Determination of coal properties using laser-induced breakdown spectroscopy combined with kernel extreme learning machine and variable selection
Rapid and online analysis of coal properties is extremely important for reasonable and clean utilization of coal. In this study, laser-induced breakdown spectroscopy (LIBS) was applied for analysis of coal properties. The kernel extreme learning machine (K-ELM) method was used to establish a nonlinear model, and particle swarm optimization (PSO) was used as the variable selection method to eliminate useless information and improve prediction ability of the model. The influence of different pretreatment methods was also investigated by 10-fold cross validation (CV); moreover, based on the optimal pretreatment method, three K-ELM models with full spectra, characteristic lines and PSO were developed and compared for predicting ash content, volatile matter content and calorific value of coal. The root mean squared error of cross-validation (RMSECV), correlation coefficient of cross-validation (RCV), root mean square error of prediction (RMSEP) and correlation coefficient of prediction (RP) were used to evaluate model performance; the corresponding RMSEP and RP values were 1.8957% and 0.9936 for ash content based on the K-ELM model with characteristic lines, 1.0874% and 0.9945 for volatile matter, and 0.6999 MJ kg−1 and 0.9872 for calorific value based on the K-ELM model with PSO. The results demonstrate that LIBS coupled with K-ELM and variable selection is a promising technique for rapid analysis of coal properties, and it will also be helpful for effective, clean utilization of traditional energy sources.