A hybrid variable selection method based on wavelet transform and mean impact value for calorific value determination of coal using laser-induced breakdown spectroscopy and kernel extreme learning machine

Abstract

Variable selection plays an important role in the multivariate analysis of laser-induced breakdown spectroscopy (LIBS). In this study, a hybrid variable selection method based on wavelet transform (WT) and mean impact value (MIV) was proposed to extract useful information from LIBS spectra for calorific value determination of coal. Firstly, WT method was employed to filter the useless or irrelevant information from the broadband LIBS spectra, and 881 wavelet coefficients were obtained by global thresholding. Then the wavelet coefficients were further eliminated by MIV method. Finally, 142 wavelet coefficients were obtained by WT-MIV method. The retained wavelet coefficients were used directly as input variables to establish a nonlinear KELM model for calorific value determination of coal. The results demonstrated a significant improvement over full spectra model, with root mean square error of prediction (RMSEP) reducing from 1.2584 MJ/kg to 0.6151 MJ/kg, correlation coefficient of prediction (RP) improving from 0.9802 to 0.9879. It indicates that LIBS coupled with WT-MIV-KELM is a feasible method for calorific value determination of coal, and the hybrid variable selection method was more efficient to reduce the calculation time and improve the model performance.