Data for: Dynamic Reconstruction Based Representation Learning for Multivariable Process Monitoring

Published: 19 August 2019| Version 1 | DOI: 10.17632/cs8gf83tn2.1
Chenglin Wen, Meiqin Liu, Feiya Lv


As a public benchmark of chemical industrial process, TEP is well suited for multivariable control problems. This simulation data can be downloaded from the website: The 41 measured variables, with the 22 continuous measurements are sampled with the sampling interval of 3mins, while the 19 composition variables are generated at time delays that vary from 6 to 15mins. 21 types of identified faults are inside. Each faulty state consists of 480 samples that are used as training dataset, and 960 samples are used as testing dataset with faults induced after 8 hours, which corresponds to 160 samples.