关键词: 诺西肽发酵, 建模, 神经网络集成, 差分进化算法, Elman神经网络

Abstract: In order to improve the poor extrapolation effect and generalizability of the single neural network, the neural network ensemble is used to develop the model of Nosiheptide fermentation process. Each individual network is trained on a bootstrap re-sampling replication of the original training data through the Bagging approach. Then, outputs of the individual neural networks are combined to form an overall output of neural network ensemble through the weighted average method, in which the weight of each individual network is determined by the differential evolution algorithm. The Elman network, a typical dynamic neural network, is applied in each individual network. The model of Nosiheptide fermentation product concentration, based on the neural network ensemble, is thus developed through combination of outputs from multi-Elman neural networks. This model is compared with the single neural network model to illustrate its high accuracy and generalizability.