Scholarlink

<p>The nonlinear synergistic interactions between biomass and plastics during co-pyrolysis make accurate prediction of bio-oil yield a significant challenge for process optimization. This study aimed to develop a high-fidelity machine learning framework capable of forecasting liquid yield based on complex physicochemical and operational variables. A comprehensive dataset of 192 experimental records was curated to train and validate eight distinct regression algorithms, including Convolutional Neural Networks (CNN), AdaBoost, and Support Vector Regression, using a rigorous 5-fold cross-validation protocol. Performance benchmarking identified CNN and AdaBoost as the superior models, achieving a test coefficient of determination (R<sup>2</sup>) of 0.982 and demonstrating exceptional generalization on unseen data. To transcend “black-box” limitations, SHapley Additive exPlanations (SHAP) analysis was applied, identifying the plastic-to-biomass blending ratio and feedstock elemental oxygen content as the dominant predictors of yield. The results quantitatively confirm that optimizing the feedstock blend ratio is the primary lever for maximizing liquid recovery, surpassing the influence of residence time. This study provides a validated, interpretable computational tool that allows operators to estimate co-pyrolysis performance and design optimal feedstock mixtures without extensive trial-and-error experimentation.</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 Bio-Oil Yield in Co-Pyrolysis of Lignocellulosic Biomass and Waste Plastics Using CNN and AdaBoost: A Machine Learning and SHAP Analysis Approach

  • 期刊论文

  • 发表日期：

    2026年3月10日

查看全文
  • Ahmad Adel Abu-Shareha,
  • Raed Alfilh,
  • Tariq Abdulkader Alrihaim,
  • Raghavendra Rao,
  • Abinash Mahapatro,
  • Karthikeyan A.,
  • Harjot Singh Gill,
  • Yashwant Singh Bisht,
  • Siya Singla,
  • Fatimah Pashtun

摘要

The nonlinear synergistic interactions between biomass and plastics during co-pyrolysis make accurate prediction of bio-oil yield a significant challenge for process optimization. This study aimed to develop a high-fidelity machine learning framework capable of forecasting liquid yield based on complex physicochemical and operational variables. A comprehensive dataset of 192 experimental records was curated to train and validate eight distinct regression algorithms, including Convolutional Neural Networks (CNN), AdaBoost, and Support Vector Regression, using a rigorous 5-fold cross-validation protocol. Performance benchmarking identified CNN and AdaBoost as the superior models, achieving a test coefficient of determination (R2) of 0.982 and demonstrating exceptional generalization on unseen data. To transcend “black-box” limitations, SHapley Additive exPlanations (SHAP) analysis was applied, identifying the plastic-to-biomass blending ratio and feedstock elemental oxygen content as the dominant predictors of yield. The results quantitatively confirm that optimizing the feedstock blend ratio is the primary lever for maximizing liquid recovery, surpassing the influence of residence time. This study provides a validated, interpretable computational tool that allows operators to estimate co-pyrolysis performance and design optimal feedstock mixtures without extensive trial-and-error experimentation.