Abstract: We address the challenge of adapting to external factors and fluctuations in incoming material characteristics in leaf-wetting machines, which traditionally rely on manual experience to regulate control parameters. Our approach predicts tobacco quality prediction and adaptively optimizes process parameters. First, we identify key control parameters and factors affecting their regulation by systematically analyzing the process flow. We then apply a Bayesian optimization extreme gradient boosting tree algorithm to model the relationship between process parameters and the moisture content and temperature of exported tobacco. Finally, using the standard quality of exported tobacco as our optimization objective, we determine the global optimal solution. To stabilize tobacco quality after wetting and improve the leaf-wetting machine's operation, we introduce an improved gray wolf algorithm with bounded stability and an adaptive penalty function. This approach accelerates convergence speed and reduces control parameter fluctuations. Experimental results show that our method reduces the moisture content and temperature fluctuation ranges of exported tobacco by 42.5% and 29.9%, respectively, compared to manual adjustments, ensuring smoother operation of the leaf-wetting machine.