Model based control of a yeast fermentation bioreactor using optimally designed artificial neural networks

Abstract

Artificial Neural Networks (ANN) have been used for a wide variety of chemical applications because of their ability to learn system features. This paper presents the use of feedforward neural networks for dynamic modeling and temperature control of a continuous yeast fermentation bioreactor. The analytical model of this nonlinear process is also presented and it was used to generate the training data. Different ANNs were trained using the backpropagation learning algorithm. To avoid over-fitting of the data and achieve the best prediction ability with the simplest structure possible, a pruning algorithm is proposed for topology optimization of the ANN. The resulting ANNs were introduced in a Model Predictive Control scheme to test the control performance of the structure. The robustness of this control structure was studied in the case of setpoint changes and noisy temperature measurement, when the network used for prediction had been trained including noisy data in the training set. Results obtained with Linear Model Predictive Control (LMPC) as well as with proportional-integral-derivative (PID) control are also presented and compared with those obtained with the neural network model based predictive control (NNMPC) strategy. The use of inverse neural models for dynamic modeling and control of this process is also discussed and exemplified via simulations.