基于FCM的青霉素发酵分时段统计建模及监控

Multi-phase statistical modeling and monitoring based improved FCM for penicillin fermentation process

对间歇过程进行实时监测具有重要的现实意义,传统的多向主元分析方法(MPCA)是用单一的统计模型来表现原始数据的信息,没有考虑到大多数间歇过程由于操作条件或反应进程的改变,不同操作阶段的数据动态特性会不同,同一操作阶段的变量也往往具有高度非线性的特性,因此会导致一些重要信息的缺失。本文针对青霉素发酵过程固有的多时段特性,提出了一种基于模糊C均值算法的分时段过程监控算法,该方法以每个时刻数据矩阵的相似度指标作为聚类输入,以便准确的判断过程特性变化,实现间歇生产过程的阶段划分,进而用MPCA建立多时段过程监控模型,最后再利用相应的统计指标进行过程监测。将该算法应用于青霉素发酵过程的在线监测,实验结果验证了该方法的有效性和可靠性。

The real-time monitoring for batch process has important practical significance. The traditional multi-way principal component analysis （MPCA） is to represent the original single data information using a single statistical model, without taking the characteristics into account as the changing of conditions or reaction process in a batch process. The characteristics include the difference of the dynamic characteristic data in the different operation stages and the high nonlinear of the variables in the same operation phase. Therefore, it will lead to some important information missing. Considering the inherent characteristics of multiple periods in the penicillin fermentation process, the sub-periods process monitoring method is put forward based on the fuzzy c-means algorithm （FCM）, which takes the similarity index with each time data matrix as input, to determine changes in the process characteristics accurately. The batch process is divided into multiple period of time. And then to establish sub-periods process monitoring model using MPCA. Finally, we use the corresponding statistical indicators to achieve process monitoring. The proposed method is used to evaluate the industrial penicillin fermentation process data. The experimental results demonstrate the validity and reliability of the method.