Abstract: The efficacy of conventional flood risk assessment methods is curtailed by extensive computational requirements, insufficient data, and difficulty in adapting to terrain changes, thereby it is urgent to quickly model and analyze flood in large scenarios. This research delineates an innovative technique that amalgamates expansive LiDAR point clouds segmentation with deep learning to expedite flood risk simulation. Our comprehensive procedural framework is comprised of data acquisition and preprocessing, sophisticated point cloud segmentation, Digital Elevation Model (DEM) reconstruction, and hydrodynamic simulation. It has been applied specifically to model flood scenarios within a designated section of China′s South-to-North Water Diversion Project. The empirical results underscore the proficiency of this method, with an mean Intersection over Union reaching 70.8% and an overall classification accuracy attaining 88.7% for the extraction of intrinsic terrain features. The method accurately projects maximum flood inundation extents of 7.0×10⁴ m² and 10.5×10⁴ m² for the respective design and check flood simulation scenarios. This approach provides a paradigm shift in rapid flood risk assessment, markedly advancing the modeling efficiency and analysis precision in flood risk management.