Research On The Prediction Method Of Urban Rainstorm Flood Depth Based On Deep Learning

With climate warming and accelerated urbanization,the frequency of extreme rainfall events has increased.Rainstorm flood has become one of the main natural disasters plaguing many cities in China,and has brought serious threats to human production,life and socio-economic activities.As an effective "non-engineering measure" to control urban flood,urban rainstorm flood forecast can win more emergency response time for urban flood prevention and mitigation.At present,the methods and models of urban rainstorm flood prediction are still in the process of construction and improvement,and a wealth of research results have been formed.Although the existing physical-driven urban rainstorm flood model can reflect the characteristics of urban flood in a certain period of time,there are still problems such as low calculation efficiency of the model,and short forecasting period.In view of this,this thesis focused on the prediction of urban flood and inundation depth from the perspective of geography.It summarized the existing urban rainstorm flood models and analyzed its shortcomings and analyzed the influencing factors of flood depth,then analyzed the characteristics of urban storm flood simulation with deep learning,generated the depth prediction feature factors for temporal and spatial prediction of flood depth,including rainfall,pipeline network density,slope,road network density,water system density and land-use type.Combining CNN and LSTM deep learning framework,a data-driven urban rainstorm flood depth prediction model was constructed,and the prediction and verification of urban rainstorm flood depth based on CNN-LSTM was realized.The results showed that the goodness of fit between the predicted water depth in the next hour and the real water depth was about 0.76.With the prolongation of the forecast period,the prediction accuracy of the model will gradually decrease.The main research content and results of this thesis are as follows:(1)This thesis summarized and analyzed the existing urban storm flood models and then analyzed the influencing factors of urban stormwater flood depth,sorted out the characteristics of urban rainstorm and flood simulation taking into account deep learning,analyzed the characteristic factors of depth prediction from multi-source hydrometeorology data.On this basis,the generation method of water depth prediction characteristic factors based on GTWR was designed to solve the correlation degree of feature factors and water depth.(2)This thesis clarified the flood time series data and its characteristics,then processed the water depth prediction feature factors,flood characteristics,and time labels into urban rainstorm flood time series data as the input data of the prediction model,Combining model input and output features and water depth prediction feature factors,and taking full advantage of CNN and LSTM,a neural network structure combining CNN and LSTM was designed,thereby established a data-driven urban rainstorm water depth prediction model.(3)This thesis took some typical areas of Nanjing as the research area,collected and organized experimental data and completed the data preprocessing.Based on GTWR,the final explanatory variables of the predictive model were determined,combined with the training and optimization of the model,the optimal hyperparameters were obtained,and the optimal Data-driven urban rainstorm flood depth prediction model was established.By substituting multiple historical data for learning,the urban rainstorm water depth prediction based on CNN-LSTM was realized.A number of model evaluation indicators were introduced,and the actual observation depth in the test set data was compared and analyzed with the model prediction depth,which verified the scientificity and rationality of the urban rainstorm flood depth prediction method based on deep learning and data-driven in this thesis.