基于遗传算法的BP神经网络模型在岩心扫描仪测定海洋沉积物多种组分中的应用研究

海洋沉积物样品成分复杂,由于基体效应的影响,利用岩心扫描仪开展X射线荧光光谱分析只能得到目标元素的强度信息,不利于该方法在成矿机制和古环境等研究领域更好地发挥作用。本文采用岩心扫描仪测定海洋沉积物中的铝硅钾钙钛锰铁钒铬铜锌铷锶钇和铅15种元素,尝试引入BP神经网络模型利用其非线性拟合能力校正基体效应。实验表明,以水系沉积物、海洋沉积物和岩石国家标准物质以及定值海洋沉积物样品为训练样本,采用遗传算法优化BP神经网络的初始权值和偏置,可以有效校正除硅之外的14种元素基体效应的影响,实现了岩心扫描仪XRF测量结果由强度到浓度的转化。本方法的精密度为0.6%~6.8%(RSD,n=11),国家标准物质和海洋沉积物实际样品中15种组分的预测值与参考值的相对偏差在0.5%~17.5%之间,适合于海洋沉积物中多种主次量组分的快速分析,拓展了岩心扫描仪的功能。

Determination of Multi-components in Marine Sediments by Core Scanner Based on BP Neural Network of Genetic Algorithm

Marine sediments have complex components. Due to the influence of the matrix effect, intensities of elements can only be acquired when using a Core Scanner to carry out X-ray Fluorescence Spectrum analysis, which restricts its application in the fields of paleoecology and mineralization. A method has been introduced for the fast determination of Al₂O₃, K₂O, CaO, TiO₂, MnO, Fe₂O₃, V, Cr, Cu, Zn, Rb, Sr, Y and Pb in marine sediments by Core Scanner, the effects of back-propagation neural network on correcting the nonlinear matrix effects have been investigated and are presented in this paper. Experimental results show that using national certified reference materials of stream sediments, marine sediments and rocks as training samples, a genetic algorithm is used to optimize the initial weight and bias of BP neural network. The matrix effect of 14 elements except Si was corrected by the GA-BP neural network method, which converts the Core Scanner X-ray Fluorescence Spectrum output results from intensities to concentrations. The relative standard deviations of this method are 0.6%-6.8% (n=11). The relative deviations between the predicted values and the reference values of the 15 components of the national standard materials and marine sediment samples range from 0.5% to 17.5%. This indicates that the proposed method is suitable for fast analysis of multi-components in marine sediments, extending the functions of the Core Scanner.