NEURAL-NETWORK MODELING IN OPTIMIZATION OF CONTINUOUS FERMENTATION PROCESSES

The capability of self-recurrent neural networks in dynamic modeling o f continuous fermentation is investigated in this simulation study. In the past, feedforward neural networks have been successfully used as onestep-ahead predictors. However, in steady-state optimisation of con tinuous fermentations the neural network model has to be iterated to p redict many time steps ahead into the future in order to get steady-st ate values of the variables involved in objective cost function, and t his iteration may result in increasing errors. Therefore, as an altern ative to classical feedforward neural network trained by using backpro pagation method, self-recurrent multilayer neural net trained by backp ropagation through time method was chosen in order to improve accuracy of longterm predictions. Prediction capabilities of the resulting neu ral network model is tested by implementing this into the Integrated S ystem Optimisation and Parameter Estimation (ISOPE) optimisation algor ithm. Maximisation of cellular productivity of the baker's yeast conti nuous fermentation was used as the goal of the proposed optimising con trol problem. The training and prediction results of proposed neural n etwork and performances of resulting optimisation structure are demons trated.