摘要
拉曼光谱强度与物质量存在的线性关系会受到许多复杂因素破坏,包括激发光源、聚焦、散射、折射等,导致定量效果不佳。各种因素的干扰效应,总体上分成加性和乘性效应,而消除乘性效应的难度会更大一些。光谱序列信号可视为向量,信号强度对应向量的模量,而体现向量本质的方向属性不会受模量变化的影响。根据这一原理,利用向量的方向确定性,将信号的强度度量转换成空间角度度量,建立了一种消除乘性效应的方法。首先,选择一个与待定量组分相近而与背景空间近似正交的基准向量,并定义移动窗口;然后,计算移动窗口内的光谱向量与基准向量的夹角,所得值存储为矩阵,完成角度描述转换。角度矩阵消除了乘性效应的干扰,而定量关系仍然近似线性,只要将该矩阵的秩满足多元统计建模要求,就可以用于多元校正,并得到良好结果。研究采用甲醇-乙醇-异丙醇混合体系,验证了消除乘性效应后改进的定量效果,对于积分时间波动的预测值与实际值,直接PLS方法的相关系数r为0.911 9,预测标准偏差(RMSEP)为0.110 2;采用MSC预处理的r为0.906 0,RMSEP为0.100 8;而本文提出的VAPLS的r为0.998 7,RMSEP为0.015 2。结果表明向量角转换度量处理后,光谱的乘性干扰得到了有效校正,拉曼定量分析准确性得到了提高。
The linear relationship between the Raman spectral intensity and the analyte amount is frequently disrupted for a variety of complex reasons,which include these variations in laser source,focusing effect,sample scattering and refracting,so that causes poor quantitative results.As a whole,these disturbing effects can be divided to be additive and multiplicative,and the multiplicative effects are generally more difficult to be eliminated.A spectrum is a series data,also can be treated as a vector.In principle,unstable motions in spectrum intensity/amplitude corresponding to the module shifts for a vector,doesn＇t impact the vector direction which is the essence of the vector,so it is reasonable to rewrite the data form on module to on space angle for the same measurement.This thesis employed a data transformation to eliminate the multiplicative effects within spectra,i.e.,the spectrum signal on its amplitude has been transformed to be on the vector angles.The first step of the transformation is the selection of a stand vector which is near to the analyte and almost orthogonal to the background within the sample space;and the next step is to define a moving window,then to find out the angle between the sample vector（i.e.the transformed spectrum）and the stand vector within the window;while the window is moved along the spectrum data series,the transformation for vector angle（VA）series has been finished.The thesis has proved that an approximate linear quantitative relationship has been remained in the VA series.Multivariate calibration need full rank matrix which is combined by spectrum from variety samples,and variety VA series also can combine a full rank VA matrix,so the approximate linear VA matrix still perfectly meeting the demand for multivariate calibration.A mixed system consisted by methanol-ethanol-isopropanol has been employed to verify the eliminations to the multiplicative effects.These measuring values of the system are obtained at different Raman integral times and have remarkable multiplicative effects.In predicting results,the correlation coefficient（r）and the root mean squared error of prediction（RMSEP）from class PLS respectively are 0.911 9and 0.110 2,and 0.906 0and 0.100 8are for the preprocessing by multiplicative scatter correction（MSC）.In contrast,rand RMSEP under the VAPLS,presented by this thesis,respectively are 0.998 7and 0.015 2and are significantly better than others.The VAPLS has eliminated the multiplicative effects of Raman spectra and improved the accuracy of Raman quantitative analysis and it owes to the preprocessing of the vector angle transformation.
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2016年第2期419-423,共5页
Spectroscopy and Spectral Analysis
基金
国家科技支撑计划课题(2011BAE11B00)
广西自然科学基金项目(2014GXNSFAA118056)
广西科技计划项目(桂科攻1355010-15)资助
关键词
拉曼光谱
乘性效应
散射校正
向量角转换
Raman spectrum
Multiplicative effects
Scatter correction
Vector angle transformation
