组合偏最小二乘回归方法在近红外光谱定量分析中的应用

针对近红外光谱数据局部效应显著,变量个数多,彼此间常存在严重的复共线性,并多与样品组分含量呈非线性关系,构建一种组合非线性偏最小二乘回归(E-S-QPLSR)方法。它采用无重复采样技术(subag-ging),从训练样本中生成若干子样,然后每个子样通过二次多项式偏最小二乘回归(QPLSR),建立其子模型,并实现对训练样本因变量的定量预测,再将它们交由线性PLS算法用于计算各子模型的组合权系数。将该法应用于80个玉米样品的水组分含量与其近红外光谱的定量关系建模,效果良好,显示出很强的学习能力,所建模型的预报性能也优于其它方法。

Combined Prediction Model Based on Partial Least Squares Regression and Its Application to Near Infrared Spectroscopy Quantitative Analysis

Aiming at the near infrared spectra (NIR) on local effect sensitivity, numerous predictor variables with serious multicollinearity and having nonlinear quantitative relationship with the chemical compositions from the spectral data, a novel ensemble model, termed as ensemble model based on the subbaggin technique and quadratic partial least squares regression (E-S-QPLSR), was constructed. Firstly, a quite quantity of forecasting sub-models had been established by using the non-linear quadratic partial least squares regression (QPLSR) method and the subagging algorithm, that was a subsampling technique without replacement from the training samples. Then, based on the groups of the training samples forecasting values from the above sub-models, all of the sub-model weighting cocefficients were calculated by using the linear PLSR algorithm. Finally, the application to the corn samples water content modeling of the proposed E-S-QPLSR method was presented in comparison with some other methods. The E-S-QPLSR method not only holds on fine learning ability, but also improves the prediction performance and steady capability.