Scholarlink

<p>Operating performance assessment is critical for improving production efficiency and ensuring economic benefits. Traditional echo state networks (ESNs), despite their strong nonlinear and dynamic modeling capabilities, are limited by their reliance on labeled samples and high computational costs when expanded into deep networks. To overcome these challenges, we propose the first deep semi-supervised ESN (DSESN) framework that synergistically integrates unlabeled and labeled data. First, unsupervised learning is performed with both labeled data and unlabeled data in the multi-layer reservoir construction stage. Then, only labeled data are used for classification in the output optimization stage. Furthermore, we propose an enhanced black-winged kite algorithm (EBKA) incorporating weight-splitting with perturbation mechanism to efficiently optimize weight parameters of the network. On this foundation, we develop a distributed operating performance assessment strategy based on DSESN. Finally, the effectiveness and feasibility of DSESN-based assessment framework is validated on the case of hot strip mill process (HSMP). Compared with three popular ESN-based methods, the proposed DSESN has the highest accuracy in two cases, reaching 97.35% and 95.50%, respectively.</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

A deep semi-supervised echo state network-based distributed operating performance assessment framework for manufacturing processes optimized by enhanced black-winged kite algorithm

  • 期刊论文

  • 发表日期：

    2025年7月2日

查看全文
  • Chuanfang Zhang,
  • Wenxiao Yin,
  • Kaixiang Peng,
  • Xueyi Zhang

摘要

Operating performance assessment is critical for improving production efficiency and ensuring economic benefits. Traditional echo state networks (ESNs), despite their strong nonlinear and dynamic modeling capabilities, are limited by their reliance on labeled samples and high computational costs when expanded into deep networks. To overcome these challenges, we propose the first deep semi-supervised ESN (DSESN) framework that synergistically integrates unlabeled and labeled data. First, unsupervised learning is performed with both labeled data and unlabeled data in the multi-layer reservoir construction stage. Then, only labeled data are used for classification in the output optimization stage. Furthermore, we propose an enhanced black-winged kite algorithm (EBKA) incorporating weight-splitting with perturbation mechanism to efficiently optimize weight parameters of the network. On this foundation, we develop a distributed operating performance assessment strategy based on DSESN. Finally, the effectiveness and feasibility of DSESN-based assessment framework is validated on the case of hot strip mill process (HSMP). Compared with three popular ESN-based methods, the proposed DSESN has the highest accuracy in two cases, reaching 97.35% and 95.50%, respectively.