Abstract: To address the issues of structural imbalance in the supply of high-quality tobacco leaves and the highly coupled inventory demand to enhance digital design and maintenance capabilities of cigarette blending formulas, a dataset including 482 tobacco leaves in the inventory, with 23 cigarette brands and a 3-year production plan was built, and a quarterly rolling calculation process was constructed. By integrating near-infrared spectroscopy, sensory experience of formula designers and computer-aided programming, a two-level similarity substitution full-chain cigarette blending formula maintenance approach was established, which combined the near-infrared-tobacco similarity algorithm (SSD-2D-CORR) and the formula designers' experience-formula similarity algorithm (PLSR-AHP-PCA-MD). The results showed that: 1) Taking the TB2 brand cigarettes as an example, the differences in various quality indicators between the five new formulas and the original formula were all relatively small, meeting the requirements of the original formula. 2) Under the overall substitution calculation method, the independent control limits of various indicators were relatively stable, significantly better than using the individual substitution calculation method. Through the comprehensive control limits, the independent control limits of various indicators were represented as a whole, and the difficulty for formulators to judge the stability of cigarette quality during maintenance could be reduced. 3) This method achieved automatic calculation for blending formula maintenance in the scenarios of full inventory, all brands, long cycle, and highly-coupling level, retaining global spectral information while integrating the brand experience of the formulator. This study provides an effective digital design solution for tobacco leaf procurement, inventory configuration, product maintenance and cost control.