Reflectance spectroscopy based rapid determination of coal quality parameters

摘要

In this work, the reflectance spectroscopy of 212 coal samples of different origins was investigated across the Vis-NIR-SWIR range (wavelength: 350-2500 nm) to estimate their ash, moisture, volatile matter, fixed carbon content and gross calorific value (GCV). Several mathematical pre-treatments were applied to each spectrum for improving the signal-to-noise ratio. Partial-least-square (PLS), random forest (RF), and extreme gradient boosting (XGBoost) based regression methods were used to capture the relationships between coal quality parameters with corresponding spectral responses. The predictive models were generated by taking a combination of a set of differently pre-processed spectra with the above-mentioned regression methods to obtain the optimal prediction performance. The results show that spectral pre-processing improves the prediction accuracy of a model. Excessive pre-processing, however, could reduce the model accuracy due to the loss of information. RF regression model works best for estimating moisture and fixed carbon content, while XGBoost shows the best result for ash content and GCV, and PLS models provide the most accurate prediction for volatile matter content.