閆汶朋 汪志濤 袁曉



摘 要 ???:時間序列的相似性度量是時間序列聚類、分類以及其他相關時間序列分析的基礎.傳統基于距離的相似性度量方法,忽視了時間序列可能存在的時間上的聯系,而將時間序列看作一系列孤立點的集合.對于序列間可能存在的前后聯系,基于分數階微分的遺傳特性和記憶特性,提出一種新的時間序列聚類的相似性度量.根據時間序列的分數階微分計算新序列間的點距離,將其作為聚類算法的輸入對時間序列進行聚類.仿真實驗結果表明,與基于原始序列矢量距離的聚類結果相比,新的分數階相似性度量方法表現更好.
關鍵詞 :時間序列; 聚類; 相似性度量; 分數階微分
中圖分類號 :TP391 文獻標識碼 :A DOI : ?10.19907/j.0490-6756.2023.043004
Time series similarity measurement based on ?fractional differential and its application
YAN Wen-Peng, WANG Zhi-Tao, YUAN Xiao
(College of Electronics and Information Engineering, Sichuan University, Chengdu 610065, China)
Similarity measures of time series are the basis for time series clustering, classification and other related time series analysis. The traditional distance-based similarity measure ignores the possible temporal connections of time series and treats time series as a series of isolated point sets. For the possible backward and forward connections between sequences, a new similarity measure for time series clustering is proposed based on the genetic and memory properties of fractional order differentiation. The point distances between the new sequences are calculated based on the fractional order differentiation of the time series, and then are used as the input of the clustering algorithm to cluster the time series. The simulation experimental results show that the new fractional-order similarity measure performs better compared with the clustering results based on the original distances.
Time series; Cluster; Similarity measure; Fractional differential
1 引 言
時間序列作為一種隨時間順序變化的數據序列,通常具有數據量大、維度高、無限遞增、結構復雜等特點.近年來,面對日益龐大的時間序列數據集,人工標記的成本日益增加,屬于無監督、半監督學習的時間序列聚類引起了越來越多研究者的興趣,并被廣泛應用于金融學 ?[1]、醫療診斷 ?[2,3]、工業生產控制 ?[4]和生物學 ?[5]等.聚類通過將相似的數據放入相關或同質的組中,將具有最小相似性的對象放入其他組中,已成為一種有用的數據分析方法.
對于時間序列的相似性研究,很多采用了歐幾里德距離或其演變,但基于矢量的歐式距離及其演變單純的將時間序列看做孤立點的集合,忽視了時間序列可能存……