1. Slowness or Autocorrelation? A serial correlation feature analysis method and its application in process monitoring.
- Author
-
Li, Qinghua, Zhao, Zhonggai, and Wang, Lei
- Subjects
- *
STATISTICAL correlation , *FEATURE extraction , *WHITE noise , *SUM of squares , *AUTOCORRELATION (Statistics) - Abstract
Slow feature analysis (SFA) has shown great advantages in process monitoring. However, there is no limitation imposed on the serial correlation of feature except the first-order autocorrelation, so the residual feature may not exhibit the white noise-like behaviors and may include useful information. In this paper, according to the properties of white noise, we propose a serial correlation analysis (SCA) method to improve the performance of feature extraction, where the residual feature is guaranteed to conform to the characteristics of white noise as much as possible by minimizing the sum of the squares of the multi-order autocorrelation coefficients, and the principal features are those features with a large sum of the squares. In addition, the determination of parameters is discussed, and monitoring indices are constructed for the principal feature and the residual feature to evaluate the static and dynamic status of the process operation. Finally, the proposed SCA method is applied in two numerical cases and the Tennessee Eastman (TE) process, and the results verify the effectiveness of the proposed method. • The statistical properties of residual features extracted by SFA are analyzed. • A serial correlation analysis method is proposed to fully extract data information. • SCA is used to evaluate the process steady operation status and dynamics change. • The proposed method is applied in two numerical cases and the TE process. [ABSTRACT FROM AUTHOR]
- Published
- 2023
- Full Text
- View/download PDF