时间序列相似性度量方法  

Time Series Clustering Method Based on DTW Distance

作　　者：同方伟 郑东良[1] 陈璇 黎东 裴会锋 TONG Fang-wei;ZHENG Dong-liang;CHEN Xuan;LI Dong;PEI Hui-feng(Equipment Management and Unmanned Aerial Vehicles Engineering College,Air Force Engineering University,Xi′an 710051,China;No.63762 Unit of PLA,Weinan 714000,China;Department of Basic Sciences,Air Force Engineering University,Xi′an 710051,China;No.77120 Unit of PLA,Chengdou 610000,China)

机构地区：[1]空军工程大学装备管理与无人机工程学院,西安710051 [2]63762部队,渭南714000 [3]空军工程大学基础部,西安710051 [4]77120部队,成都610000

出　　处：《价值工程》2021年第35期189-193,共5页Value Engineering

摘　　要：本文基于时间序列的相似性度量方法提出了一种新的改进方法PAA_divided+SDTW,针对于原始的DTW度量方法时间复杂度大,计算效率低下的问题,本文采用时间序列建模提取时间序列的特征,基于这些特征进行相似性度量。本文所提出的改进方法在很大程度上降低了时间序列相似性度量的效率,相比于原始的DTW方法的精度有所提高。同时,PAA_divided方法相比于原始的PAA方法的误差更小,更精确地保留了时间序列的本质,对PAA的优化也有一定的借鉴意义。This paper is a time series classification algorithm based on DTW distance.Through similarity measurement of time series.The measurement distance is set as the classification basis.K-means is used to classify the time series in the dataset.At the same time,this paper improves the traditional DTW algorithm.Considering the PAA method used to the time series,we come up with a new method named PAA_divided algorithm to improve the precision of PAA method.As for the traditional K-means algorithm,we prove the improved algorithm have higher classification efficiency on sparse object sets.Finally,we compare the efficiency between new algorithm and traditional algorithm and prove that the efficiency of new method is higher than the traditional algorithm.

关 键 词：时间序列 动态时间弯曲 K-MEANS 逐段聚集平均 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]