基于多模型共识的近红外光谱波长选择方法

Wavelength Selection in Near Infrared Spectral Analysis Based on Consensus Modeling

本文提出了一种基于多模型共识的近红外光谱波长选择方法。该方法首先建立多个无信息变量消除法(UVE-PLS)模型进行波长选择,然后用这些模型的共识结果对波长变量的可靠性进行衡量,进而选择可靠、有效地波长变量。对模拟数据和真实近红外光谱数据的测试结果表明,该方法能够有效地对波长变量进行选择,优于传统的UVE-PLS方法。

Wavelength selection is one of the most important problems in near infrared spectral analysis. In this paper, a new method consensus uninformative variable elimination-partial least square（CUVE- PLS） based on consensus modeling was proposed to select informative wavelengths in near infrared spectral analysis. Firstly, the new method constructed several UVE-PLS models, then the wavelength was evaluated by using the consensus of the UVE-PLS models. The results obtained by testing on simulated data and real near infrared spectral data show that the proposed method can select wavelength effectively and performs better than the traditional UVE-PLS.