Scholarlink

Owing to preeminent capabilities of modeling temporally correlated data, linear dynamical systems (LDS) have been widely utilized as soft sensors for predicting industrial quality-related variables that are difficult to measure in real-time. However, outliers in industrial data prevent the LDS from learning appropriate parameters, leading to compromised generalization performance of the LDS. In order to deal with this issue, this paper proposes a semi-supervised robust LDS (SsRLDS). In the SsRLDS, a novel probabilistic graphical model is first designed for improving the immunity of the LDS upon outliers contaminated training data, and an efficient semi-supervised learning algorithm without resorting to mathematical approximations is developed for training the SsRLDS. Finally, the effectiveness of the SsRLDS is verified through a numerical case and an actual industrial case.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Semi-supervised Robust Linear Dynamical System for Industrial Quality Variable Prediction

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Wenxue Han,
  • Weiming Shao,
  • Chihang Wei

摘要

Owing to preeminent capabilities of modeling temporally correlated data, linear dynamical systems (LDS) have been widely utilized as soft sensors for predicting industrial quality-related variables that are difficult to measure in real-time. However, outliers in industrial data prevent the LDS from learning appropriate parameters, leading to compromised generalization performance of the LDS. In order to deal with this issue, this paper proposes a semi-supervised robust LDS (SsRLDS). In the SsRLDS, a novel probabilistic graphical model is first designed for improving the immunity of the LDS upon outliers contaminated training data, and an efficient semi-supervised learning algorithm without resorting to mathematical approximations is developed for training the SsRLDS. Finally, the effectiveness of the SsRLDS is verified through a numerical case and an actual industrial case.