Multimodal modelling of uneven batch data

Atli Freyr Magnusson*, Jari Pajander, Gürkan Sin, Stuart M. Stocks

*Corresponding author for this work

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


This work explores the application of a novel tri-linear regression methodology known as Shifted Covariates REgression Analysis for Multi-way data (SCREAM) to predict the quality of a fed-batch process. The SCREAM model shows promise as it is the only known multilinear regression tool that can directly handle three-way data arrays of different lengths. Thus, it provides an alternative modelling tool that does not require complicated time warping methods as a preprocessing step. The model was tested on a simulated fed-batch dataset based on industrial simulation of penicillin production. Variations were intentionally included in the simulations to create uneven data arrays. The SCREAM model outperforms traditional staples of multivariate models like NPLS and UPLS when warping is not considered and thus shows promise for application in fed-batch processes.
Original languageEnglish
Title of host publicationProceedings of the 14th International Symposium on Process Systems Engineering
EditorsYoshiyuki Yamashita, Manabu Kano
Place of PublicationAmsterdam
Publication date2022
ISBN (Electronic)978-0-443-18726-1, 978-0-323-85159-6
Publication statusPublished - 2022
Event14th International Symposium on Process Systems Engineering (PSE 2021+) - Kyoto, Japan
Duration: 19 Jun 202223 Jun 2022


Conference14th International Symposium on Process Systems Engineering (PSE 2021+)
SeriesComputer Aided Chemical Engineering


  • Fed-Batch
  • Multimodal Modelling
  • PLS
  • Multivariate analysis


Dive into the research topics of 'Multimodal modelling of uneven batch data'. Together they form a unique fingerprint.

Cite this