基于列文伯格-马夸尔特-反向传播人工神经网络的X射线荧光光谱定量分析方法
目的 建立一种基于列文伯格-马夸尔特-反向传播人工神经网络(Levenberg-Marquardt back-propagation artificial neural networks, LM-BP-ANN)的X射线荧光光谱(XRF)的定量检测分析方法。方法 采集84个土壤样品光谱数据, 预处理后应用主成分分析(PCA)提取特征参数, 随机选取训练集、校正集、预测集样品个数分别为42、21、21。以均方差(MSE)、校正决定系数(R2)、校正标准差(SEC)、验证决定系数(r2)、预测标准差(SEP)和相对预测误差(RPD)为评价指标, 同时分析比较LM-BP-ANN、BP-ANN、PLS三种算法的建模结果, 并利用模型预测土壤重金属含量。结果 实验确定隐含层神经元数目、学习率和迭代次数值依次为: 6、0.1和8, 3种建模方法中LM-BP-ANN效果最优, 模型的相关系数高于0.98, 表明模型有效。结论 模型分析快速, 可用于实际土壤样品中重金属含量的检测, 对于改进X射线荧光光谱仪的检测准确度有着重要的意义。
Objective To establish a quantitative detection analysis method based on Levenberg-Marquardt back-propagation artificial neural network (LM-BP-ANN) for X-ray fluorescence spectrometry (XRF). Methods The spectrometry data of all 84 soil samples were collected. The principal component analysis (PCA) method was used to extract the characteristic variables after preprocessing the spectrometry data. The sample number of training set, calibration set and prediction set were 42, 21 and 21, respectively, which were chosen randomly. Mean square error (MSE), adjusted determination coefficient (R2), square error of calibration (SEC), verify determination coefficient (r2), standard error of prediction (SEP) and relative prediction error (RPD) were used as evaluation indexes in LM-BP-ANN, BP-ANN and PLS algorithm, which were applied for modeling. The modeling results were analyzed and compared, and the quantitative model was used to predict the heavy metals contents in actual soils. Results The values of the hidden layer neuron number, learning rate and iterations count were confirmed as 6, 0.1 and 8 through the experiments. The effects of LM-BP-ANN were prior than the other two modeling methods by the comparison of the modeling results. The correlation coefficients were higher than 0.98, indicating that the calibration model was feasible. Conclusion The rapid analysis method is suitable for the detection the actual samples. It has important significance on improving detection accuracy of X-ray fluorescence spectrometer.
标题：基于列文伯格-马夸尔特-反向传播人工神经网络的X射线荧光光谱定量分析方法
英文标题：Quantitative analysis method based on Levenberg-Marquardt back-propagation artificial neural network for X-ray fluorescence spectrometry
作者：
李芳 北京农业质量标准与检测技术研究中心
陆安祥 北京农业质量标准与检测技术研究中心
王纪华 北京农业质量标准与检测技术研究中心
中文关键词:列文伯格-马夸尔特算法,反向传播神经网络,X射线荧光光谱,
英文关键词:Levenberg-Marquardt algorithm,back-propagation neural network,X-ray fluorescence spectrometry,
发表日期：2015-11-18
- 文件大小:
- 1009.39 KB
- 下载次数:
- 60
-
高速下载
|
|
