TS-SEA：用于時間序列分類的時域-頻域-季節(jié)性聯(lián)合對比學(xué)習(xí)

打開文本圖片集

中圖分類號：TN911.7-34；TP301.6 文獻(xiàn)標(biāo)識碼：A 文章編號：1004-373X（2025）16-0038-07

TS-SEA： temporal-frequency-seasonal joint contrastive learning for timeseriesclassification

LIKun’，TANJun2，GUINing2，ZHUZhaowei3 （1.School ofSoftware，XinjiangUniversity，Urumqi 83oo91，China; 2.SchoolofComputer Scienceand Engineering，Central South University，Changsha 41oo83，China; 3.School of Computer Scienceand Technology，Zhejiang Sci-Tech University，Hangzhou 311241，China）

Abstract：Timeseriesclasification （TSC）isthetaskofcategorizing sequentialdata intopredefined clasesaccording to their temporalpaterns.Real-world timeseriesusuallcontaincomplexcouplingoftrend terms，seasonalcomponents，outliers， andnoise，anditsacurateecompositioniscrucialtoimproveclasificationpeforance.Terefore，atimesriescassiication method，S-Ssoe，ichooiisnteioalydol FFTandSTL.Basedontheseviews，iterativelearningisrealizedbymeansofcontrastlearningbetweenencoders.Theresults indicate that incomparison with existing methods，theproposedTS-SEAmethodcanexhibitthebetterperformance whendealing with diverse time series applications.

Keywords：TS-SEA;timeseriesclasification;multi-viewjointlearning;contrastivelearnng;Fouriertransform;tieseries decomposition

0 引言

時間序列數(shù)據(jù)在現(xiàn)實(shí)世界中隨處可見[1，對許多應(yīng)用至關(guān)重要，特別是在諸如醫(yī)療保健[2-3]、金融[4、交通[5-6]和工業(yè)生產(chǎn)等領(lǐng)域。（剩余11295字）