Adaptive database management based on the database monitoring index for long-term use of adaptive soft sensors

• Published in: Chemometrics and intelligent laboratory systems Vol. 146; pp. 179 - 185
• Main Authors: Kaneko, Hiromasa, Funatsu, Kimito
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 15.08.2015
• Subjects: Adaptive model; Database management; Model degradation; Process control; Soft sensor; Soft sensor; Model degradation; Adaptive model; Database management; Process control
• ISSN: 0169-7439; 1873-3239
• DOI: 10.1016/j.chemolab.2015.05.024

Soft sensors are an essential tool for controlling chemical and industrial plants. To adapt to new process states, these soft sensors require adaptation mechanisms. As the performance of adaptive soft sensors depends on the quality of its database and the size of which will have some upper limit, a database monitoring index (DMI) has been developed. Additional information is only added to the database if it has a sufficiently high DMI value. In this study, we propose a DMI optimization method and adaptive database management scheme. DMI hyperparameters are set automatically using an initial database, and the main database adapts to new process states by storing data with large prediction errors. Case studies using simulated and real industrial datasets confirm that the proposed scheme enables soft sensors to operate with high predictive accuracy over the long term. •Performance of adaptive soft sensors depends on the quality of the database.•A database monitoring index (DMI) has been developed for database management.•DMI can monitor the amount of information in the database.•We proposed a DMI optimization method and adaptive database management scheme.•The proposed scheme enables soft sensors to operate with high predictive accuracy over the long term.