基于有偏估计的近红外定量分析方法及应用

Near-infrared Quantitative Analysis Method Based on Biased Estimation and Its Application

根据工况需求建立稳健性好的定量分析模型是使用近红外分析技术的关键。近红外光谱的波长点数多,实际工况中难以获得精确的无偏估计,可控的有偏估计更便于后续优化控制对产品目标属性的调优。鉴于此,提出一种融合互信息的光谱波长选择方法:根据光谱波长间的最大信息系数,选择与目标属性相关性大的波长,实现有监督波长选择。此外,融合即时学习,以有偏最小最大概率机建立定量分析模型,在保证有偏估计的同时提高了模型的自适应能力。工业案例验证结果表明,所提方法可有效提高近红外定量分析模型的精度,减少累积偏差。

The key of near-infrared analysis technology is to establish a robust quantitative analysis model according to the corresponding process conditions.It is difficult to obtain an accurate unbiased estimation in real-world because there are a large number of wavelength points in the near-infrared spectrum.The controllable biased estimation is more convenient for Optimization of product target attributes by optimization control.In view of this,a spectral wavelength selection method based on mutual information is proposed.According to the maximum information coefficient between the spectral wavelengths,the wavelength which has high correlation with the target attribute is selected.In this way,supervised wavelength selection is realized.Besides,the biased minimax probability machine combined with just-in-time learning approach is adopted to establish the quantitative analysis model,which improves the adaptive ability of the model while performing the biased estimation The industrial case shows that the proposed method can effectively improve the accuracy of near-infrared quantitative analysis model and reduce the cumulative error.