Scholarlink

<p>This study presents a machine learning-driven framework for performance prediction of ammonia-fueled intermediate-temperature solid oxide fuel cells (IT-AP-SOFCs), integrating numerical simulation with advanced data-centric validation techniques. A comprehensive 2D axisymmetric model incorporating the Temkin-Pyzhev kinetic model for ammonia decomposition was developed, generating 1030 parametric simulations across six operational variables. Eleven machine learning algorithms were evaluated, with XGBoost demonstrating superior predictive accuracy for both power density (R<sup>2</sup> = 0.999) and maximum temperature (R<sup>2</sup> = 0.9999). Permutation importance analysis revealed inlet temperature as the dominant parameter (82.3 % contribution), followed by air velocity (14.4 %) and electrolyte porosity (2.7 %). Rigorous 5-fold cross-validation confirmed model generalizability, while Gaussian noise injection (±10 %) demonstrated robustness under experimental uncertainty (R<sup>2</sup> &gt; 0.87). The synergy of feature importance interpretation, noise resilience validation, and ensemble regularization mechanisms positions XGBoost as an optimal tool for SOFC digital twin development, enabling accurate performance prediction despite complex thermo-electrochemical couplings and real-world measurement variability.</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

Feature importance, K-fold validation, and Gaussian noise analysis in machine learning modeling of ammonia-fueled IT-AP-SOFC performance: A numerical-data synergy

  • 期刊论文

  • 发表日期：

    2025年7月1日

查看全文
  • Mahdi Keyhanpour,
  • Majid Ghassemi

摘要

This study presents a machine learning-driven framework for performance prediction of ammonia-fueled intermediate-temperature solid oxide fuel cells (IT-AP-SOFCs), integrating numerical simulation with advanced data-centric validation techniques. A comprehensive 2D axisymmetric model incorporating the Temkin-Pyzhev kinetic model for ammonia decomposition was developed, generating 1030 parametric simulations across six operational variables. Eleven machine learning algorithms were evaluated, with XGBoost demonstrating superior predictive accuracy for both power density (R2 = 0.999) and maximum temperature (R2 = 0.9999). Permutation importance analysis revealed inlet temperature as the dominant parameter (82.3 % contribution), followed by air velocity (14.4 %) and electrolyte porosity (2.7 %). Rigorous 5-fold cross-validation confirmed model generalizability, while Gaussian noise injection (±10 %) demonstrated robustness under experimental uncertainty (R2 > 0.87). The synergy of feature importance interpretation, noise resilience validation, and ensemble regularization mechanisms positions XGBoost as an optimal tool for SOFC digital twin development, enabling accurate performance prediction despite complex thermo-electrochemical couplings and real-world measurement variability.