CIESC Journal ›› 2022, Vol. 73 ›› Issue (2): 782-791.DOI: 10.11949/0438-1157.20210791

• Process system engineering • Previous Articles     Next Articles

Near-infrared spectroscopy online detecting for 2,6-dimethylphenol purity based on transfer learning

Yunfei WU(),Xiaoli LUAN(),Fei LIU   

  1. Key Laboratory of Advanced Process Control for Light Industry of the Ministry of Education, Institute of Automation, Jiangnan University, Wuxi 214122, Jiangsu, China
  • Received:2021-06-10 Revised:2021-09-03 Online:2022-02-18 Published:2022-02-05
  • Contact: Xiaoli LUAN

基于迁移学习的2,6-二甲酚纯度近红外光谱在线检测

邬云飞(),栾小丽(),刘飞   

  1. 江南大学自动化研究所,轻工过程先进控制教育部重点实验室,江苏 无锡 214122
  • 通讯作者: 栾小丽
  • 作者简介:邬云飞(1998—),男,硕士研究生,yfwuu@vip.jiangnan.edu.cn
  • 基金资助:
    国家自然科学基金项目(61991402)

Abstract:

Near-infrared spectroscopy is used to on-line detect the product purity of distillation purification process of 2,6-dimethylphenol (2,6-DMP) monomer separation section. Because the product purity of the 2,6-DMP product tower is relatively high (usually 99.10%—99.95%), the purity value of the samples is concentrated in a small range, the coefficient of variation between samples is small, the discrimination of NIR spectrum is low, and the correlation between NIR spectrum and physical property concentration is small, a reliable model cannot be established. To accurately detect the product purity of the 2,6-DMP product tower, thereby real-time control of the product quality, the migration learning algorithm is introduced to make full use of the similarity of the near-infrared spectroscopy data between the different towers in the 2,6-DMP distillation and purification process. With the help of the near-infrared spectroscopy data of the lower 2,6-DMP purity in other towers, the performance of the near-infrared model of the higher 2,6-DMP purity in the product tower is improved. Finally, the validity of the method is verified by establishing near infrared on-line detection models for the purity of 2,6-DMP in a synthetic material company. The results show that with the help of NIR data of different number and purity range, the accuracy of the purity model after transferring is different. It illustrates that the transfer learning algorithm can not only solve the modeling problem in the case of small spectral discrimination effectively, but also the accuracy of the model is related to the number of spectra and the purity range of product.

Key words: 2,6-dimethylphenol, near-infrared spectroscopy, online detecting, transfer learning, reactive distillation, algorithm

摘要:

使用近红外光谱分析技术,对2,6-二甲酚(2,6-DMP)单体分离工段精馏提纯过程的产品纯度进行在线检测。由于2,6-DMP产品塔的产品纯度较高(通常为99.10%~99.95%),样本的纯度值集中分布在一个较小的范围,样本数据之间的变异系数小,样本之间的区分度低,导致所采集光谱与产品纯度之间缺乏相关性,因而无法建立可靠的近红外模型。为了准确检测2,6-DMP产品塔的产品纯度,从而对产品质量进行实时调控,引入迁移学习算法,充分利用2,6-DMP精馏提纯过程中不同塔之间近红外光谱数据的相似性,借助于其他塔中较低2,6-DMP纯度的近红外光谱数据,改善产品塔中较高2,6-DMP纯度的近红外模型性能。通过对某合成材料公司2,6-DMP精馏提纯过程的产品纯度建立近红外在线检测模型,验证了所提方法的有效性。分析结果表明,借助于不同样本数量以及纯度范围的近红外光谱数据,所建模型精度有所不同,说明迁移学习算法不仅可以有效解决样本区分度低情形下的建模问题,且建模精度与所借助光谱的样本数量以及纯度范围密切相关。

关键词: 2,6-二甲酚, 近红外光谱, 在线检测, 迁移学习, 反应精馏, 算法

CLC Number: