激光诱导击穿光谱(LIBS)结合机器学习算法快速测定石油焦中微量元素

石油焦中微量元素对其作为预焙阳极的性能起着决定性的作用。首先，通过基于LIBS光谱构建用于石油焦中铁(Fe)和铜(Cu)定量分析的PLS校正模型。然后，考察了不同光谱预处理(归一化、多元散射校正、标准正态变换、一阶导数和二阶导数)以及变量选择算法(粒子群优化算法和变量重要性投影)对PLS校正模型预测性能的影响。建立了一种基于激光诱导击穿光谱(Laser-induced breakdown spectroscopy, LIBS)结合偏最小二乘(Partial least squares, PLS)的石油焦中微量元素定量分析方法。结果显示，与其他PLS校正模型相比，基于二阶导数和变量重要性投影的PLS模型对Fe的预测性能最优，最优的交叉验证相关系数(R-squared cross validation,R²cv)为0.966 7,均方根误差(Root mean squared error cross validation, RMSEcv)为10.282 1 mg/kg,预测集的相关系数(R-squared prediction,R²p)为0.86...

Rapid Determination of Trace Elements in Petroleum Coke by Laser-induced Breakdown Spectroscopy Combined with Machine Learning Algorithms

Trace elements in petroleum coke play a crucial role in determining its performance as a prebaked anode. Firstly, a calibration model using partial least squares (PLS) was constructed to quantitatively analyze the element of Fe and Cu in petroleum coke, employing laser-induced breakdown spectroscopy (LIBS) spectra. The study investigated the impact of various spectral pretreatments (normalization, multiple scattering correction, standard normal variate, first derivative, and second derivative) as well as variable selection algorithms (particle swarm optimization algorithm and variable importance projection) on the predictive accuracy of the PLS calibration model. Consequently, a method for quantitative analysis of trace elements in petroleum coke was developed by combining LIBS with PLS. The results revealed that the PLS model based on the second derivative and variable importance projection demonstrated the optimal prediction accuracy for Fe, exhibiting the coefficient of determination R-squared cross validation (R2cv) of 0.9667, root mean squared error cross validation (RMSEcv) of 10.2821 mg/kg, prediction set coefficient of determination R-squared prediction (R2p) of 0.8638, and root mean squared error prediction (RMSEp) of 14.5078 mg/kg. Likewise, for Cu, the PLS model employing multiplicative scatter correction and particle swarm optimization exhibited the highest predictive performance, with the R2cv of 0.9553, RMSEcv of 6.3300 mg/kg, prediction set R2p of 0.9269, and RMSEp of 7.6502 mg/kg. Hence, the combination of LIBS and PLS offers a viable approach for the rapid detection of trace element content in petroleum coke.