Hyperspectral measurement technique based rapid determination of coal quality parameters of Jharia and Raniganj basin coal

Hyperspectral technology has a striking ability to detect the quality of materials, and consequently, research on coal quality using this technique has increased many folds in recent years. For the purpose of assessing the quality of coal, hyperspectral signatures, proximate analysis, and gross calorific value of 78 coal specimens from different parts of the Jharia and Raniganj basins (India) were studied. All the coal samples demonstrate welldefined absorption features at different wavelengths. Based on the spectral responses, seven wavelength regions, namely 400-900 nm, 950-1050 nm, 1350-1500 nm, 1840-1950 nm, 2120-2250 nm, 2260-2350 nm, and 2450-2500 nm, were chosen for this study. Three techniques were applied to capture the relationship between coal spectral absorption features and coal proximate parameters, i.e., XGBoost, Random Forest, and Partial-leastsquare. It has been concluded that the result obtained for random-forest-based regression work best for predicting ash, moisture, and GCV. In comparison, the PLS regression model shows the most accurate predictions for the volatile matter.

Automated summary

Hyperspectral technology is highly effective in assessing coal quality, leading to an increase in research on coal quality using this technique. A study was conducted on the hyperspectral features, proximate analysis, and gross calorific value of 78 coal samples from different parts of the Jharia and Raniganj basins in India to evaluate coal quality. The study found that all coal samples exhibit well-defined absorption features at different wavelengths. Based on spectral responses, seven wavelength regions were chosen for the study. XGBoost, Random Forest, and Partial-leastsquare techniques were used to capture the relationship between coal spectral absorption features and proximate parameters. The study concluded that the random-forest-based regression model provided the best predictions for ash, moisture, and GCV, while the PLS regression model showed the most accurate predictions for volatile matter.