Enhancing LIBS Analysis of Heavy Metals in Fly Ash: A Hybrid Strategy Integrating Data Augmentation with Selective Ensemble

Abstract

The rapid and accurate quantification of heavy metal elements in fly ash is essential for environmental monitoring and resource recovery. As a rapid and in-situ analytical technique, laser-induced breakdown spectroscopy (LIBS) provides an effective approach for the detection of heavy metals in fly ash, yet its quantitative analysis is often constrained by spectral interference and limited samples. To address these issues, this study developed a whale optimization algorithm-selective ensemble (WOA-SE) model based on data augmentation for the accurate quantification of Cu and Zn in fly ash. First, LIBS was used to collect spectral data of fly ash samples, followed by the application of the synthetic minority oversampling technique (SMOTE) to expand the training set to compensate for sample insufficiency. Then, multiple preprocessing methods were applied to the spectra, and competitive adaptive reweighted sampling (CARS) was employed for feature selection to reduce spectral redundancy. Finally, a WOA-SE model was established using the processed data to quantitatively analyze Cu and Zn. The results indicate that the WOA-SE model significantly outperforms individual models, with determination coefficients of prediction (R2 P) for Cu and Zn reaching 0.9639 and 0.9614, respectively, and corresponding root mean square errors of prediction (RMSEP) of 0.0013 wt% and 0.0099 wt%. The analytical strategy proposed in this study significantly improves the accuracy and robustness of LIBS in quantifying heavy metals in fly ash, providing a reliable approach for rapid detection of heavy metals in solid waste.