CIESC Journal ›› 2018, Vol. 69 ›› Issue (6): 2567-2575.DOI: 10.11949/j.issn.0438-1157.20171388

Previous Articles     Next Articles

Prediction approach for terminal batch process quality based on process transfer

CHU Fei1, CHENG Xiang1, DAI Wei1, ZHAO Xu1, WANG Fuli2   

  1. 1. School of Information and Control Engineering, China University of Mining and Technology, Xuzhou 221116, Jiangsu, China;
    2. State Key Laboratory of Integrated Automation of Process Industry, Northeastern University, Shenyang 110819, Liaoning, China
  • Received:2017-10-17 Revised:2017-12-12 Online:2018-06-05 Published:2018-06-05
  • Supported by:

    supported by the National Natural Science Foundation of China (61503384, 61603393), the Natural Science Foundation of Jiangsu Province (BK20150199, BK20160275, BK20150204), the Fundamental Research Funds for the Central Universities (2015QNA65) and the Jiangsu Provincal Postdoctoral Fund (1501081B).

基于过程迁移的间歇过程终点质量预报方法

褚菲1, 程相1, 代伟1, 赵旭1, 王福利2   

  1. 1. 中国矿业大学信息与控制工程学院, 江苏 徐州 221116;
    2. 东北大学流程工业综合自动化国家重点实验室, 辽宁 沈阳 110819
  • 通讯作者: 褚菲
  • 基金资助:

    国家自然科学基金项目(61503384,61603393);江苏省自然科学基金项目(BK20150199,BK20160275,BK20150204);中央高校基本科研业务费专项资金(2015QNA65);江苏省博士后基金项目(1501081B)。

Abstract:

Adequate process data is the foundation for implementing a data-driven modeling approach. Nevertheless, it is often impossible to meet this requirement for a new batch process. A new quality prediction approach based on process transfer was proposed to establish an accurate prediction model for new batch processes without sufficient data. With application of multivariate statistical regression analysis model, JY-PLS (Joint-Y partial least squares) regression model, this approach realized rapid modeling and quality prediction of new batch processes by construction of common latent variable space between similar batch processes and transfer of present data information from similar batch processes to new and non-modeled batch processes. During online application, the process transfer model was updated with online data and simultaneously estimated confidence interval of prediction error to determine stability of the prediction error. In order to overcome adverse effects on process transfer model caused by possible differences between batch processes, similar process data was eliminated gradually according to data similarity. Finally, effectiveness of the proposed approach was verified by penicillin process simulation.

Key words: batch-wise, process transfer, model, prediction, model update, data elimination, algorithm

摘要:

提出了一种基于过程迁移的间歇过程质量预报方法,旨在解决新间歇过程数据不足难以建立准确预报模型的问题。该方法基于多元统计回归分析模型,通过构建相似间歇过程间的共同潜变量空间,将已有相似间歇过程的数据信息迁移应用到未建模的新间歇过程中,实现新间歇过程的快速建模和质量预报。在线应用时,利用在线数据不断更新过程迁移模型;同时,实时估计模型预测误差的置信区间,判断预报模型预测误差的稳定性;为克服相似过程间可能存在的差异给迁移模型带来的不利影响,根据数据相似度逐步剔除相似间歇过程数据。最后,通过仿真实验验证了所提方法的有效性。

关键词: 间歇式, 过程迁移, 模型, 预测, 模型更新, 数据剔除, 算法

CLC Number: