Reactors often exhibit high vibration intensity due to the alternating stacking of iron core cakes and air-gap pads. To address the limitations that existing vibration reduction measures based on dimensional optimization neglect in-situ replacement requirements and lack mechanistic explanations for vibration mitigation

this paper proposes a reactor vibration suppression optimization method in which electromagnetic-structural coupling modeling is integrated with intelligent prediction algorithms

and presents the interpretability analysis of the predictive model. First

after considering the anisotropic magnetostrictive characteristics of silicon steel sheets

a parametric vibration calculation model incorporating segmented iron core-air gap structures is established

identifying iron core cake height and air gap height as key design variables. Second

a Latin hypercube n−1 sampling strategy is employed to circumvent equality constraints

and a dataset is constructed on the basis of computational results. Subsequently

the sparrow search algorithm is adopted to optimize hyperparameters of the extreme gradient boosting (XGBoost) model and to develop a high-precision vibration prediction model

meanwhile

the segmented iron core column reconfiguration is achieved under constraints via the sparrow search algorithm. Finally

leveraging Shapley additive explanation (SHAP) attribution theory

the contribution degrees of feature variables to vibration displacement are quantitatively analyzed

with simulation results elucidating the influence mechanisms of iron core cake height and air gap height on vibration. Prototype testing shows the optimized reactor reduces vibration displacement amplitude by 12.1% without changing overall dimensions

guiding vibration-optimized reactor design for installation compatibility.