摘要
基于斜率提取边缘点的时间序列分段算法在斜率波动频率剧烈时易陷入局部最优,不能保持原始时间序列的整体特征。针对该问题,提出基于一阶滤波的时间序列分段线性表示方法 PLR_SFWF。将信号处理中的滤波引入一维时间序列,通过平滑序列细微波动显现序列基本轨迹,从而捕获到能够保持序列整体特性的序列点。在此基础上通过优先队列将不同重要程度点分类存储,得到最终分段线性表示。实验结果表明,在斜率波动频率平缓时,SFWF与传统分段线性算法相比拟合误差更小;在斜率波动频率剧烈时,其分段结果比SEEP算法具有更好的全局特性。
For the time series whose slope fluctuation frequency is relatively fierce, time series piecewise algorithm with edge point extraction based on slope is easy to fall into local optimum. It cannot keep the overall features of original time series. For this problem, this paper proposes a Piecewise Linear Representation （PLR） method of time series based on first-order filtering, which is named fluotuation PLR_SFWF. It brings filtering in signal processing into unidimensional time series,revealing the elementary track of time series by smoothing slight fluctuation,so as to capture the points which keep the overall features of time series. Based on the priority queue, it classifies points with different degrees into different queues, getting the final time series PLR. Experimental results show that, for the time series whose slope fluctuation frequency is gentle, compared with other PLR algorithms, the fitting error of SFWF is smaller. For other time series whose slope fluctuation frequency is relatively fierce,compared with SEEP, SFWF has better global characteristics.
出处
《计算机工程》
CAS
CSCD
北大核心
2016年第9期151-157,共7页
Computer Engineering
关键词
时间序列
分段线性表示
滤波
平滑
优先队列
time series
Pieeewise Linear Representation （PLR）
filtering
smoothing
priority queue
