摘要
针对烟叶中K、Cl两个无机元素含量与近红外光谱数据之间存在复杂非线性关系、常规的线性模型效果差、预测误差大等问题,在传统偏最小二乘法(PLS)的基础上,将基于核变换的非线性PLS建模方法(NPLS)引入到烟叶无机元素的建模中,建立了K、Cl等指标数学模型,并对模型的可行性和有效性进行了研究,同时与PCR、PLS等其它方法线性建模方法建模准确度和误差分布进行了对比。实验证明,该方法对烟叶光谱与无机元素之间的非线性关系进行了处理,模型的准确率更高,预测误差分布更合理。
Due to complicated nonlinear relationship between near infrared spectra of tobacco and contents of inorganic elements such as potassium and chloride, which resulted in poor modeling effects and high prediction errors of conventional linear model, this paper introduced nonlinear partial least squares （NPLS） method based on kernel function transformation into calibration. Mathematical models of potassium and chloride were built and their feasibility and effectiveness were studied. The accuracy and distribution of errors were also compared with other linear calibration methods such as PCR and PLS. Results showed that NPLS processed nonlinear relationship between spectra and inorganic elements of tobacco. The model achieved higher prediction accuracy and more reasonable error distribution. It provided an alternative analytical method and technique for cognizing quality of tobacco in a comprehensive, fast and in-depth manner.
出处
《中国烟草学报》
EI
CAS
CSCD
北大核心
2016年第3期67-71,共5页
Acta Tabacaria Sinica
基金
中国烟草总公司山东省公司科技重大专项和重点资助项目(合同编号:KN223)
关键词
近红外光谱
高斯核函数
非线性PLS
校正分析模型
near infrared spectra
Gaussian kernel function
nonlinear PLS
calibration analysis model
