| To address the shortcomings of current dam seismic dynamic response analysis systems, which often neglect the impact of construction quality on the physical and mechanical parameters of dam materials and struggle with rapid analysis and visualization of the temporal and spatial distribution of dynamic responses, this study proposes a method for constructing a spatiotemporal analysis system for seismic dynamic response under the digital twin framework. This approach leverages the strengths of digital twins, such as virtual-physical mapping and rapid analysis. The system's data foundation integrates high-fidelity geometric models, material parameters of the dam body, and seismic motion information through parameterized modeling, intelligent monitoring of construction quality, and virtual sensors. A novel non-homogeneity analysis model for dam material parameters based on transfer learning and ResNet is introduced to accurately analyze the spatial distribution of material parameters across the entire dam, considering the effects of construction quality. For specialized modeling, a pix2pixHD-based surrogate model for spatiotemporal analysis of dam seismic response is developed. This model uses multi-layer generative adversarial networks to learn the temporal distribution characteristics of seismic response data and employs local enhancers to improve the authenticity of spatial distribution features in the analysis results, thereby overcoming the limitations of existing surrogate models that can only predict a few points and fail to characterize the overall spatiotemporal distribution of seismic dynamic responses. In terms of system visualization, a CUDA-accelerated 3D physical field calculation method is proposed for visualizing dam seismic dynamic responses, addressing the issue of using pre-made animations for earthquake process visualization in current research. Finally, the study develops a dam seismic analysis system within the digital twin framework, exemplified by an earth-rockfill dam project in southwestern China. The system achieves rapid analysis and visualization of the entire dam's seismic dynamic response in just 148 ms, providing robust support for seismic safety assessments and decision-making for dams. |