The evolution process of urban waterlogging under different levels of rainfall was simulated using a hydrodynamic model, and then the waterlogging evolution process data was used as the training set of the K-nearest neighbor algorithm machine learning model for model training. The rainfall predicted by the atmospheric numerical model was used to drive the trained K-nearest neighbor machine learning model for rapid urban waterlogging forecasting. Taking Qinhan New City in Shaanxi Province as an example, the predictive performance of the model was tested through three measured rainfall events. The results show that the model can quickly predict urban waterlogging within 17 seconds, with an average error of no more than 8% for the predicted waterlogging area, and no more than 15% for the average error of waterlogging amount and depth. The model has good predictive performance and can enhance urban disaster prevention and reduction capabilities, effectively reduce life and property losses.