改进粒子群算法优化SVR的LIBS钢液元素定量分析  

Quantitative Analysis of Liquid Steel Element in LIBS Using SVR Improved by Particle Swarm Optimization

作　　者：杨友良 王禄 马翠红 Yang Youliang;Wang Lu;Ma Cuihong(College of Electrical Engineering,North China University of Science and Technology,Tangshan,Hebei 063210,China)

机构地区：[1]华北理工大学电气工程学院,河北唐山063210

出　　处：《激光与光电子学进展》2020年第5期256-263,共8页Laser & Optoelectronics Progress

基　　金：国家自然科学基金(61171058)。

摘　　要：通过激光诱导击穿光谱(LIBS)对钢液表面的不同位置进行激发检测,对得到的光谱数据进行归一化预处理。通过主成分分析法筛选出4个代表性因素,将得到的4个因素作为输入信息,针对钢液中Mn、Ni、Cr和Si四种元素,训练并建立定标模型。利用Cat-fish粒子群(PSO)算法选出最优参数值,最后用测试集来验证模型的预测效果。实验结果表明:Cat-fish PSO-支持向量回归(SVR)的决定系数R^2大于0.95,相对标准偏差RSD均值为3.53%,均方根误差RMSE在1.5%以内;所提模型优于普通SVR预测模型,能够快速精确检测出元素含量。该研究为LIBS在线准确定量分析钢液元素提供了借鉴性较高的优化算法。The laser induced breakdown spectrum(LIBS)is used to excite and detect the different positions at liquid steel surface,and normalization pretreatment is performed for the spectral data.The four representative factors are screened out by principal component analysis and used as input information.Aiming at the four elements of Mn,Ni,Cr,and Si in liquid steel,the calibration model is trained and established,and the optimal parameter value is selected by Cat-fish particle swarm optimization(PSO)algorithm.Finally,the test set is used for verifying the prediction effect of the model.The experimental results show that the determination coefficient R^2 of Cat-fish PSOsupport vector regression(SVR)is greater than 0.95,the mean value of relative standard deviation RSDis 3.53%,and the root-mean-square error RMSEcan be controlled within 1.5%.The proposed model is superior to the ordinary SVR prediction model,and it can detect the element content quickly and accurately.This study provides an optimization algorithm for the on-line and accurate quantitative analysis of liquid steel elements by LIBS,which has high reference value.

关 键 词：光谱学 激光诱导击穿光谱 Cat-fish粒子群 支持向量回归预测 定量分析 

分 类 号：TN247[电子电信—物理电子学]