.png){kind=icon}
.png){kind=icon}
Return
Decoding the molecular complexity of coal: A machine learning framework for component classification and solvent selection
.png){kind=icon}
PRE
AI
DOI：10.1016/j.fuel.2025.138142.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Molecular-level characterization of coal organic matter is essential for devising low-emission, high-value utilization strategies. In this study, 12 coals with different ranks and origins were treated by two-stage sequential thermal dissolution (260 degrees C and 300 degrees C) using cyclohexane (CYH) and isopropanol (IPA) as solvents, and 48 dissolution fractions were obtained and analyzed using gas chromatography/mass spectrometry to acquire molecular information. Unsupervised hierarchical clustering analysis and principal component analysis first revealed that CYH selectively enriches aliphatic and aromatic hydrocarbons, whereas IPA favors esters and phenols. Six supervised algorithms, logistic regression, decision tree, random forest (RF), K-nearest neighbor, gradient boosting and support vector machine (SVM), were then trained to predict chemical class. A solvent-partitioned dataset raised the number of reliably predicted compound families from four to seven. Model-driven insights allowed us to propose two directional conversion routes: a phenolic feedstock obtained with IPA at 300 degrees C predicted by RF and an aliphatic feedstock obtained with CYH at 300 degrees C predicted by SVM. These findings provide quantitative guidelines for selecting solvents to target specific chemicals during coal liquefaction.
Keywords:
Thermal dissolution
Gas chromatography/mass spectrometry
Unsupervised learning algorithm
Supervised learning algorithm
Coal
Classification prediction