论文部分内容阅读
目的:利用拉曼光谱技术建立误差反向传播人工神经网络(back-propagation artificial neural networks,BP-ANN)定量分析模型,用于紫石英中CaF_2含量的快速检测。方法:采集紫石英及紫石英不同CaF_2含量配比品共计54批样品的拉曼光谱,光谱在OPUS 7.0软件中作Fourier自去卷积(Fourier self-deconvolution,FSD),并以一阶导数(FD)+矢量归一化(VN),9点平滑的方法进行预处理,逐级组合区间偏最小二乘法(si PLS)优选建模谱段,以乙二胺四乙酸滴定法测定各批样品中CaF_2含量结果为参考值,在Matlab 2014a软件中以BP-ANN算法建立CaF_2快速定量模型。结果:建立了以1 675~1 625,1 525~1 475,850~800,750~700,650~600 cm~(-1)为特征建模谱段的3层BP-ANN定量分析模型,模型预测均方根误差(RMSEP)2.73%,R~2=85.64%,预测结果最大相对偏差5.55%,平均相对偏差2.30%,平均回收率99.74%,表明模型的预测效果良好。结论:在si PLS优选所得谱段基础上,建立了紫石英BP-ANN定量分析模型,该模型预测能力较好,可用于紫石英中CaF_2含量的快速、准确测定。
OBJECTIVE: To establish a quantitative analysis model of back-propagation artificial neural networks (BP-ANN) by Raman spectroscopy for the rapid detection of CaF_2 in amethyst. Methods: The Raman spectra of 54 batches of samples with different CaF 2 content were collected. The spectra were Fourier self-deconvolution (FSD) in OPUS 7.0 software and the first derivative FD) + vector normalization (VN) and smoothing method at 9 o’clock were used for preprocessing. Sequential partial least squares (si PLS) The results of CaF_2 content were used as reference values. A rapid quantitative model of CaF_2 was established by Matlab-Matlab software in BP-ANN. Results: A 3-layer BP-ANN quantitative analysis model was established with 1675 ~ 1625, 1225 ~ 1475, 850 ~ 800, 750 ~ 700 and 650 ~ 600 cm ~ (RMSEP) 2.73%, R ~ 2 = 85.64%. The maximum relative deviation 5.55%, the average relative deviation 2.30% and the average recovery 99.74%, which indicate that the model has a good predictive effect. Conclusion: Based on the optimized si PLS spectra, a BP-ANN quantitative analysis model was established. The model has good predictive ability and can be used for the rapid and accurate determination of CaF_2 in amethyst.