Scholarlink

<p>In actual industrial operation process, some key performance indicators (KPIs) are tricky to detect online due to the characteristics of the detection equipment and the nature of the parameters. Moreover, these KPIs usually present small sample attributes. In this article, a stable and efficient soft measuring model for the KPIs of industrial processes is proposed using deep forest regression (DFR) and multi-layer state transition algorithm (STA). First, DFR is used to build soft measuring models for KPIs with random initial hyperparameters. Second, an improved dynamic STA (DSTA) is developed to optimize the DFR’s hyperparameters. Furthermore, the probability parameters of the DSTA structure are optimally selected using a STA. Finally, gradient refinement is utilized to fine-tune the state factor, which achieves a more accurate optimization process during the internal iteration process. The proposed algorithm is evaluated on the benchmark function, dataset, and an actual industrial problem. Results prove that the use of our method in soft measuring modeling can be effective.</p>

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

Using deep forest regression and multi-layer state transition algorithm to soft measuring modeling with small sample data

  • 期刊论文

  • 发表日期：

    2025年2月13日

查看全文
  • Heng Xia,
  • Jian Tang,
  • Wen Yu

摘要

In actual industrial operation process, some key performance indicators (KPIs) are tricky to detect online due to the characteristics of the detection equipment and the nature of the parameters. Moreover, these KPIs usually present small sample attributes. In this article, a stable and efficient soft measuring model for the KPIs of industrial processes is proposed using deep forest regression (DFR) and multi-layer state transition algorithm (STA). First, DFR is used to build soft measuring models for KPIs with random initial hyperparameters. Second, an improved dynamic STA (DSTA) is developed to optimize the DFR’s hyperparameters. Furthermore, the probability parameters of the DSTA structure are optimally selected using a STA. Finally, gradient refinement is utilized to fine-tune the state factor, which achieves a more accurate optimization process during the internal iteration process. The proposed algorithm is evaluated on the benchmark function, dataset, and an actual industrial problem. Results prove that the use of our method in soft measuring modeling can be effective.