Abstract: To extract the characteristic information of total acid and improve model accuracy, simulated annealing algorithm was used to optimize window width and wavenumber in a total acid model obtained by the partial least squares analysis. Ninety different brands of vinegar samples collected from the whole country were used to acquire near infrared spectral data, and total acid content was measure by the general analytical method. Pre-processing methods were used to smooth the vinegar NIR spectra. A total of 17 wavenumbers were selected and the predictive model for total acid model revealed a correlation coefficient of prediction set of 0.9210 and its prediction ability was superior to that of the whole spectral partial least square model and interval partial least square model. These results showed near infrared spectroscopy in combination with simulated annealing algorithm allows to quickly predict total acid content of vinegar.

Key words: vinegar, total acid, near infrared spectroscopy, simulated annealing algorithm, wavenumber selection