KIT | KIT-Bibliothek | Impressum | Datenschutz

Time Series Comparison with Dynamic Time Warping, Convolutional Neural Network and Regression

Yu, Yuncong; Mayer, Thomas; Knoch, Eva-Maria; Frey, Michael ORCID iD icon; Gauterin, Frank

Abstract (englisch):
This paper introduces a novel method for comparison of similar time series, especially measurement and simulation data to identify problems in the observed system. It employs the technique for time series segmentation proposed in [1] together with Dynamic Time Warping (DTW) to jointly segment pairs of measurement and simulation time series. Further, a Convolutional Neural Network (CNN) is used to identify the characteristics of segments. It is trained with synthetic data generated by the time series generator presented in [1]. Finally, the essential parameters are estimated with regression. Performance evaluation of each step is conducted and shows a high accuracy. The usage of this method is not restricted to evaluation of measurement and simulation time series, but can be extended to serve the general purpose of sequence data comparison.


Zugehörige Institution(en) am KIT Institut für Fahrzeugsystemtechnik (FAST)
Publikationstyp Proceedingsbeitrag
Publikationsjahr 2019
Sprache Englisch
Identifikator ISBN: 978-3-8169-3463-9
KITopen-ID: 1000132813
Erschienen in International Conference on Calibration Methods and Automotive Data Analytics. Ed.: K. Röpke
Veranstaltung International Conference on Calibration Methods and Automotive Data Analytics (2019), Berlin, Deutschland, 21.05.2019 – 22.05.2019
Verlag expert-Verlag
Seiten 10-20
KIT – Die Forschungsuniversität in der Helmholtz-Gemeinschaft
KITopen Landing Page