摘要
为了提高液态钢多种成分的检测精度,利用LIBS技术结合极限学习机对液态钢中多种成分进行定量分析。为了降低仪器和环境带来的干扰,采用分析元素特征谱线的积分强度和基体Fe元素内标归一化作为极限学习机(ELM)的输入,分析元素浓度和Fe元素浓度比作为极限学习机的输出。将实验数据分为两部分,一部分用来对极限学习机进行训练,另一部分作为验证。从MATLAB仿真中可以看出,ELM的学习速度快、精度高、泛化能力强,可用于本次实验研究。利用ELM建立的数学模型可以快速、方便地定量分析液态钢的多种成分。分析结果表明,基于极限学习机的LIBS钢液定量分析Ni、Si、Mn元素的相对标准偏差(RSD)分别为8.37%、8.21%和5.3%,均方根误差分别为0.601%、0.422%、0.411%。该种分析法较BP神经网络和SVM精度均有一定的提高。
In order to improve the accuracy of a variety of compositions in liquid steel,the quantitative analysis of compositions in liquid were measured with LIBS technology and extreme learning machine.In order to reduce the disturbance of instruments and environment,the element spectral lines of the integral strength and internal standard unitary matrix Fe element were used as extreme learning machine(ELM)input,analysis element concentrations and Fe element concentration ratio were as the output of the extreme learning machine.Experimental data was divided into two parts,one were used for training,the other as the validation.Can be seen from the simulation in MATLAB,ELM had the characteristics of learning speed,high precision and strong generalization ability,and could be used for this experiment research.The mathematical model of ELM could analyze quickly and easily a variety of ingredients in liquid steel.The results showed that the relative standard deviation(RSD)of quantitative analysis of liquid steel with technology of LIBS based on ELM were 8.37%,8.21% and 5.3%,root mean square error(MSE)were 0.601%,0.422%,0.411%.Compared with quantitative analysis methods of precision of BP neural network and SVM,ELM had a higher accuracy.
出处
《河北联合大学学报(自然科学版)》
CAS
2016年第2期19-25,共7页
Journal of Hebei Polytechnic University:Social Science Edition
基金
国家自然科学基金项目(61271402)
关键词
激光诱导击穿光谱技术
极限学习机
液态钢
多元素定量分析
laser induced breakdown spectroscopy
extreme learning machine
liquid steel
multi-element quantitative analysis