Unveiling the predictive power of machine learning in coal gross calorific value estimation: An interpretability perspective

The calorific value of coal is a fundamental parameter for assessing its economic viability and environmental impact as a fuel source. Traditional empirical methods, such as Dulong's formula, often fall short in accuracy across diverse coal types and geographic regions. Although machine learning models can significantly improve predictive accuracy, their “black-box” nature often poses challenges in terms of transparency and interpretability, hindering their adoption in industrial applications. This study addresses these dual challenges by proposing a highly accurate and interpretable framework for predicting gross calorific value of coal. Four machine learning models, including Random Forest Regression (RFR), Support Vector Machine (SVM), Gradient Boosting Regression Tree (GBRT), and eXtreme Gradient Boosting (XGB), are employed to predict the gross calorific value of coal. A total of 3,344 coal samples from the U.S. Geological Survey Coal Quality Database are included in the study. The XGB model achieved the highest predictive performance with an R2 of 0.9908, demonstrating its capability to capture complex, non-linear relationships. To enhance interpretability, Explainable Artificial Intelligence (XAI) techniques, such as Local Interpretable Model-agnostic Explanations (LIME), Accumulated Local Effects (ALE), and Individual Conditional Expectation (ICE), were employed. These methods elucidated the influence of key variables, with carbon, hydrogen, and pyritic sulfur identified as major contributors to gross calorific value, while moisture, oxygen, and major oxides exhibited negative impacts. By bridging the gap between predictive accuracy and model transparency, this study provides a novel framework for coal quality analysis, advancing sustainable and informed energy resource management.