Abstract:[Objective] The aim of the study is to explore a digital blending method based on spectral information, thereby improving the efficiency of black tea blending and ensuring the stability of the finished tea quality. [Method]'Ying Hong No. 9' black tea was taken as the research subject. Partial least squares regression was employed to compare the effects of different spectral preprocessing methods and feature wavelength selection algorithms on prediction model performance. Quantitative prediction models for the mass fraction of tea polyphenols, the mass fraction of free amino acids, and the mass of water-soluble extractives, as well as the taste scores, were established accordingly. Furthermore, a multi-objective optimization algorithm was employed to transform the tea blending process into a mathematical model that balanced both quality and cost considerations. The NSGA-Ⅱ (non-dominated sorting genetic algorithm Ⅱ) was utilized for global optimization of blending recipes, with manual sensory evaluation conducted to validate the results. [Result] The prediction models for the mass fraction of tea polyphenols, the mass fraction of free amino acids, and the mass of water-soluble extractives, as well as the taste scores, achieved determination coefficients of 0.890, 0.810, 0.802, and 0.863, respectively, in their predictive sets, all outperforming the full-spectrum prediction results. The optimized blending recipe derived from digital blending strategy optimization not only met sensory evaluation standards but also reduced blending costs by 28.5%. [Conclusion]The combination of near-infrared spectroscopy and the multi-objective optimization algorithm enables digitization of the whole process from sensory evaluation to tea blending. Near-infrared spectroscopy effectively captures chemical information related to key flavor compounds and taste quality of black tea. The multi-objective optimization algorithm, together with near-infrared spectroscopy, enables rapid development of optimal recipes, serving as an effective auxiliary method in the tea blending process. The findings hold significant theoretical and practical value in advancing intelligent upgrading in the tea industry.
