RBF网络光度法与BP网络光度法的比较

目的：探讨神经网络光度法用于复方制剂的含量测定。方法：训练集为按L25（5^6）正交表制备的25组标准混合液的吸光度数据和各组分的浓度数据，混合液中各组分的5个浓度水平分别为80％、90％、100％，110％和120％。预报集采用复方制剂的吸光度数据。网络的输入为混合物的吸光度，网络的输出为各组分的浓度。分别用径向基函数网络和Levenberg-Marqurdt优化算法的BP网络处理数据。结果：复方阿司匹林片和联磺甲氧苄啶片的紫外分光光度法测定结果表明，径向基函数网络在网络训练时间和测定精度等方面好于Levenberg-Marqurdt优化算法的BP网络。结论：径向基函数网络光度法测定复方制剂简便，准确。

Comparision Between RBF Network Spectrophotometry and BP Network Spectrophotometry

To delve neural network spectrophotometry appling to content determination of compound preparation. The train set was consisted of data of 25 group standard mixtrue which prepared according to orthogonal layout. The predict set was obsorbance data of compound preparation. The input of network was mixture obsorbance and the output was component concentration. The ab-sorbance data of three component compound preparation was processed with radial basis function network and BP network of Leven-berg-Marqurdt optimazation algorithm, respectively. Compound aspirin tablets and tablets of sulfamethoxazole sulfadiazine and trimethoprim have been determined by UV spectrophotometry. The results obtained with radial basis function network are better than those provided with the BP network in traing time and precision of determination. The radial basis function network spectrophotometry is simple and accuracy in the determination of compound preparation.