光谱学与光谱分析, 2017, 37 (2): 429, 网络出版: 2017-06-20  

基于SiPLS特征提取和信息融合的汽油中乙醇含量的多光谱检测

Multi Spectral Detection of Ethanol Content in Gasoline Based on SiPLS Feature Extraction and Information Fusion
作者单位
1 燕山大学信息科学与工程学院, 河北省特种光纤与光纤传感重点实验室, 河北 秦皇岛 066004
2 内蒙古民族大学物理与电子信息学院, 内蒙古 通辽 028000
摘要
采用紫外可见(ultraviolet/visible, UV-Vis)光谱技术和近红外(near-infrared, NIR)光谱技术及信息融合技术对乙醇汽油中乙醇含量进行了检测。 首先采用组合区间偏最小二乘(synergy interval PLS, SiPLS)算法作为特征提取方法, 分别建立了基于UV-Vis和NIR光谱的偏最小二乘(PLS)回归模型; 再根据油品的实际情况, 运用信息融合理论将UV-Vis和NIR光谱信息进行融合, 建立了数据级融合(low level data fusion, LLDF)和特征级融合(mid-level data fusion, MLDF)模型, 并与单谱源模型效果进行了比较, 确定了最优模型为数据级融合后再进行矢量归一化的模型(LLDF-VN1); 最后分别用高乙醇含量样品和市售汽油样品的光谱数据对该最优模型进行了通用性检验。 结果表明: UV-Vis和NIR光谱数据单独建模均能很好的检测并提供较好的预测结果; 而UV-Vis和NIR光谱数据直接融合在基于校正集的回归模型中效果最好, 其校正集相关系数rc=0999 9, 校正集交叉验证均方差RMSECV=0125 8, 校正集整体评价偏差Biasc=0000 6; 而采用数据级融合后再进行矢量归一化的模型(LLDF-VN1)的预测效果为最佳, 其rp=0999 1, RMSEP=0352 7, Biasp=-0073 8; 自配溶液对最优模型(LLDF-VN1)的通用性验证中, rp=0999 7, RMSEP=0329 1, Biasp=0102 2; 市售汽油对最优模型(LLDF-VN1)的通用性验证中, rp=0990 1, RMSEP=0892 7, Biasp=0675 1。 实验结果说明通过将UV-Vis和NIR光谱信息进行数据级融合可以快速、 准确的检测出乙醇汽油中乙醇的含量, 并能实现乙醇浓度的宽范围检测, 为进一步实现混合油品中物质的快速检测奠定了基础。
Abstract
The ethanol content in ethanol gasoline was detected with ultraviolet/visible(UV/vis) and near-infrared (NIR) spectroscopy while information fusion technology and synergy interval PLS(SiPLS) algorithm were used as the feature extraction method with the establishment of partial least squares(PLS) regression model. Using the information fusion theory, UV/vis and NIR spectra were used for data fusion, the data level fusion (Low level data fusion, LLDF) and feature level fusion(Mid-level data fusion, MLDF) model were established. The results were compared with the single source modelwith low level data fusion before vector normalization(LLDF-VN1) selected for the optimal model. Finally, the optimal model was tested using the spectral data collected from the samples of high ethanol content and commercial gasoline. The results showed that both UV/vis and NIR can be used to detect and provide good prediction results, whereas direct fusion of the UV/vis and NIR spectral data provided the best results in the regression model based on the calibration set, with the highest correlation coefficient rc, the smallest Biasc and RMSECV values, as 0999 9, 0125 8 and 0000 6, respectively. And the prediction effect of the model of LLDF-VN1(low level data fusion before vector normalization) was the best, rp=0999 1, Biasp=0352 7, RMSEP=-0073 8. In the verification of the optimal model (LLDF-VN1) by the self distribution solution, rp=0999 7, Biasp=0102 2, RMSEP=0329 1; and that for gasoline sold on market, rp=0990 1, RMSEP=0675 1, Biasp=0892 7, respectively. It showed that the data level fusion based on UV/vis and NIR spectral information could be used to detect the content of ethanol in ethanol-gasoline quickly and accurately, achieving a wide range of ethanol concentration detection, which laid a foundation for further realization of the rapid detection of substances in the blended fuel oil.

周昆鹏, 毕卫红, 邢云海, 陈俊刚, 周彤, 付兴虎. 基于SiPLS特征提取和信息融合的汽油中乙醇含量的多光谱检测[J]. 光谱学与光谱分析, 2017, 37(2): 429. ZHOU Kun-peng, BI Wei-hong, XING Yun-hai, CHEN Jun-gang, ZHOU Tong, FU Xing-hu. Multi Spectral Detection of Ethanol Content in Gasoline Based on SiPLS Feature Extraction and Information Fusion[J]. Spectroscopy and Spectral Analysis, 2017, 37(2): 429.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!