神经群结构,算法与X射线荧光光谱分析研究

研究了一种多组分型神经群网络结构，根据多元体系各变量间的内在规律，可在神经网络中由相互间具有紧密联系的一些神经元的集合形成群结构，采用这种全连接方式的神经群网络结构，减少了连接权重，剔除了噪音，从而增强了模型稳定性，提高了Ｘ射线荧光光谱预测准确度，显著增加了神经网络的外推预测能力，降低了训练模型所需的标样数。

Neural Cluster Structure with Multiple Component Prediction Based on Back Error Propagation in X Ray Fluorescence Spectrometry

A neural cluster structure with multiple component prediction based on back error propagation was proposed. The neural cluster structure is the collection of a group of the neurons that have the close relationship among one another in the neural network. Compared with the classical multiple component prediction based on back error propagation, the stability of the neural cluster structure with disturbance factors increases. Its prediction accuracy to unknown samples and to outlier improves significantly. The structure decreases also the number needful to real standard samples.