Selection of Inferential Models for Controlling Product Compositions in a Distillation Column

• Published in: KAGAKU KOGAKU RONBUNSHU Vol. 26; no. 1; pp. 94 - 99
• Main Authors: HASHIMOTO, IORI, KANO, MANABU, Showchaiya, Natthaphon, HASEBE, SHINJI
• Format: Journal Article
• Language: English; Japanese
• Published: The Society of Chemical Engineers, Japan 01.01.2000
• Subjects: distillation; inferential control; inferential model; partial least squares
• ISSN: 0386-216X; 1349-9203
• DOI: 10.1252/kakoronbunshu.26.94

For controlling product quality, which is difficult to measure on-line, an effective approach is to build an inferential control system. In the inferential control system, quality variables are estimated from measured process variables, and the estimates are controlled. Since the performance of the inferential control system depends on the inferential model, appropriate selection of the model is crucial for effective performance of the inferential control system. In this article, the performance of inferential control systems using steady-state, static, and dynamic models are compared. From the viewpoint of estimation, dynamic models are much better than both steady-state and static models. Nevertheless, the best control performance cannot be achieved by using the dynamic model. In addition, it is shown that manipulated variables must not be used as input variables of the inferential model.