Fermenter modelling using local linear models

  1. Francisco, M.
  2. Vega, P.
  3. Gil, A. B.
Proceedings:
Modelling, Identification, and Control Conference (MIC 2001) : proccedings

ISSN: 1025-8973 2293-5126

ISBN: 0-88986-316-4

Year of publication: 2001

Volume: 1

Pages: 308-313

Congress: IASTED International Conference on Modelling, Identification, and Control (MIC 2001), Innsbruck, Austria, February 19-22, 2001

Type: Conference paper

Abstract

In this work, some advanced methods have been developed to model a fermenter for baker’s yeast production. Firstly, the process and its mathematical model are described, explaining the reasons that make necessary the use of new techniques. Secondly, a state space neural network (SSNN) is used to model the process globally, giving good results for typical validation data. Then, a linear local modelling approach is presented, and two methods are considered to identify this local models. The first one allows to train a linear SSNN for every local model, and the second one allows to linearize the global model obtained with the SSNN around the points where local models are placed. Both methods give good results, but the accuracy of the second is better due to the robustness of the SSNN global model.