重要的作用,但大多数的时间序列利用基于距离的方法进行序列间的相似性度量,忽略了时间序列本身的形态特征。为此,提出了一种基于趋势对比的异常序列检测算法,利用重要点和分段线性相结合的自底向上的线性逼近方法,并以最小化两个目标函数为目的进行相邻分段融合,从而使提取的趋势特征有较高的准确度。而且为了降低提取算法的复杂度问题,对采集到的时间序列先进行道格拉斯-普克算法的冗余点删除,保持序列整体形态的同时从一定程度上减少了计算量。最后通过仿真实验,验证了所提出的检测算法的有效性,不仅提高了检测的准确率,还增强了序列趋势变化观测的直观性。As a special mode of time series,abnormal sequence plays a very important role. But most time series use the distance-based method in similarity measure,ignore the morphological feature of time series itself. Therefore,this paper proposes an abnormal series detection algorithm based on trend contrast,makes use of the bottom-up linear approximation method with combination of important point and the piecewise linearization. And to extract trend features with higher accuracy, the paper fuses the adjacent subsection on the premise of minimizing the two new optimal object function. Moreover, for the purpose of reducing the complexity of the algorithm, time series were collected after deleting redundant point by Douglas-Peucker algorithm,which keeps the whole trend feature of the sequence and to some extent reduces the amount of calculations. The results of simulation demonstrate that the detection algorithm is feasible,and it not only improves the detection accuracy,but also enhances the sequence visu-alization of the change in trend.

关 键 词：传感器网络 异常序列 离群数据 趋势特征 分段线性逼近 分段融合 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]