Kohonen-Elman网络在同时测定铸铁中锡钼锑中的应用

多组分同时测定时,由于组分间的相互影响,特征波长的选择是影响计算精度的重要因素。SnⅣ、MoⅥ和SbⅢ均可与水杨基荧光酮(SAF)和溴化十六烷基三甲胺(CTMAB)发生高灵敏度的显色反应,生成稳定的三元胶束化合物,但紫外吸收光谱重叠严重。实验提出将Kohonen神经网络与Elman网络相结合建立了铸铁中3种金属同时测定的定量分析方法。方法利用Kohonen神经网络的聚类能力选择特征波长点,然后用优化后的Elman神经网络对优选特征波长点处的吸光度数据建立预测模型。结果表明,用从全谱中选出的26个波长点吸光度数据建模,整体预测效果最好。将实验方法用于合成样测定,预测结果与实际浓度的平均相对误差绝对值在2.24%～3.10%之间;用于铸铁样中Sn、Mo和Sb同时测定,测定值与原子吸收光谱法测定值吻合较好,相对标准偏差(RSD,n=7)在1.2%～2.7%之间。

Application of Kohonen-Elman network in simultaneous determination of tin,molybdenum and antimony in cast iron

The selection of characteristic wavelength is the important factor to influence the calculation precision during simultaneous determination of multi-components due to the interaction effect among components. Sn(IV), Mo(VI) and Sb(III) can all react with salicyl fluorone (SAF) and cetyltriethylammnonium bromide (CTMAB) to form stable ternary micellar compounds. The sensitivity of reactons are high. However, the ultraviolet absorption spectrua were seriously overlapped. In this study, the combination of Kohonen neural network with Elman network was proposed to establish the quantitative analysis method for the simultaneous determination of three metal elements in cast iron. In this method, the characteristic wavelength point was firstly selected based on the clustering ability of Kohonen neural network. Then, the absorbance data at optimized characteristic wavelength were used to establish prediction model using the optimized Elman neural network. The results showed that the whole prediction results were best when the model was created by the absorbance data at 26 characteristic wavelength points from the whole spectra. The experimental method was applied to the determination of synthetic sample. The absolute values of average relative error between prediction results and actual concentrations were between 2.24% and 3.10%. The proposed method was applied to the simultaneous determination of tin, molybdenum and antimony in cast iron sample, and the found results were consistent with those obtained by atomic absorption spectrometry (AAS). The relative standard deviations (RSD, n=7) were between 1.2% and 2.7%.