[Wavelength Variable Selection Method in Near Infrared Spectroscopy Based on Discrete Firefly Algorithm].

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Context:

• Near-infrared (NIR) spectral data presents challenges due to its large size, impacting computational complexity and model performance.
• Effective wavelength variable selection is crucial for developing accurate and robust spectral calibration models.
• Existing methods may not fully optimize the selection process for complex spectral datasets.

Purpose:

• To develop and evaluate a novel wavelength variable selection method for NIR spectroscopy using a discrete firefly algorithm (DFA).
• To reduce computational complexity and enhance the accuracy and robustness of spectral calibration models.
• To compare the performance of the DFA-based method against traditional full spectrum modeling and genetic algorithm approaches.

Summary:

• A discrete firefly algorithm (DFA) was developed for wavelength selection in NIR spectroscopy, incorporating improvements like adaptive attractiveness and elitist strategies for faster convergence.
• Outlier removal using Monte Carlo and calibration/prediction set selection via Kennard-Stone method were employed.
• The DFA method was applied to build a partial least squares (PLS) calibration model for succinic acid concentration in fermentation broth, demonstrating superior performance.

Impact:

• The DFA-based wavelength selection significantly improved the performance of the PLS-NIR calibration model, achieving high correlation coefficients (R2c=0.986, R2p=0.969) and low errors (RMSEC=0.409, RMSEP=0.458).
• This method offers a superior alternative to full spectrum modeling and genetic algorithms for NIR spectral data filtering and analysis.
• The findings highlight the potential of DFA for efficient and accurate chemometric analysis in various applications involving large spectral datasets.