In response to the problem of overly cumbersome and complex calibration process of urban rainstorm flood models, a BIC-KMeans algorithm was constructed by coupling Bayesian information criterion (BIC) and K-means clustering machine learning algorithm (KMeans). According to the distribution law of sample parameters in different urban functional areas, combined with the storm water management model(SWMM), an urban rainstorm flood rapid simulation method was proposed. Six, three, and four historical observation flood events were selected for verification in the main campus of Zhengzhou University, the southern part of Jinshui District in Zhengzhou City, and the central urban area of Zhengzhou. The results show that the proppsed urban rainstorm flood rapid simulation method has good applicability. The values of uncertainty parameters such as depression storage capacity, surface Manning coefficient, infiltration rate, and attenuation coefficient vary from small to large in different urban functional areas, with the order of industrial and commercial areas, residential areas, and public use areas. The relative error of simulated flood discharge in the three catchment areas is less than 12%, the Nash efficiency coefficient is greater than 0.75, and the determination coefficient is greater than 0.80. All indicators are better than traditional parameter adjustment methods. The simulation accuracy of SWMM will decrease with the increase of the spatial scale of the catchment area. The peak time error is on the minute level in small-scale rainwater and flood simulation, and on the hour level in larger scale.