摘要
快速准确测定铁矿石中的硅、铝、钙、镁含量对铁矿石质量评价具有重要作用。受制于多变量分析方法过拟合现象以及不同种类样品基体效应,使用激光诱导击穿光谱(LIBS)准确测定铁矿石中硅、铝、钙、镁含量仍然是当前存在的挑战。采用变量重要性-反向传播人工神经网络(VI-BP-ANN)辅助LIBS定量分析铁矿石中硅(以SiO_(2)计)、铝(以Al_(2)O_(3)计)、钙(以CaO计)和镁(以MgO计)的含量。在这项研究中,收集了12种244批铁矿石代表性样品的LIBS光谱,优化了光谱预处理方法,使用随机森林(RF)对LIBS光谱特征的重要性进行了测量,使用袋外(OOB)误差优化RF模型参数,变量重要性阈值用于优化BP-ANN校准模型的输入变量。变量重要性阈值和神经元数量通过五折交叉验证(5-CV)的测定系数(R^(2))和均方根误差(RMSE)进行优化。结果显示测试样本SiO_(2)、Al_(2)O_(3)、CaO和MgO含量预测均方根误差(RMSEP)分别为0.3772 wt%、0.1339 wt%、0.0592 wt%和0.1411 wt%,R^(2)分别为0.9701、0.9554、0.9871、0.9975。相比于使用相同的预处理方法作为PLS、SVM、RF和BP-ANN四种模型的输入,VI-BP-ANN在校准集和预测集都显示出出色的预测能力。结果表明LIBS与VI-BP-ANN的结合有潜力在实际应用中实现铁矿石硅、铝、钙、镁含量的快速准确预测。
The rapid and accurate determination of calcium,magnesium,aluminium and silicon content in iron ore plays an important role in iron ore quality assessment.The accurate determination of calcium(CaO),magnesium(MgO),aluminium(Al_(2)O_(3))and silicon(SiO_(2))in iron ore using laser-induced breakdown spectroscopy(LIBS)remains a challenge due to the overfitting of multivariate analysis methods and matrix effects between different types of samples.In this paper,variable importance-back propagation artificial neural network(VI-BP-ANN)assisted LIBS was used for the first time to quantify the content of SiO_(2),Al_(2)O_(3),CaO and MgO in iron ore.In this study,LIBS spectra of 12 representative samples of 244 batches of iron ore were collected,spectral pre-processing methods were optimised,the importance of LIBS spectral features was measured using random forest(RF),RF model parameters were optimised using out-of-bag(OOB)errors,and variable importance thresholds were used to optimise the input variables for the BP-ANN calibration model.The variable importance thresholds and the number of neurons were optimised by five-fold cross-validation(5-CV)of the coefficient of determination(R^(2))and root mean square error(RMSE).The results showed root mean square error of prediction(RMSEP)for the SiO_(2),Al_(2)O_(3),CaO,MgO content of the test samples were 0.3723 wt%,0.1298 wt%,0.0524 wt%and 0.1490 wt%respectively,with R^(2)of 0.9771,0.9504,0.9878 and 0.9977,respectively.Compared to using the same preprocessing method as input to the three PLS,SVM and RF models,the VI-BP-ANN model showed excellent performance in both the calibration dataset and prediction dataset.The results indicate that the combination of LIBS and VI-BP-ANN has the potential to achieve fast and accurate prediction of calcium,magnesium,aluminium and silicon content of iron ore in practical application.
作者
刘曙
金悦
苏飘
闵红
安雅睿
吴晓红
LIU Shu;JIN Yue;SU Piao;MIN Hong;AN Ya-rui;WU Xiao-hong(Technical Center for Industrial Product and Raw Material Inspection and Testing of Shanghai Customs District,Shanghai 200135,China;College of Materials&Chemistry,University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2023年第10期3132-3142,共11页
Spectroscopy and Spectral Analysis
基金
海关总署科研项目(2020HK253)资助。
关键词
铁矿石
反向传播人工神经网络
变量重要性
定量分析
激光诱导击穿光谱
Iron ore
Back propagation artificial neural network
Variable importance
Quantitative analysis
Laser-induced breakdown spectroscopy
