Scholarlink

<p>This work introduces a robust machine learning framework for accurately predicting CO<sub>2</sub> adsorption capacity (AC) in porous adsorbent materials, offering a significant advancement over conventional experimental and analytical approaches. Specifically, we combine a cascaded forward neural network (CFNN) with advanced local and global explainable artificial intelligence (XAI) techniques to achieve both high predictive accuracy and interpretability. The model is trained on an expanded dataset exceeding 2700 data points, encompassing diverse families, such as metal-organic frameworks (MOFs), zeolites, porous organic polymers (POP), and carbon-based materials (CBM). The CFNN is further optimized through three distinct learning algorithms, namely Levenberg–Marquardt, Bayesian Regularization, and Scaled Conjugate Gradient. Quantitatively, the CFNN-LM model achieved a determination coefficient (R<sup>2</sup>) of 0.9991 and a root mean square error (RMSE) of 0.0659, outperforming alternative ML frameworks and surpassing the models previously reported in the literature. Moreover, by integrating SHAP and LIME for global and local interpretability, we provide physical insights into the factors influencing adsorption performance under varying conditions. Beyond modeling accuracy, this framework offers tangible applications enabling rapid, low-cost pre-screening of adsorbents for industrial sectors, while also equipping researchers with a transparent and scalable tool for accelerating material discovery. By bridging the gap between computational intelligence and practical deployment, this work contributes a scientifically credible and operationally valuable asset to the evolving landscape of carbon capture solutions.</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

Modeling CO2 adsorption capacity of diverse porous adsorbent materials: robust machine learning frameworks and insights from local and global explainable artificial intelligence

  • 期刊论文

  • 发表日期：

    2025年11月2日

查看全文
  • Menad Nait Amar,
  • Saad Alatefi,
  • Ahmad Alkouh,
  • Okorie Ekwe Agwu

摘要

This work introduces a robust machine learning framework for accurately predicting CO2 adsorption capacity (AC) in porous adsorbent materials, offering a significant advancement over conventional experimental and analytical approaches. Specifically, we combine a cascaded forward neural network (CFNN) with advanced local and global explainable artificial intelligence (XAI) techniques to achieve both high predictive accuracy and interpretability. The model is trained on an expanded dataset exceeding 2700 data points, encompassing diverse families, such as metal-organic frameworks (MOFs), zeolites, porous organic polymers (POP), and carbon-based materials (CBM). The CFNN is further optimized through three distinct learning algorithms, namely Levenberg–Marquardt, Bayesian Regularization, and Scaled Conjugate Gradient. Quantitatively, the CFNN-LM model achieved a determination coefficient (R2) of 0.9991 and a root mean square error (RMSE) of 0.0659, outperforming alternative ML frameworks and surpassing the models previously reported in the literature. Moreover, by integrating SHAP and LIME for global and local interpretability, we provide physical insights into the factors influencing adsorption performance under varying conditions. Beyond modeling accuracy, this framework offers tangible applications enabling rapid, low-cost pre-screening of adsorbents for industrial sectors, while also equipping researchers with a transparent and scalable tool for accelerating material discovery. By bridging the gap between computational intelligence and practical deployment, this work contributes a scientifically credible and operationally valuable asset to the evolving landscape of carbon capture solutions.