Developing robust hybrid-models

Peter Jul-Rasmussen, Xiaodong Liang, Xiangping Zhang, Jakob Kjøbsted Huusom

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

Abstract

With the increased availability of process data, data-driven methods are becoming more appealing to use in process modelling. Data-driven methods are especially interesting in systems where the existing first-principles models are not adequate in describing the full system dynamics. This work compares a serial semi-parametric hybrid-modelling approach, integrating Machine Learning and first-principles modelling, to a stochastic greybox modelling approach proposed by (Kristensen et al. 2004). Through a CSTR case study both modelling approaches shows accurate predicting performance and robustness towards noisy data. The stochastic greybox modelling is found to be more robust towards measurement frequency as the true model structure is provided when fitting the parameters. Through this work, it is found that good modelling performance can be achieved without providing an explicit model structure by utilizing process data.
Original languageEnglish
Title of host publicationProceedings of the 33rd European Symposium on Computer Aided Process Engineering
EditorsAntonis Kokossis, Michael C. Georgiadis, Efstratios N. Pistikopoulos
Volume52
PublisherElsevier
Publication date2023
Pages361-366
ISBN (Print)978-0-443-23553-5, 978-0-443-15274-0
DOIs
Publication statusPublished - 2023
Event33rd European Symposium on Computer Aided Process Engineering - Athens, Greece
Duration: 18 Jun 202321 Jun 2023

Conference

Conference33rd European Symposium on Computer Aided Process Engineering
Country/TerritoryGreece
CityAthens
Period18/06/202321/06/2023
SeriesComputer Aided Chemical Engineering
Volume52
ISSN1570-7946

Keywords

  • Design methods
  • Mechanistic Modelling
  • Gradient SMB

Fingerprint

Dive into the research topics of 'Developing robust hybrid-models'. Together they form a unique fingerprint.

Cite this