Time Series Recovery Using Adjacent Channel Data Based on LSTM: A Case Study of Subway Vibrations

Xin, Tao (author): School of Civil Engineering, Beijing Jiaotong University, Beijing 100044, China; Beijing Key Laboratory of Track Engineering, Beijing Jiaotong University, Beijing 100044, China

Yang, Yi (author): School of Civil Engineering, Beijing Jiaotong University, Beijing 100044, China

Zheng, Xiaoli (author): Beijing Municipal Institute of City Planning and Design, Beijing 100045, China

Lin, Jing (author): Mälardalens universitet,Luleå tekniska universitet,Drift, underhåll och akustik,School of Innovation, Design and Engineering, Mälardalen University, 63220 Eskilstuna, Sweden,Innovation och produktrealisering,Department of Civil, Environmental and Natural Resources Engineering, Luleå University of Technology, 97187 Luleå, Sweden,IDT

Wang, Sen (author): School of Civil Engineering, Beijing Jiaotong University, Beijing 100044, China; Frontiers Science Center for Smart High-Speed Railway System, Beijing 100044, China

Wang, Pengsong (author): School of Civil Engineering, Beijing Jiaotong University, Beijing 100044, China; Collaborative Innovation Center of Railway Traffic Safety, Beijing 100044, China

(creator_code:org_t)

Related links:: https://doi.org/10.3...; show more...; https://urn.kb.se/re...; https://doi.org/10.3...; https://urn.kb.se/re...; show less...

Abstract Subject headings

Multi-sensor technology has been widely applied in the condition monitoring of rail transit. In practice, the data of some channels in the high channel counts are often abnormal or lost due to the abnormality and damage of the sensors, thus resulting in a large amount of data waste. A method for the data recovery of lost channels by using adjacent channel data is proposed to solve this problem. Based on the LSTM network algorithm, a data recovery model is established based on the “sequence-to-sequence” regression analysis of adjacent channel data. Taking the measured vibration data of a subway as an example, the network is trained with multi-channel measured data to recover the lost channel data of time-series characteristics. The results show that this multi-channel data recovery model is feasible, and the accuracy is up to 98%. This method can also further reduce the number of channels that need to be collected.

By the author/editor: Xin, Tao; Yang, Yi; Zheng, Xiaoli; Lin, Jing; Wang, Sen; Wang, Pengsong

About the subject

ENGINEERING AND TECHNOLOGY: ENGINEERING AND ...; and Electrical Engin ...; and Signal Processin ...

Kungliga biblioteket hanterar dina personuppgifter i enlighet med EU:s dataskyddsförordning (2018), GDPR. Läs mer om hur det funkar här.
Så här hanterar KB dina uppgifter vid användning av denna tjänst.