Research On Shanghai Flood Simulation And Forecasting Coupling CAESAR-Lisflood And CNN

Currently,flooding has become a serious threat to human society and economic development.With the increase of global flood-affected areas and population,it is estimated that by 2030,more than 700 million people will be affected.Due to the advantageous location and population concentration of coastal cities,they are more vulnerable to flooding than inland cities.Global warming has led to rising sea levels and an increase in extreme weather events,further exacerbating the flood risk in coastal areas.Therefore,accurate simulation and rapid prediction of coastal city flooding are crucial.Foreign scholars have extensively studied the application of hydrological and hydraulic models in simulating flooded areas and depths in inland regions.However,due to the limitations of the models,the application of rapid response and prediction of flood inundation under different scenarios is restricted.In recent years,deep learning has received attention in flood simulation,but its training requires a comprehensive and accurate flood dataset,which poses limitations for further application.Therefore,how to combine hydraulic models with AI-based deep learning models to carry out accurate and fast flood simulation and prediction is of great significance for flood control and disaster reduction in coastal cities.In this study,Shanghai’s main urban area was taken as the research area to conduct coupled hydraulic and deep learning-based coastal city flood simulation and prediction research,and a flood prediction model based on deep learning network was built.The specific research content and achievements are as follows:(1)Conduct flood simulation based on the CAESAR-Lisflood model.By comparing with the classic hydrodynamic model LISFLOOD-FP,the accuracy of the CAESAR-Lisflood model in simulating floods in coastal cities is studied.The study found that the CAESAR-Lisflood model has high accuracy and reliability in flood simulation in coastal areas,and its calculation speed is relatively fast.Therefore,in practical applications,this model can be used to simulate the flood range and depth,and provide data support for CNN-based flood prediction models.(2)Conduct research on the prediction of coastal urban floods by coupling CAESAR-Lisflood and CNN.A CNN-based coastal urban flood prediction model is established,and automatic hyperparameter optimization is carried out using Bayesian optimization method to improve the generalization ability of the flood prediction model and reduce application difficulty.Using the Huangpu River water level data as input,the CAESAR-Lisflood model is used to simulate floods multiple times to generate a large amount of flood data,which drives the CNN-based coastal urban flood prediction model to predict the flood situation in coastal cities based on water level data,realizing rapid prediction of coastal urban flood submergence.(3)A CNN-based coastal urban flood prediction model is used to conduct flood prediction experiments with different regularization methods,with the aim of finding the most suitable regularization method for the model.The experimental results show that the model with the Early stopping regularization method performs the best,with the highest prediction accuracy and efficiency.