Application of Artificial Neural Network to State Estimation of Process of Erythromycin Fermentation

The application of dynamic BP network and RBF network to the state estimation of process of erythromycin fermentation was studied. Their converge speed and learning capability were compared. The research indicated that each of BP network and RBF network had quite good learning capability, but the converge speed of RBF was faster. The potency of erythromycin, glucose concentration, NH 2 N concentration, propanol concentration and cell concentration can be online estimated by these trained artificial neural networks. Therefore, these trained networks can be used in the research of online optimization and control of erythromycin fermentation processes in the future.