RBF网络在地质样品非线性基体效应校正中的应用

APPLICATION OF RBF NETWORK FOR CORRECTION OF NONLINEAR MATRIX EFFECTS IN GEOLOGICAL SAMPLES

基于RBF(Radial Basis Function)网络的正交最小平方算法,对攀枝花地质样品进行动态非线性基体效应校正。首先采用EDXRF仪器对样品进行测量,并对样品的荧光计数进行归一化,设计出自动分类模型和动态非线性数据基体效应校正模型的二级串联结构,用自组织神经网络(SOFM)进行分类,然后用RBF网络预测了未知样品元素Ti的百分含量。结果表明:对所预测样本的Ti的百分含量与化学分析相比的误差均小于0.5%。由此可见,用RBF方法可以有效地对地质样品进行非线性基体效应校正,能够达到工矿生产的要求。

The orthogonal least-squares algorithm of the RBF network is applied to dynamic correction of the nonlinear matric effect in the geological samples of Panzhihua. First the EDXRF instrument is used to measure the samples, and then the fluorescence counting of samples is normalized. The two cascaded framework of auto classily model and dynamic correction the nonlinear matric effect model is designed, and then the self-organizing network is adopted to classify and the RBF network is used to forecast the content of Ti in the unknown sample. The results show that the error of forecast content of Ti comparing to the chemistry analysis is lower of 0.5%, which suggests that the RBF network can effectively correct the nonlinear matric effect to the geological samples, and can achieve the request of the mine production.