In response to the difficulty of balancing the efficiency and accuracy of physical mechanism models in urban waterlogging simulation, a fast and precise simulation method for urban waterlogging based on U-Net super-resolution reconstruction was proposed. By using coarse accuracy waterlogging results as physical priors, a fast mapping of waterlogging distribution from coarse accuracy to fine accuracy was achieved. The Minzhi area of Shenzhen City was selected as the study area. Two sets of hydrodynamic simulation grids, coarse and fine, were constructed, combining terrain and its derived features as multi-channel inputs to train the deep learning model, and the simulation results were compared with the one-dimensional two-dimensional coupled hydrodynamic model. The results show that the super-resolution reconstruction method can accurately restore the spatial pattern and peak characteristics of ponding under most design rainstorm scenarios. The Nash efficiency coefficient generally exceeds 0.9, and the root mean square error is less than 0.15. Compared with the one-dimensional two-dimensional coupled hydrodynamic model, the efficiency of the simulation of the 2h rainstorm process is 50.71 times higher. With the increase of rainstorm return period, the super-resolution reconstruction method is more stable for the reconstruction of the overall ponding characteristics, but there is still a deviation in the local low-lying areas and areas with complex pipe network.