An ensemble model for rapid quantitative determination of vanadium (V) in petroleum coke by laser-induced breakdown spectroscopy

Abstract

Precise detection and analysis of trace elements in materials are critical for various industrial applications. This study integrates laser-induced breakdown spectroscopy (LIBS) with advanced machine learning algorithms—random forest (RF) and gradient boosting decision tree (GBDT)—to develop an ensemble model for the rapid quantitative prediction of vanadium (V) in petroleum coke. A 1064 nm laser was employed to ablate petroleum coke samples, generating plasma, with the resultant spectral data collected via a spectrometer. The spectral data underwent preprocessing to isolate vanadium-specific information. Optimization of the regression prediction parameters for RF and GBDT was achieved through a recursive feature elimination method. Subsequently, an ensemble model was constructed to predict vanadium concentration. The results indicate that the ensemble model demonstrates excellent predictive performance, with R² = 0.99976, RMSECV = 3.47145 mg kg⁻¹, and RMSEP = 3.38779 mg kg⁻¹. Hence, integrating RF and GBDT with LIBS offers a robust and precise methodology for vanadium concentration detection and analysis, providing significant insights and methods for monitoring trace element concentrations in petroleum coke.