Scholarlink

<p>Simulated Moving Bed (SMB) chromatography separation is a novel absorptive separation technique with high separation capacity, and it is challenging to make the process run stably at the desired operating point for a long time. Therefore, powerful, and adaptive soft sensor models are important for the overall stability, efficiency, and optimization of the SMB chromatographic separation process. Extreme Learning Machine (ELM), known for its robust generalization capability, can serve as a valuable soft-sensor model for predicting economic and technical indicators such as purity and yield in SMB chromatography separation processes. To improve the prediction accuracy of ELM, avoid stochasticity and learn incrementally, a bidirectional online sequential ELM (BOSELM) is proposed. BOSELM learns data with fixed or varying block sizes on a one-by-one or block-by-block basis (data chunks) without the need to define the size of the network beforehand and determines the output weights based on the analysis of the sequentially arriving data. On the other hand, BOSELM also makes use of the bidirectional ELM's ability to explore the number of hidden nodes to reduce the number of hidden nodes without affecting the learning efficiency, improving the speed and adaptability of model training. The moving window (MW) strategy was adopted to adaptively correct the BOSELM (MW-BOSELM) to address the issue of decreasing model prediction accuracy caused by changes in process conditions. Additionally, the MW strategy kernel-extreme learning machine (MW-KELM) exhibits higher prediction accuracy than MW-BOSELM at certain instances. To enhance the adaptability of the model, an adaptive hybrid soft-sensor model is proposed to intelligently switch between BOSELM and KELM. Comparative analysis with previous models like BELM, OSELM and BOSELM highlights the superiority of the proposed hybrid adaptive soft-sensor model. Thus, these models help to improve the operational efficiency, product quality and economics of the SMB chromatographic separation process.</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

Bidirectional online sequence extreme learning machine and switching strategy for soft-sensor model of SMB chromatography separation process

  • 期刊论文

  • 发表日期：

    2025年1月1日

查看全文
  • Yong-Cheng Sun,
  • Jie-Sheng Wang,
  • Cheng Xing,
  • Yi-Peng Shang-Guan,
  • Xiao-Tian Wang,
  • Song-Bo Zhang

摘要

Simulated Moving Bed (SMB) chromatography separation is a novel absorptive separation technique with high separation capacity, and it is challenging to make the process run stably at the desired operating point for a long time. Therefore, powerful, and adaptive soft sensor models are important for the overall stability, efficiency, and optimization of the SMB chromatographic separation process. Extreme Learning Machine (ELM), known for its robust generalization capability, can serve as a valuable soft-sensor model for predicting economic and technical indicators such as purity and yield in SMB chromatography separation processes. To improve the prediction accuracy of ELM, avoid stochasticity and learn incrementally, a bidirectional online sequential ELM (BOSELM) is proposed. BOSELM learns data with fixed or varying block sizes on a one-by-one or block-by-block basis (data chunks) without the need to define the size of the network beforehand and determines the output weights based on the analysis of the sequentially arriving data. On the other hand, BOSELM also makes use of the bidirectional ELM's ability to explore the number of hidden nodes to reduce the number of hidden nodes without affecting the learning efficiency, improving the speed and adaptability of model training. The moving window (MW) strategy was adopted to adaptively correct the BOSELM (MW-BOSELM) to address the issue of decreasing model prediction accuracy caused by changes in process conditions. Additionally, the MW strategy kernel-extreme learning machine (MW-KELM) exhibits higher prediction accuracy than MW-BOSELM at certain instances. To enhance the adaptability of the model, an adaptive hybrid soft-sensor model is proposed to intelligently switch between BOSELM and KELM. Comparative analysis with previous models like BELM, OSELM and BOSELM highlights the superiority of the proposed hybrid adaptive soft-sensor model. Thus, these models help to improve the operational efficiency, product quality and economics of the SMB chromatographic separation process.