Study On Optimization Method Of Coal Property Analysis By Laser-Induced Breakdown Spectroscopy Based On Coal Characteristics

Rapid analysis of coal property plays an important role in development of low carbon,flexible and intelligent operation of coal-fired power plant.Laser-induced breakdown spectroscopy(LIBS)is promising for rapid analysis of coal property due to its advantages such as good safety,simple sample preparation and synchronous analysis of multiple elements.However,the complex physicochemical characteristics of coal aggravate the interference of spectral data uncertainty on quantitative analysis of coal property.Moreover,coal property cannot be comprehensively profiled by individual spectral information.These problems limit the improvement of the accuracy of coal property detection.Therefore,this dissertation studied the optimization methods of coal property detection by LIBS based on the characteristics of coal to meet the application requirement of rapid and accurate detection of coal property.The heterogeneity of coal aggravates the impact of the spectral data uncertainty on the quantitative analysis of coal property.To solve this problem,the method of exploiting data uncertainty to optimize quantitative analysis model of coal property was proposed.The method retained some data uncertainty in the calibration data matrix by representing each sample with several groups of spectral data.It was expected to raise the model toleration to spectral signal variation when data uncertainty was involved in training model.In this work,the performance of the quantitative analysis models trained by different level of data uncertainty was compared with each other.As a result,the quantitative analysis models trained by the spectral data uncertainty has better accuracy and robustness than the traditional model without training with spectral data uncertainty for determination of ash content.Comparing the traditional model with the model trained by appropriate level of data uncertainty,root mean square error of prediction(RMSEP)of ash content reduced from 1.718 wt.% to 1.152 wt.%.It is difficult to improve the accuracy of LIBS detection of volatile matter content and calorific value.Meanwhile,it is well known that the proximate analysis indices and calorific value are associated with each other.Therefore,we proposed the hybrid modeling methods based on the combination of coal property indices and spectral information to determine coal property.The method combines the values of ash and moisture content with LIBS data to establish the quantitative analysis models of volatile matter and calorific value.As a result,the proposed method has better accuracy than modeling method only using LIBS data.The RMSEP of calorific value and volatile matter decreased from 0.451 MJ/kg and 1.043 wt.% to 0.351MJ/kg and 0.690 wt.%.Individual spectroscopy is difficult to provide the element and molecular information of coal synchronously.However,the coal property is related to both element composition and molecular structure.Therefore,we proposed a method combining LIBS and infrared spectroscopy to improve the analysis of coal property.The combination of Fourier transform mid-infrared(FTIR)transmission spectroscopy and LIBS was firstly studied to improve the determination of calorific value and volatile matter content.The results indicated that the combination of FTIR and LIBS show better accuracy than individual spectroscopy.It confirmed the feasibility of improving the detection of coal property by combination of LIBS and infrared spectroscopy.Moreover,the combination of LIBS and near infrared diffuse reflectance spectroscopy(NIRS)was studied to improve the detection of coal property.NIRS,LIBS,and data fusion of NIRS and LIBS(NIRS&LIBS)were used to determine calorific value,volatile matter content,ash content and moisture content.The analysis results of coal property based on different spectral data were compared with each other.The NIRS&LIBS showed best accuracy in prediction of calorific value and volatile matter content.RMSEP of calorific value and volatile matter content is 0.192 MJ/kg and 0.672 wt.%.However,the best accuracy for prediction of ash content and moisture content was achieved by LIBS and NIRS,respectively.RMSEP of ash content and moisture content is 0.774 wt.% and 0.308 wt.%.Finally,the best analysis results of volatile matter,ash and moisture content were used to calculate fixed carbon content.The results indicate that the combination of LIBS and NIRS can realize the synchronous detection of multiple coal properties.At last,comprehensively utilizing the spectral data uncertainty,the correlation among coal property indices and the complementary effect of NIRS and LIBS,a stepwise modeling method based on NIRS and LIBS was proposed to improve the analysis of coal property.First,the ash content and moisture content were analyzed by LIBS and NIRS data with some level of data uncertainty,respectively.Then,the predictive results of ash and moisture combined with NIRS,LIBS and NIRS&LIBS data with some level of data uncertainty was used to determine calorific value and volatile matter content.The results indicated that the combination of NIRS&LIBS data and the predictive results of ash and moisture has the best performance.Compared with the results derived from NIRS&LIBS,the RMSEP of calorific value decreased from 0.536MJ/kg to 0.176 MJ/kg,the RMSEP of volatile matter content decreased from 1.125 wt.% to0.662 wt.%.Finally,the predictive results of volatile matter,ash and moisture content was used to calculate fixed carbon content.The results indicate that the proposed method can realize the synchronous and accurate detection of multiple coal properties.