Wavelength selection methods for NIR calibration in petrochemicals: Status, characteristics, and scenarios analysis

Wavelength selection is the crucial step in constructing a near-infrared (NIR) quantitative analysis model, which plays a crucial role in addressing the curse of dimensionality. Effective wavelength selection eliminates redundant and irrelevant wavelengths, significantly improving the speed and efficacy of data analysis. Given the complexity and collinearity of the spectral data from petrochemical products, numerous wavelength selection methods have been proposed to address the issue of dimensionality. However, no unified standard exists to select the appropriate wavelength selection algorithm for specific problems. Currently, the trial-and-error method is primarily used for screening in practical applications. To facilitate researchers and engineers in understanding the characteristics of existing wavelength selection technologies and make more informed choices, this paper categorizes wavelength selection methods into continuous wavelength-based and projection-based methods, depending on whether the original feature space is maintained. Furthermore, experiments are conducted on several methods for predicting gasoline properties, including wavelength point selection (WPS), wavelength interval selection (WIS), and projection techniques, and their status, characteristics, and scenarios are analyzed. Finally, this paper provides some recommendations for potential future research directions, encompassing hybrid methods, interpretability, and generality, to advance the development and application of wavelength selection technologies in petrochemicals.