根据市售鼠药样品成分各异且相对复杂,建立6种不同成分体系和9个不同样本容量的校正集,运用小波变换压缩鼠药的近红外透射光谱数据,结合BP反向神经网络算法对压缩的数据进行建模,考察校正集样品特性对模型预测能力的影响.试验结果表明:采用BP神经网络算法建立定量模型时,只要校正集样品中包含了与预测样品性质相似的样本,就能准确地对复杂样品进行近红外定量分析.当校正集容量分别为72和84时,模型预测结果趋于平稳.当校正集数量为96时,模型的最大相关系数为0.959 8,预测最小标准差和平均相对误差分别为1.893%和1.92%.

According to the diversity and complexity of components in various rodenticides on market,6 calibration sets with different componental systems and 9 calibration sets with different sample capacities were established for the NIRS analysis of the rodenticides on market for their tetramine contents,to reduce the NIR spectral data by the wavelet transform (WT) method,and to establish the model for the determination by using in combination with the method of back propagation neural network (BP) and various methods for pretreatment of the NIR spectra. It was shown that in modelling for quantification by the BP method,accurate NIRS determination of tetramine in complex samples was attained when samples having properties similar to the predictive samples were included in the calibration sets. Predictive results obtain by the model were found to attain to steadiness when the values of capacity of calibration sets were taken as 72 and 84. When the number of calibration sets taken was 96,maximum value of 0.959 8 was found for the correlation coefficient,and minimum values of RMSEP and average relation deviation obtained were 1.893% and 1.92% respectively.