Bioprocess control using a recurrent neural network model

The paper deals with the identification and the control of a continuous biotechnological process using dynamic neural networks. The process considered in the paper is the growth process of Candida lipolytica population on an ammonium sulfate substrate and its model includes a mean age equation. The neural network used for identification is trained at every hour, based on the experimental data from the process and the process parameters are given by the neural network weights determined at every training step. The mean age model has been validated based on the fact that the parameters of the mean age equation are the same with the ones from the other model equations (biomass, substrate and enzyme-substrate complex). The mean age control is of PI type. The feedback contains an on-line identification recurrent neural network, together with a mean age observer.