Ann methods are feasible for the verification measurements in nuclear safeguards . experimental data sets have been used to study the performance of neural networks involving radial basis function neural network and generalized regression neural network ( grnn ) . the optimization of the parameter spreads have been given and the analysis error of grnn no more than 0 . 2 % 分析结果表明，使用泛化能力较高的混合训练集训练神经网络，网络给出的富集度值与标准样品的标称值之间的相对差异小于13 % ；使用泛化能力相对较弱的分组训练集训练神经网络，网络给出的分析结果的不确定度小于2 % ；使用分组训练集和广义回归神经网络，网络给出的分析结果的不确定度小于0
