A method was proposed for the fast determination of Al2O3,SiO2,K2O,CaO,TiO2, MnO,Fe2O3,V,Cr,Cu,Zn,Rb,Sr,Y,Zr,Ba and Pb in marine sediments by core scanner,and the effect of back-propagation neural network on correcting the nonlinear matrix effects were also investigated. The intensity was obtained in the determination of national certified reference materials of stream sedi-ments,marine sediments and rocks by core scanner coupled with pressed powder sample preparation. And the matrix effect of 17 elements was corrected by the methods of GA-BP neural network(17-18-17)using the intensity and certified reference value as training samples. Then forecasting model was established for semiquantitative analysis. The in-situ analysis data of sediment core were evaluated by comparing with the results obtained from artificially layered powder. Since the comparative results based on the two different methods indicated the similar distribution features of multi-components,the described method is appropri-ate for fast semiquantitative analysis of multi-components in marine sediments.%采用岩芯扫描仪测定海洋沉积物中的Al2O3、SiO2、K2O、CaO、TiO2、MnO、Fe2O3、V、Cr、Cu、Zn、Rb、Sr、Y、Zr、Ba和Pb等17种元素,尝试引入遗传算法(GA)优化的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校准样品,采用压片法制样,测定校准样品中相关组分的强度,并作为训练样本代入GA-BP神经网络(17-18-17),可以有效校正基体效应的影响,建立半定量预测模型.海洋沉积物实际岩芯样品的分析结果表明,各层位待测组分预测值与分层取样定值结果的变化趋势基本吻合,适合于海洋沉积物中多种主次量组分的快速分析.
展开▼
