Research on Urban Storm Flood Simulation by Coupling K-means Machine Learning Algorithm and GIS Spatial Analysis Technology into SWMM Model

Accurate flood simulation has significant practical implications for urban flood management. The focus of this study is to develop a new flood model (K-SWMMG) based on the Storm Water Management Model (SWMM), which innovatively couples the K-means clustering machine learning algorithm and GIS spatial analysis techniques. The K-means clustering machine learning algorithm is used to determine the uncertain parameters of the SWMM model, while GIS spatial analysis techniques enhance the two-dimensional realism of flood simulation. We applied the K-SWMMG model to six historical observed flood events in a specific catchment area in Zhengzhou City, using rainfall and flow data. The study shows that: 1) K-SWMMG optimizes the sub-basin division method of urban stormwater models, avoiding the tedious and complex parameter calibration process, and improving modeling efficiency to some extent. 2) The two-dimensional visualization of inundation provided by GIS spatial analysis techniques better meets the production requirements of current urban flood simulation. 3) K-SWMMG outperforms SWMM in terms of simulation performance, with improvements in absolute error (AE), relative error (RE), Nash-Sutcliffe efficiency coefficient (NSE), and coefficient of determination (R²) by 0.019m, 5.36%, 0.068, and 0.042, respectively. The findings can provide scientific decision-making references for urban flood forecasting and early warning.