Research On Chemometrics Methods In Laser-induced Breakdown Spectroscopy And Its Application In Field Analysis Of Coal-based Energy Materials

Field analysis of coal-based energy materials is of great significance for the sustainable development of coal and other fossil energy materials.Laser-induced breakdown spectroscopy（LIBS）,with its unique advantages,is considered to be an effective tool for field analysis of coal-based energy materials.However,due to the complexity of composition and structure of sample,LIBS will produce a large number of complex spectra,and how to extract effective information from these complex spectral data for accurate quantitative or qualitative analysis is still one of the difficulties.Based on the actual needs of field analysis of coal-based energy materials,this paper takes coal as the research object,conducts research on quantitative analysis method of coal property based on LIBS spectrum,and focuses on the kernel extreme learning machine（KELM）model based on variable selection and data fusion strategy to solve the complex spectral analysis problem.This study will provide a theoretical basis and technical support for field analysis of coal-based energy materials.The full paper is divided into four chapters,and the main contents are as follows:1.Taking coal as the research object,a method based on LIBS and KELM was established for determination of carbon and sulfur in coal.First,the pretreatment methods,input variables and model parameters were optimized by 5-fold cross-validation.Then,the optimized KELM model was used for the quantitative analysis of carbon and sulfur elements in coal.Compared with support vector machine（SVM）,least squares support vector machine（LS-SVM）and back propagation artificial neural network（BP-ANN）models,the KELM model has the best prediction performance for both carbon and sulfur elements,and the optimal RMSEP and R_P²of carbon are 0.3762%and 0.9994,the optimal RMSEP and R_P² of sulfur are 0.7704%and0.9832.Compared with other methods,KELM has the advantages of fast learning speed and good generalization performance.2.Taking coal as the research object,a method based on LIBS combined with KELM and variable selection was established for rapid determination of coal property.（1）The effects of different pretreatment methods and parameters on the KELM model was explored,and the KELM models based on full spectrum,characteristic lines and particle swarm optimization（PSO）was established to measure the ash content,volatile content and calorific value of coal.The results show that the optimal calibration model for volatile matter and calorific value is KELM model based on PSO,and the optimal calibration model for ash content is KELM model based on characteristic lines;（2）In order to overcome the shortcomings of PSO method,a hybrid variable selection method based on WT-MIV was proposed,and the WT-MIV-KELM model was established for calorific value determination.Compared with KELM and WT-KELM models,the WT-MIV-KELM model has the best prediction performance,and the optimal RMSEP and R_P² are 0.6151 MJ/kg and 0.9759,respectively;（3）In order to overcome the limitation of low computational efficiency of wrapper variable selection method,a hybrid variable selection method（V-WSP-PSO）based on filter method（V-WSP）and wrapper method（PSO）was proposed.Compared with the KELM model based on other variable selection methods,the V-WSP-PSO-KELM model has the best prediction performance,and the optimal RMSEP and R_P² are 0.3534 MJ/kg and 0.9894,respectively.It shows that suitable variable selection method can improve the calculation efficiency and prediction performance of the model,and compared with single variable selection method,hybrid variable selection method has the advantages of high computational efficiency and high accuracy.3.Taking coal as the research object,a method based on the fusion of LIBS and FTIR was established for rapid determination of coal property.（1）KELM models based on low-level data fusion and mid-level data fusion was established for the determination of ash content,volatile matter and calorific value,and the prediction performance was compared with KELM models based on LIBS and FTIR.The results show that the optimal calibration model for ash content is KELM model based on low-level data fusion,and the optimal calibration model for volatile matter and calorific value is KELM model based on FTIR spectrum;（2）In view of large difference in the dimension of two spectral matrix and more interference information in the fusion spectrum,a hybrid variable selection method based on MI-PSO was proposed,and on this basis,the low-level data fusion based on MI-PSO,mid-level data fusion based on MI-PSO and mid-level data fusion based on MI-PSO and PCA were proposed for determination of ash content,volatile matter and calorific value of coal.The results show that the optimal calibration model for ash content and volatile matter is low-level data fusion model based on MI-PSO,and the optimal calibration model for calorific value is mid-level data fusion model based on MI-PSO.It shows that the fusion spectrum has the advantage of high prediction accuracy,and variable selection method can further improve the prediction performance of the fusion model.