Neural networks modelling of two biotechnological processes

Neural models of a batch (with DO-uncontrolled and controlled culture) and a fed-batch (with the control action being the flow rate of feed-glucose) fermentation processes for the production of the enzyme superoxide dismutase SOD) and of a continuous process of anaerobic digestion of organic wastes (stimulated by the addition of glucose) have been developed on the basis of experimental and generated data (from a deterministic models). The specialized software package NNSYSID20 for MATLAB 5.3 has been used for training and testing the neural models. ANNs with different structures were tested. The criteria for the choice of the structures were: (a) models validity for experimental data unknown to the network and (b) statistical characteristics. Especially the influence of the number of experimental data in the process of developing of ANN models has been investigated. The models obtained may be useful for simulation and automatic control of these fermentation processes.