小波包变换广义回归神经网络分光光度法同时测定铜铅锌  

Simultaneous determination of copper(Ⅱ),lead(Ⅱ) and zinc(Ⅱ) in spectrometry using a wavelet packet transform based generalized regression neural network

作　　者：高玲[1] 任守信[1] 

机构地区：[1]内蒙古大学化学化工学院,内蒙古呼和浩特010021

出　　处：《冶金分析》2007年第10期6-10,共5页Metallurgical Analysis

基　　金：国家自然科学基金(20667002);内蒙古自然科学基金(200408020210)

摘　　要：本文建立了一种小波包变换广义回归神经网络(WPTGRNN)方法,同时测定Cu(Ⅱ),Pb(Ⅱ)和Zn(Ⅱ)。该法结合小波包变换(WPT)和广义回归神经网络(GRNN)改进除噪质量和预测能力。信号的小波包描述可提供信号的局部时间和空间信息,从而提高了信号和噪音之间的分离能力。除噪质量还可经最好基原理和阈值操作得到进一步改进。广义回归神经网络能克服反传训练所面临的收敛问题及促进非线性计算。通过最佳化,选择了小波函数、小波包分解水平及GRNN的平滑因子,偏最小二乘法(PLS)用于比较研究。编制了3个程序(PWPTGRNN,PGRNN和PPLS)进行相关计算,所有组分的预测标准误差(SEP)和相对预测标准误差(RSEP)分别为8.0×10-7mol/L和5.5%,WPTGRNN法是成功的且优于GRNN及PLS方法。A wavelet packet transform based generalized regression neural network （WPTGRNN） was developed to perform simultaneous spectrophotometric determination of copper（Ⅱ）, lead（Ⅱ） and zinc（Ⅱ）. This method combines wavelet packet transform （WPT） with generalized regression neural network （GRNN） for improving the effect of the noise removal and enhancing its ability of prediction. Wavelet packet signals provided a local time-frequency formation and enhanced the obility to separate signal and noise. The effect of the noise removal can be further improved by using best-basis algorithm and thresholding operation. Generalized regression neural network （GRNN） was applied for overcoming the convergence problem in back propagation training and facilitating nonlinear calculation. By optimi- zation, wavelet function, decomposition level and smoothing factor of GRNN were selected. The partial least squares （PLS） was used for comparative study. Three programs including PWPTGRNN, PGRNN and PPLS, were designed to perform relative calculations. The SEP and RSEP for total elements are 8. 0×10^-7 mol/L and 5.5%, respectively. Experimental results showed that WPTGRNN was successful and more than others.

关 键 词：小波包变换 广义回归神经网络 Cu(Ⅱ) Pb(Ⅱ) Zn(Ⅱ) 

分 类 号：TG115.3[金属学及工艺—物理冶金]