系统聚类—反向传播(BP)神经网络用于铅相态分析

Application of systematic clustering-back propagation network in the phase analysis of lead

针对传统化学物相分析对浸取剂高选择性要求及严重串相问题等缺点，以稳定的浸取常数为基础，建立了用于铅相态分析的数学模型，将聚类法与反向传播（BP）神经网络相结合，用于模拟地质样品中铅的相态分析。对隐层节点数、学习步长、学习样本、过拟合现象等进行了探讨，为分析结果的准确性提供了保证。对5个模拟样品进行了分析，相对误差小于土4％，相对标准偏差为0．99％～1．55％，结果优于传统化学物相分析结果。

Aim at the weakness of traditional chemical phase analysis method which has high selectivity request of extraction solvent and serious problem of cluster phase, a new mathematics model of combining systemic clustering and back propagation network is worked out. Base on the stability of extracting constant, the method can be used to the phase analysis of lead in simulative geological samples. The hidden nodes,study step,learning sample and over-fitting , which offer assurance of accuracy of analysis result were discussed. The proposed method has been used for the determination of lead in five simulative geological samples with relative errors of less than ±4% and RSD of 0. 99%-1. 55%. The method is superior to traditional method.