Scholarlink

<p>The detailed characterization of solid carbonaceous fuels (SCFs) is crucial for optimal production of second-generation fuels through the thermochemical conversion process. Hence, a comprehensive machine learning (ML) modelling approach is presented for accurately predicting the higher heating value (HHV) of various solid carbonaceous fuels (SCFs). To encompass a wide and diverse range of SCFs, 3771 samples comprising 16 distinct fuel types, categorised into 5 major fuel classes, are used here. Six supervised ML models are used for predicting the HHV and their performance is evaluated. The multilayer perceptron (MLP) has emerged as the most effective model with the highest prediction accuracy among all the ML models. The study proposed a novel two-stage method for predicting the HHV from easily accessible proximate analysis (PA) data. Here, ultimate analysis (UA) data are first estimated from experimentally measured PA data, and then, HHV is predicted from the combination of PA and estimated UA data. The two-stage model shows high accuracy (R<sup>2</sup> = 0.99, MAE = 0.48, and MSE = 0.68) compared to the previous studies reported in the literature. The robustness and generalisation of the model are validated through bootstrapping-based uncertainty analysis. Shapley additive explanations (SHAP) is employed to assess the global and local interpretability by analysing the summary plot, violin plot, and force plot. The applied interpretable ML framework driven by a large and diverse dataset offers a scientifically robust and cost-effective solution for accurately estimating the HHV of a wide range of SCFs from easily available PA data.</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

An interpretable machine learning framework for high-accuracy prediction of higher heating value of diverse solid fuels from proximate analysis

  • 期刊论文

  • 发表日期：

    2026年6月30日

查看全文
  • Ayush Garg,
  • Satyajit Pattanayak,
  • Meraj Alam,
  • Segun E. Ibitoye,
  • Arup Nandi,
  • Chanchal Loha

摘要

The detailed characterization of solid carbonaceous fuels (SCFs) is crucial for optimal production of second-generation fuels through the thermochemical conversion process. Hence, a comprehensive machine learning (ML) modelling approach is presented for accurately predicting the higher heating value (HHV) of various solid carbonaceous fuels (SCFs). To encompass a wide and diverse range of SCFs, 3771 samples comprising 16 distinct fuel types, categorised into 5 major fuel classes, are used here. Six supervised ML models are used for predicting the HHV and their performance is evaluated. The multilayer perceptron (MLP) has emerged as the most effective model with the highest prediction accuracy among all the ML models. The study proposed a novel two-stage method for predicting the HHV from easily accessible proximate analysis (PA) data. Here, ultimate analysis (UA) data are first estimated from experimentally measured PA data, and then, HHV is predicted from the combination of PA and estimated UA data. The two-stage model shows high accuracy (R2 = 0.99, MAE = 0.48, and MSE = 0.68) compared to the previous studies reported in the literature. The robustness and generalisation of the model are validated through bootstrapping-based uncertainty analysis. Shapley additive explanations (SHAP) is employed to assess the global and local interpretability by analysing the summary plot, violin plot, and force plot. The applied interpretable ML framework driven by a large and diverse dataset offers a scientifically robust and cost-effective solution for accurately estimating the HHV of a wide range of SCFs from easily available PA data.