Performing parameter optimization and variable selection simultaneously in Fourier transform infrared and laser-induced breakdown spectroscopy data fusion

Abstract

The increasing complexity of analysing objects presents higher and more severe challenges to analysts, and the integration strategy of multiple technologies has attracted more and more attention. In this study, low-, mid- and high-level data fusion strategies combined with kernel extreme learning machine (KELM) were applied to take advantage of the synergistic effect of information obtained from Fourier transform infrared (FTIR) spectroscopy and laser-induced breakdown spectroscopy (LIBS) for coal property analysis. In order to further improve the prediction performance of the KELM model, the marine predators algorithm (MPA) was proposed to simultaneously optimize parameters and input variables. The results showed that the prediction performance of all models was further enhanced after simultaneous parameter optimization and variable selection. The high-level data fusion model with simultaneous parameter optimization and variable selection was the most effective strategy, achieving optimal results for three of the four coal properties (carbon, ash content and calorific value). The only exception was volatile matter, where the high-level data fusion model with simultaneous parameter optimization and variable selection obtained the optimal cross-validation results, but the prediction results were slightly worse than those of the FTIR model with simultaneous parameter optimization and variable selection. The overall results presented in this work evidence that utilizing the synergy of FTIR and LIBS is a promising method for the determination of coal properties, while simultaneous optimization of parameters and input variables based on the MPA can further effectively improve the prediction performance of the quantitative model.