炉渣是一种典型的冶金固体废弃物,主要由氧化镁、氧化铁和氧化铝等金属氧化物组成.炉渣成分的快速定量分析有助于确定炉渣中有价元素或成分的含量,进而选择合适的资源化途径,实现高效利用,同时减少环境污染.本研究提出了一种基于激光诱导击穿光谱(Laser induced breakdown spectroscopy,LIBS)结合机器学习算法定量分析炉渣中Fe、Si和Ti的方法.首先,采集了炉渣样品的LIBS光谱,并通过国家标准技术研究院数据库(National institute of standards and technology,NIST)鉴定了相关元素的特征谱线.然后,考察了不同光谱预处理方法对偏最小二乘(Partial least squares,PLS)模型预测性能的影响,重点对光谱预处理方法的组合性能进行了探讨.提出了变量重要性投影(Variable importance in projection,VIP)结合灰狼优化算法(Grey wolf algorithm,GWO)的混合变量选择算法,并用于炉渣样品LIBS光谱特征变量筛选.基于交叉验证对预处理方法和特征筛选方法的参数、阈值、输入变量和模型参数等进行了优化.根据优化的参数及输入变量构建了基于LIBS技术的炉渣中Fe、Si和Ti的定量分析模型.结果表明,优化的模型相较于基于原始光谱的模型具有更好的预测性能,其预测决定系数(R2p)分别为0.9525、0.9604和0.9972,预测均方根误差(RMSEp)分别为0.0461、0.0141和0.1963.LIBS结合机器学习算法可为炉渣元素的现场快速检测提供一种可行方法,有望为冶金固废资源化利用提供技术参考.
Rapid Quantitative Analysis of Slag Elements by Laser Induced Breakdown Spectroscopy Combined with Mixed Variable Selection
Slag is a typical metallurgical solid waste,mainly composed of magnesium oxide,iron oxide,alumina oxide and other metal oxides.The rapid quantitative analysis of slag components is helpful to determine the content of valuable elements or components in slag,and then choose a suitable resource utilization way to achieve efficient utilization and reduce environmental pollution.In this study,a quantitative analysis method of Fe,Si and Ti in slag was proposed based on laser induced breakdown spectroscopy(LIBS)combined with machine learning algorithm.Firstly,LIBS spectra of slag samples were collected,and the characteristic spectral lines of related elements were identified through the National Institute of Standards and Technology(NIST)database.Then,the influence of different spectral preprocessing methods on the predictive performance of PLS model was investigated,and the combined performance of spectral preprocessing methods was discussed.On this basis,a mixed variable selection algorithm combining variable importance in projection(VIP)and grey wolf algorithm(GWO)was proposed to screen LIBS spectral characteristic variables of slag samples.Based on cross-validation,the parameters,thresholds,input variables and model parameters of the preprocessing method and feature screening method were optimized.A quantitative analysis model of Fe,Si and Ti in slag based on LIBS technique was established based on the optimized parameters and input variables.The results showed that the optimized model had better prediction performance than the original spectral model,with R2p of 0.9525,0.9604 and 0.9972,and RMSEp of 0.0461,0.0141 and 0.1963,respectively.It was proved that LIBS combined with machine learning algorithm provided a feasible method for the field rapid detection of slag elements.The research is expected to provide some theoretical basis and technical reference for the resource utilization of metallurgical solid waste.
国家科技期刊平台
NSTL
万方数据
