基于低秩重构表示的动态回归迁移模型
CSTR:
作者:
作者单位:

太原理工大学 电气与动力工程学院,太原 030024

作者简介:

通讯作者:

E-mail: yangaowei@tyut.edu.cn.

中图分类号:

TP273

基金项目:

国家自然科学基金面上项目(61973226,62073232,62003233);山西省自然科学基金项目(201901D211083, 20210302123189);格盟集团科技创新基金项目(2022-05).


Dynamic transfer regression model based on low-rank reconstruction representation
Author:
Affiliation:

College of Electrical and Power Engineering,Taiyuan University of Technology,Taiyuan 030024,China

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对实际流程工业过程存在动态时变和概念漂移特性,导致软测量模型预测精度下降的问题,提出基于低秩重构表示的动态迁移回归模型.为了更好地描述动态过程,在动态内模型偏最小二乘框架下,将高维过程数据映射到低维潜变量空间中,以捕获质量变量与潜变量之间的动态相关性.为了减小概念漂移,在获得动态相关性的同时,通过增强不同工况质量变量估计值之间的相关性实现数据的条件分布对齐.在3个公开工业数据集上的实验结果表明:所提出模型的预测精度与静态基模型和动态基模型相比均有所提升,可以有效地提高模型的预测精度和泛化能力.

    Abstract:

    Aiming at the problem of the actual process industry process with dynamic time-varying and concept drift characteristics, which leads to a decrease in the prediction accuracy of the soft sensor model, a dynamic regression migration model based on low-rank reconstruction representation is proposed. In order to better describe the dynamic process, under the dynamic internal model partial least squares framework, the high-dimensional process data is mapped to the low-dimensional latent variable space to capture the dynamic correlation between quality data and latent variables. In order to reduce concept drift, while obtaining dynamic correlation, the conditional distribution alignment of data is achieved by enhancing the correlation between the estimated values of quality variables in different working conditions. Compared with the static base model and the dynamic base model, the experimental results on the three public industrial datasets improved, indicating that the proposed method can effectively improve the prediction accuracy and generalization ability of the model.

    参考文献
    相似文献
    引证文献
引用本文

霍海丹,阎高伟,程兰,等.基于低秩重构表示的动态回归迁移模型[J].控制与决策,2024,39(8):2511-2520

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2024-07-16
  • 出版日期: 2024-08-20
文章二维码