基于变点检测理论的制丝过程稳态识别方法

摘要

For the purpose of intelligent identification of the steady state of a batch process in primary processing, a method based on change point detection theory was proposed. The primary processing data of a whole processed batch of a selected specification of cigarette brand"Yunyan"were researched via being divided dynamically into subintervals by mean-variance change point detection model and PELT algorithm, then the threshold of variance and/or the mean of the subintervals were set according to technology standard, the subintervals which conformed with the threshold conditions were extracted to form a steady state data set. Finally, the formed data set was compared with the data sets identified by two existing rules of steady state defining. The results showed that: 1)Different steady state identification methods significantly affected the accuracy of batch process data measurement in primary processing. 2)The identification method based on change point detection theory featured better adaptability and identification effect. 3)Combining the said method with Shapiro-Wilk normality test via the built-in R language program in the information system, the online intelligent assessment of process capability was realizable. The proposed method provides a technical support for the intelligent identification of steady state of process in primary processing.%为实现卷烟制丝工序批过程数据稳态的智能识别,提出了一种基于变点检测理论的制丝过程稳态识别方法.以"云烟"某一规格卷烟制丝过程全批次数据为研究对象,首先利用均值方差变点检测模型和PELT算法动态划分子区间,然后根据工艺标准设定子区间方差和(或)均值的阈值,通过筛选出符合阈值条件的子区间从而形成稳态数据集,最后与现行的两种稳态截取规则识别的数据集进行测度指标对比分析.结果表明:①不同稳态识别方式对制丝批过程数据测度的准确性影响较大;②基于变点检测理论的稳态识别方法适应性较好,识别效果较优;③通过信息系统内置的R语言程序,将基于变点检测理论的稳态识别方法与Shapiro-Wilk正态性检验相结合,可实现在线智能判别过程能力.该方法的建立为制丝过程稳态的智能识别提供了技术支持.