遗传神经网络结合LIBS技术对钢液Mn元素定量分析  

Quantitative analysis of Mn element in liquid steel by means of combination of genetic neural network and LIBS technology

作　　者：马翠红 赵士超 MA Cuihong;ZHAO Shichao(College of Electrical Engineering,North China University of Science and Technology,Tangshan 063000,China)

机构地区：[1]华北理工大学电气工程学院,河北唐山063000

出　　处：《现代电子技术》2018年第15期169-173,共5页Modern Electronics Technique

基　　金：国家自然科学基金(61271402)~~

摘　　要：遗传神经网络与激光诱导击穿光谱技术(LIBS)相结合的方法能够更好地对钢液成分进行定量分析检测。建立基于遗传算法为核心的三层误差反向传播(BP)分析模型,由于BP网络的初始权值和阈值是随机数,因此存在收敛速度慢、不能保证收敛全局最优解等缺点,而遗传算法能够优化出最佳的初始权值和阈值,可以较好地克服这些问题。网络的输入选取几种元素的峰值强度与Fe元素的峰值强度进行峰值归一化处理;网络的输出为元素浓度。构建遗传神经网络定量分析模型对钢液中的Mn元素进行定量分析,得到相对标准差(RSD)为7.46%,相关系数为0.996。实验结果表明,遗传神经网络结合LIBS技术相比传统LIBS定标分析法检测的结果精确度有了一定提高。The method combined with genetic neural network and laser?induced breakdown spectroscopy(LIBS)can perform the quantitative analysis and detection for the composition of liquid steel.A three?layer error back propagation(BP)analysis model based on genetic algorithm is established.The BP network has slow convergence speed and can′t guarantee the global optimal solution because its initial weights and thresholds act as the random numbers,but the genetic algorithm can optimize the best initial weights and thresholds,and perfectly overcome these problems.The peak intensities of Fe element and other elements are selected as the inputs of the network for peak normalization.The element concentration acts as the network output.The genetic neural network quantitative analysis model is constructed to quantitatively analyze the Mn element in liquid steel,which can obtain that the relative standard deviation(RSD)is 7.46%,and the correlation coefficient is 0.996.The results show that the quantitative analysis method combined with genetic neural network and LIBS technology has higher result accuracy than the traditional LIBS calibration analysis method.

关 键 词：光谱学 激光诱导击穿光谱技术 实验装置 神经网络 遗传算法 定量分析 

分 类 号：TN247-34[电子电信—物理电子学]