To resolve the contradiction between massive monitoring data of twin models and high-fidelity rapid response, a rapid calculation method for dam deformation based on a POD-RBFNN reduced-order model was proposed taking an arch dam as an example. Based on static response data under different water level conditions obtained through finite element simulations, dominant modes and modal coefficients were extracted using the proper orthogonal decomposition (POD) method. Combined with a radial basis function neural network (RBFNN), a high-accuracy surrogate mapping between water level parameters and modal coefficients was established, achieving rapid and accurate prediction of the static displacement field of the arch dam. Calculation results show that under the condition of considering only hydrostatic pressure and structural self-weight, the prediction results of the POD-RBFNN reduced-order model agree well with finite element (FEM) simulation results, with an average relative error of 0.35% and an average absolute error of only 0.003 2 mm. Meanwhile, the calculation efficiency is improved by 3 004 times compared with the finite element simulation.