Wavelength variable selection method in near Infrared Spectroscopy based on discrete firefly algorithm

Taking into consideration of the large size of near-infrared spectral data, the spectral data has to be compressed to reduce the computational complexity of the established spectral calibration model and improve accuracy and robustness of the model. Near Infrared Spectroscopy wavelength variable selection method based on discrete firefly algorithm is presented. First, the Monte Carlo method was used to exclude outliers, and Kennard-Stone method was chosen for the selection of calibration set and prediction set. General firefly algorithm was discretized, by improving the attractiveness of adaptive formula, increasing traction weights in mobile formula and so on. In order to adapt to the effects of discretization and optimize algorithm, elitist strategy was added in the discrete firefly algorithm, to acceleratethe convergence rate. The optimum value of the DFA algorithm parameters was found in the experiment. With wavelength variables selection based on discrete firefly algorithm, succinic acid concentration of the fermentation broth partial least squares NIR calibration model was built, which was compared with genetic algorithm method. The results showed that the correlation coefficient of calibration set (R_c²) of PLS calibration model based on discrete wavelengths firefly algorithm is 0.986, RMSEC of which is 0.409. Correlation coefficient of prediction set (R_p²) is 0.969 while RMSEP is 0.458. It is superior to full spectrum modeling and calibration model using genetic algorithm method. DFA shows superiority of the near-infrared spectral data filtering.