Accuracy improvement of iron ore analysis using laser-induced breakdown spectroscopy with a hybrid sparse partial least squares and least-squares support vector machine model

Abstract

The quantitative analysis of iron ore by laser-induced breakdown spectroscopy (LIBS) is usually complicated due to nonlinear self-absorption and matrix effects. To overcome this challenge, a hybrid sparse partial least squares (SPLS) and least-squares support vector machine (LS-SVM) model was proposed to analyze the content of total iron (TFe) and oxides SiO₂, Al₂O₃, CaO, and MgO in iron ore. In this study, 24 samples were used for calibration and 12 for prediction. Sparse partial least squares was used for variable selection and establishing the multilinear regression model between spectral data and concentrations; LS-SVM was used to fit the residual errors of the SPLS regression model to compensate for the nonlinear effects. The model parameters were determined by using the tenfold cross-validation (CV) method. With the hybrid model, the root-mean-square-error of prediction (RMSEP) values of TFe, SiO₂, Al₂O₃, CaO, and MgO were 0.6242, 0.3569, 0.0456, 0.0962, and 0.2157 wt%, respectively. The results showed that the hybrid model yielded better performance than only the conventional SPLS or LS-SVM model. This study demonstrated that the hybrid model is a competitive data processing method for iron ore analysis using LIBS.