Abstract: In order to be able to quickly provide timely peak ground acceleration (PGA) data support for dam safety assessment after an earthquake, and to improve the efficiency of earthquake emergency response. In this study, a PGA prediction model for dam seismic motion is constructed based on the adaptive neuro-fuzzy inference system (ANFIS), which integrates the seismic source parameters (including magnitude M, azimuth θ, and depth H), the propagation path parameter (epicentral distance R), and the site response characteristics (PGA at the measurement points), and forms a multidimensional nonlinear prediction framework. Based on the measured data of 500 seismic events recorded within 350 km around a hydropower plant, the model prediction performance of 23 methods, including ANFIS, random forest, extreme random tree, and ridge regression, was compared and analyzed. The results show that the ANFIS model has a coefficient of determination (R²) of 0.827 on the test data, which is outstanding compared with other prediction models in terms of prediction effect, has a high training efficiency, and demonstrates a better generalization ability. The ANFIS model has higher prediction accuracy and excellent stability in predicting the strong vibration PGA of dams, and can be used as an effective method for the rapid assessment of the structural safety of dams.