Abstract: The urban inundation disasters caused by extreme rainfall events are becoming increasingly severe.However,numerical models,which are currently one of the main technical means for urban inundation risk analysis and early warning,face challenges in meeting the demands for timely forecasts due to their slow computational speed,particularly in the case of short-duration heavy rainfall events.A pressing issue is how to use AI technology to quickly and accurately predict urban inundation processes.To address this issue,this paper utilizes a high-precision hydrodynamic model accelerated by GPU to simulate inundation results under different rainfall return periods in the study area.These inundation results then drive a GA-BP neural network model to build a rapid urban inundation prediction model.The results indicate that the rapid forecasting model exhibits high prediction accuracy,with the maximum water depth compared to observed data,The RE does not exceed -5.4%,the RMSE is no higher than 0.025,and the NSE is not less than 0.891.The computation speed is approximately 349 times faster than numerical model simulations.This method can meet the daily needs of urban early warning forecasting tasks,enhance the city’s disaster prevention and mitigation capabilities,and effectively reduce losses in terms of human lives and property.