Abstract:

In order to further perfect and improve the effect of pattern-based trend similarity measurement of time series, on the basis of piecewise linear representation of time series, this paper realizes the piecewise patterning of time series according to the mean of piecewise subsequence and the derivative sign of its linear fitting function, defines the pattern matching distance by the similarities and differences between patterns, proposes a Dynamic Pattern Matching（DPM） method based on the dynamic programming principle of Dynamic Time Warping（DTW）. The experimental results show that this method can accurately measure the trend similarity of equal-length time series under different compression rates, and the time consumption is low. The unequal length of time series is a manifestation of data missing. The relationship between the measurement effect of this method and the proportion of data missing deserves further study.

Key words: time series, trend similarity, pattern matching, Dynamic Time Warping（DTW）

摘要：

为了进一步改善和提高基于模式的时间序列趋势相似性度量效果，在时间序列分段线性表示的基础上，依据分段子序列的均值及其线性拟合函数的导数符号，实现时间序列的分段模式化，以模式之间的异同性定义模式匹配距离，借鉴动态时间弯曲（Dynamic Time Warping，DTW）的动态规划原理，提出一种动态模式匹配方法（Dynamic Pattern Matching，DPM）。实验结果表明，该方法能够在不同压缩率条件下，准确度量等长时间序列的趋势相似性，而且时间消耗较低。时间序列不等长作为存在数据缺失的一种表现形式，该方法的度量效果与数据缺失比例之间的关系值得进一步的深入研究。

关键词: 时间序列, 趋势相似性, 模式匹配, 动态时间弯曲（DTW）