Loading...

Model Predictive Control Based on Hybrid Neural Networks of a Fedbatch Yeast Production Fermenter

Faramarzi Babadi, Alireza | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57207 (06)
  4. University: Sharif University of Technology
  5. Department: Chemical and Petroleum Engineering
  6. Advisor(s): Pishvaie, Mahmood Reza
  7. Abstract:
  8. Due to underlying dynamic factors, many bioprocesses operate in semi-batch or batch processes without fixed points of operation. As a result, linear model-based control is not straightforward for these processes. Additionally, optimal policy design and successful control of constrained systems depend on having an appropriate dynamic model within the process scope. In bioprocessing, synthesizing and controlling such dynamic models presents numerous challenges, as most relationships are quasi-monod and involve parameter estimation from nonlinear reactor data. One approach to addressing this issue is using black box models like neural networks. However, employing purely artificial neural networks independent of physical problems and data structure limitations is not free of problems. With weak information, there's a need for large amounts of training data, long learning times, and difficulties in identifying out-of-model points, which can only be exemplified by limited cases when working with neural network structures. Recently, hybrid network architectures have emerged, where parts of fundamental laws and specific material behavior are incorporated into the framework. Within these composites, basic principles like partial mass balance appear, along with synthetic expressions visible at the structural level. Furthermore, the term "hybrid" refers to the combination of modeling based on raw data and models, emphasizing their interdependence. In this project, a semi-batch bioreactor producing bioethanol was used as a reference system, and model-predictive control based on its dynamics was employed
  9. Keywords:
  10. Fermentation ; Hybrid Modeling ; Bioethanol ; Nonlinear Predictive Control ; Fed-Batch Process

 Digital Object List

 Bookmark

No TOC