X射线光谱与神经网络中单组分型神经群结构研究

Neural Cluster Structure with Single Component Prediction in Multiple Variable Systems for X ray Fluorescence Spectrometry

研究、比较了神经群结构与常规神经网络算法的预测性能，考察了过拟合与最佳拟合态等的关系。结果表明，在多元体系中，将神经网络单组分预测模型应用于Ｘ射线荧光光谱分析时，在预测准确度、模型稳定性和外推预测能力方面，神经群结构优于常规神经网络模型。

A neural cluster structure with single component prediction (NCSCP) was proposed for X ray fluorescence spectrometry in a multivariable system.The neural cluster structure is built by the collection of a group of neurons which have close relationships among one another.In X ray fluorescence analysis,the structure is constructed by choosing the elements in which there exist serious matrix effects,and deleting the components containing large noise.The predictability of the neural cluster structure was compared with that of the classical backward error propagation algorithm with single component prediction.The results show that the nerual cluster structure is significantly superior to the classical algorithm in prediction accuracy,antidisturbance and the predictabilty to outliers.