基于子空间算法的批量生产过程配方分析

Analysis of batch process recipe based on subspace clustering

在批量生产过程中配方的切换会对系统造成阶跃干扰、对固定参数的控制器和软测量仪表带来一定影响,文章针对此问题用配方聚类的方式对配方分类。针对配方数据是由不同数据类型的大规模高维数据组成的特点,提出了一种改进的子空间算法,该算法通过对子空间基本聚类的最相似定义,确定了子空间搜索方向,从而达到配方聚类的目的。实验仿真结果表明:对比于传统的高维数据子空间聚类算法,该算法提高了子空间聚类的精度和伸缩性;对比于前人的配方聚类算法,该算法更适合于高维大规模的分类型配方数据。

In batch process the recipe switch can lead step interference to the system and make an error on the soft measurement instrument and controller with fixed parameters. To solve this problem, the recipe clustering method is used to classify the recipe. According to the characteristic that the recipe data consist of different large-scale and high-dimensional data types, an improved subspace clustering algorithm is adopted. The algorithm defines the most similar subspace clustering to determine the search direction of the subspace, leading to the recipe clustering. The results of the experimental simulation show that compared with the traditional high-dimensional data subspace clustering algorithms, this algorithm can improve the precision and scalability of subspace clustering, and compared with the previous recipe clustering algorithms, this algorithm is more suitable for large-scale and high-dimensional recipe data.