Abstract: Batch process plays an important role in the processing of specialty chemical,semi-conductor,food and biology industries for producing high-value-added products to meet rapidly changing market,which causes their monitoring and control to emerge as essential techniques.A novel method for uneven-length-stage batch process was proposed by using a new stage segmentation strategy based on k-means clustering method.First,the three-dimension data array is unfolded in variable ways,resulting in two-dimension forms.Then according to data correlations,two-dimension training data set is divided into many clusters by applying the conventional k-means clustering method,and sequentially principal component analysis (PCA) is performed for each separated cluster.For online monitoring,a similarity index is introduced to choose the most suitable local model.The effectiveness and utility of the proposed method was validated through the simulation benchmark of fed-batch penicillin production.

Key words: batch processes, process control, fault diagnosis, uneven-length stage, k-means clustering, principal component analysis, batchwise

摘要： 针对阶段不等长的多阶段间歇过程，提出了一种基于k-均值聚类方法的阶段分段策略，可以将不等长的阶段准确分类。首先，将间歇过程的三维训练数据按变量方向展开成二维矩阵，再通过k-均值聚类的方法按照相关性将数据聚成多类并运用主元分析（PCA）方法分别对每一类建立模型。在线监控时，通过计算样本与模型之间的相似系数以选择最合适的模型进行在线监控。此方法可以将不同批次在同一采样时刻的过程数据按照相关性分到多个阶段，更符合生产过程中常见的过程数据阶段不等长的情况。最后利用青霉素仿真验证了该方法的有效性。

关键词: 间歇过程, 过程控制, 故障诊断, 阶段不等长, k-均值聚类, 主元分析, 间歇式