基于重要点的时间序列趋势特征提取方法

Trend feature extraction method based on important points in time series

现有的基于时间序列分段线性表示的趋势特征提取算法不能完全正确提取时间序列基本趋势特征.为此定义时间序列基本趋势的自然分割点为重要点,提出一种新的分割目标函数.新的目标函数包括2个优化目标:最小化时间序列基本趋势与其分段基本趋势的不一致,最小化时间序列趋势值与其分段趋势值之间的误差平方和.根据新的目标函数,设计了一种重要点和自底向上分割相结合的时间序列分段线性化趋势特征提取方法.按重要点分段可以正确提取基本趋势,而自底向上分割方法以初始小尺度分割后迭代融合拟合误差平方和增加最小的相邻分段,可以得到较高的拟合精度.仿真实验表明,该方法克服了现有分段线性化方法的缺点,在分段数相同的情况下提取精度比现有方法高.

The existing algorithms to extract trend features based on time series piecewise linearization representation cannot extract completely correct basic trend features of time series.This work defined important points as the natural segment point of basic trend features of time series,and proposed a new segment objective function.The objective function includes two optimal objects,i.e., minimize the difference between the basic trend of time series and that of its corresponding segment,and minimize the sum of ...