Citation

Hu J, Zhang S, Zeng R, Liu Z. China Saf. Sci. J. 2020; 30(9): 108-114.

Copyright

(Copyright © 2020, China Occupational Safety and Health Association, Publisher Gai Xue bao)

DOI

10.16265/j.cnki.issn.1003-3033.2020.09.016

PMID

unavailable

Abstract

In order to achieve early warning of sand plugging accidents during shale gas fracturing construction，and to reduce fracturing operation cost，an early⁃warning method based on deep learning was proposed. Firstly，based on analysis of construction parameter characteristics and data，a multi⁃variable time⁃series prediction model was put forward by using LSTM network and encoder⁃decoder architecture. Then，integrated pressure parameters and other strongly related construction parameters were utilized to explore and analyze hidden information of time series data. Finally，a comparison was made between prediction results of LSTM model and autoregressive moving average（ARIMA）model with certain shale gas fracturing data as an example. The results show that compared with traditional model，LSTM can predict change trend of fracturing operation curves more accurately with its accuracy rising by 21.75%. And its prediction time is greatly reduced compared with artificial estimation model. © 2020 China Safety Science Journal. All rights reserved.

Language: zh