在能量色散X荧光分析技术中, 常用基本参数法、 经验系数法、 人工神经网络等方法建立计数率和元素含量之间的物理模型, 此外, GMDH(group method of data handing)作为一种新型的处理复杂非线性问题的方法, 被大量理论和实验证明优于大部分的计算统计方法。 GMDH是一种自组织学习的前馈型网络, 自动筛选并在训练过程中确定其结构, 对GMDH进行改进并对结果进行定量预测, 参考值与预测值的相对误差在5%以内, 方法简洁、 合理、 可靠。

The fundamental parameter method, empirical coefficient method, artificial neural network and some other methods are commonly used to establish the physical model between the count rate and the content of elements in the energy dispersive X-ray fluorescence analysis technique. Besides, through a large number of theoretical and experimental proof, as a new method of dealing with complex nonlinear problems, GMDH (group method of data handing) is better than most of statistical methods of calculation. And is a self-organizing learning in feed forward network, which could auto filter and determine its structure in the training process. Here, we are going to improve GMDH and give a quantitative prediction of the results. And both the reference values and forecast values of relative errors will be less than 5%, which make the method simple, reasonable, and reliable.