Scholarlink

<p>Predicting catalyst activity and melt flow index (MFI) is critical for optimizing industrial polyethylene (PE) production but remains challenging due to complex nonlinear relationships between process parameters and polymer properties. This study employs machine learning (ML) to predict MFI and catalytic activity in a gas-phase fluidized bed reactor producing linear low-density polyethylene (LLDPE) and high-density polyethylene (HDPE). Using an industrial dataset (903 data points), seven regression models were evaluated including Linear Regression (LR), Random Forest (RF), AdaBoost (AB), Light Gradient Boosting Machine (LightGBM), eXtreme Gradient Boosting (XGB), Categorical Boosting (CatBoost) and the pretrained transformer-based TabPFN. TabPFN demonstrated superior performance, achieving test R² values of 0.957 for MFI and 0.993 for catalytic activity, significantly outperforming conventional models. SHAP (SHapley Additive exPlanations) analysis revealed hydrogen concentration ([H<sub>2</sub>]) as the dominant feature for MFI prediction, acting primarily as a chain transfer agent, while temperature exerted the strongest influence on catalytic activity. Correlation analysis identified key operational clusters, including thermal-catalytic coordination and hydrogen management. The results highlight TabPFN’s robustness and generalizability for real-time process optimization, providing actionable insights for tuning polymerization conditions within fixed catalyst constraints to enhance product quality and efficiency.</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

Predicting polymer melt flow index and catalytic activity in ethylene polymerization using a pretrained transformer-based model

  • 期刊论文

  • 发表日期：

    2026年2月28日

查看全文
  • Soleiman Esmaeilzadeh,
  • Sima Rezvantalab,
  • Sara Mihandoost

摘要

Predicting catalyst activity and melt flow index (MFI) is critical for optimizing industrial polyethylene (PE) production but remains challenging due to complex nonlinear relationships between process parameters and polymer properties. This study employs machine learning (ML) to predict MFI and catalytic activity in a gas-phase fluidized bed reactor producing linear low-density polyethylene (LLDPE) and high-density polyethylene (HDPE). Using an industrial dataset (903 data points), seven regression models were evaluated including Linear Regression (LR), Random Forest (RF), AdaBoost (AB), Light Gradient Boosting Machine (LightGBM), eXtreme Gradient Boosting (XGB), Categorical Boosting (CatBoost) and the pretrained transformer-based TabPFN. TabPFN demonstrated superior performance, achieving test R² values of 0.957 for MFI and 0.993 for catalytic activity, significantly outperforming conventional models. SHAP (SHapley Additive exPlanations) analysis revealed hydrogen concentration ([H2]) as the dominant feature for MFI prediction, acting primarily as a chain transfer agent, while temperature exerted the strongest influence on catalytic activity. Correlation analysis identified key operational clusters, including thermal-catalytic coordination and hydrogen management. The results highlight TabPFN’s robustness and generalizability for real-time process optimization, providing actionable insights for tuning polymerization conditions within fixed catalyst constraints to enhance product quality and efficiency.