Abstract: The data of a fermentation process is large, a single data-based soft sensor model suffers from heavy burden calculation and poor precision. In order to solve these problems, an improved multi-model soft sensor modeling method is proposed based on neural network and affinity propagation (AP) clustering. AP clustering algorithm is presented to solve the existed problems in original clustering algorithms, such as clustering number should be determined in advance and cluster accuracy depends on data distribution. Sub neural network models can be constructed based on the clusters. The proposed modeling method was applied to monitor the biomass concentration of an erythromycin fermentation process. Case studies show that the approach has better performance on calculation and accuracy.

Key words:  Soft sensor, affinity propagation (AP) clustering, multi-mode; neural network, fermentation process

摘要： 发酵过程数据量庞大，基于该数据建立的单一化软测量模型存在计算负担沉重，计算精度不良等问题。为有效解决以上问题，结合仿射传播聚类算法和神经网络提出一种改进的多模型软测量建模方法。采用仿射传播聚类算法能较为准确地确定系统的划分数目，有效地解决了传统聚类算法中聚类数目需提前给定、分类精度取决于数据分布、收敛速度慢等问题；针对已划分好的聚类个数建立相应的神经网络子模型。以红霉素发酵为工程背景，将所提方法运用在生物量浓度监测上，结果阐明所建改进的多模型软测量方法计算时间和预测精度得到了良好的改善。

关键词: 软测量, 仿射传播聚类, 多模型, 神经网络, 发酵过程