基于极限学习机的生化过程软测量建模

针对极限学习机方法隐层神经元数目过多的缺陷，提出一种改进的极限学习机方法。在单隐层前向神经网络的隐层中，增加一类分类神经元，从而形成了一种新的单隐层神经网络结构。针对不同类样本数不相同的问题，提出了处理方法，使得可以利用相同的隐层神经元对不同类的学习样本进行拟合，这使得网络的隐层神经元数目大大降低，从而简化了模型的结构，提高了神经网络的计算速度。将这一方法应用于诺西肽发酵过程，建立了菌体浓度的软测量模型，实现了菌体浓度的在线预估。

Soft Sensing Modeling Based on Extreme Learning Machine for Biochemical Processes

An improved extreme learning machine (ELM) method was proposed for soft sensing modeling, and the disadvantage of profuse hidden nodes in ELM network was overcome. By adding a kind of sorting neurons into the hidden layer of the single-hidden-layer feed forward neural network (SLFN), a new SLFN structure was proposed. A method was proposed to solve the problem of the number difference of the sample sorts, and then the same hidden neural nodes could be used to fit the different kind of samples. By this way, the number of hidden nodes was decreased, the model structure was simplified, and the on-line computation speed was increased markedly. The proposed method provided a new approach to build soft sensing models, and it was successfully applied to the soft sensing modeling the of thalli concentration for the Nosiheptied fermentation process to realize the on-line prediction of the thalli concentration.