Quantitative Detection And Analysis Research Of Coal Quality Internal Components On Near Infrared Spectroscopy

As an important indicator of quality of coal, moisture, ash, volatile and fixed carbon playa vital role in coal production, trade, processing, utilization and fundamental research.However, the disadvantage of current coals is that the long cycle of rapid detection methodsare analyzed, which contains different levels of radioactive materials. Therefore, rapid andnondestructive detection of coal internal components has become an urgent problem. Nearinfrared spectral can effectively solve the lack of traditional detection method of coal quality,to realize the analytic study of the coal quality internal components.This paper studies on near infrared nondestructive testing technology of coal qualityinternal components. Collected diffuse reflection spectra from180coal samples, respectivelycarryed out smooth spectrum, differential treatment, Kramers-Kronigv (K-K transformation)and Attenuated Total Reflection (ATR correction) based on the original spectrum, choose140samples from coal as a calibration set, which respectively established the mathematical modelof moisture, ash, volatile and fixed carbon, the40remaining coal samples was selected asprediction set, which to verify the stability and predictive power.Removal the abnormal samples through mahalanobis distance and modeling results, aprincipal component regression (PCR) and partial least squares regression (PLSR) wasestablished. Comparison that the evaluation indexes of PLSR model is better than that of PCRmodel, selected the partial least square method for the subsequent quantitative analysis. Basedon different series of smooth of the coal quality internal components, Root mean square errorcorrection (RMSEC) increases smooth points with increasing girth, decreases with the theincrease of the number of polynomial and correlation coefficient (R) decreases with the theincrease of smooth points, increases the number of polynomial with increasing girth by PLSRmodel; The model of fixed carbon content has good stability after first derivative,0.33will bethe RMSEC and the root mean square prediction error (RMSEP) phase difference, the rest index cannot improve the model of precision after differential treatment; The prediction andstability model of volatile and fixed carbon have improved better after multiple scatteringcorrection (MSC), ash and fixed carbon model are better after standard normal variate (SNV)preprocessing by PLSR; In order to further optimization analysis, the qualitative screeningwith different wavelengths by PLSR model was setted up, founding that the effect of totalmoisture and ash content modeling are best at800~1200nm, volatile and fixed carbon arerespectively at1600~2000nm.