Study On Prediction And Early Warning Of Urban Flood Waterlogging Process Based On Deep Learning

Under the background of global warming and rapid urbanization,frequent urban flood disasters have become a common challenge faced by many cities.Therefore,Shanghai,Beijing,Wuhan,Zhengzhou and other cities have established relatively perfect urban flood monitoring systems to guide urban flood control.However,for urban flood prevention and control,early warning information is more useful than monitoring information.Therefore,it is necessary to use the monitoring data to predict the process of urban flood,especially for the prediction of the process of ponding,which is of great significance to guide the flood forecast and early warning and reduce the loss caused by flood disaster as much as possible.Based on the key project of National Natural Science Foundation of China,"Research on the theory and method of urban flood disaster prediction and early warning based on big data"(No.: 51739009),using the collected rainfall and ponding process data,this paper puts forward a sensitivity index combination scheme suitable for the prediction of waterlogging depth,and constructs the prediction model of ponding process based on the Gradient Boosting Decision Tree(GBDT)algorithm.Based on the rainfall forecast data,the real-time forecast and early warning of water accumulation process in water accumulation point is realized.The main work and achievements are as follows:(1)The spatiotemporal correlation between rainfall and ponding was analyzed.Based on the formation mechanism of urban flood,the spatial and temporal distribution characteristics of rainfall and ponding are analyzed.Based on the spatial autocorrelation theory,the spatial autocorrelation of rainfall and ponding is analyzed by spatial analysis software Geo Da.The spatial autocorrelation of each ponding point is clarified,and the feasibility of constructing the prediction model of ponding process of ponding point is expounded,which provides theoretical support for the construction of prediction model of ponding process of ponding point.(2)The sensitivity index and combination scheme suitable for prediction of water depth are put forward.Guided by the construction of the prediction model of ponding process,the statistical analysis and random combination of the sensitive indexes affecting ponding depth are carried out by using the statistical analysis method;for each index combination scheme,the prediction logic regression model of ponding depth is constructed,and the index combination scheme suitable for ponding depth prediction is determined by taking the average relative error and qualified rate as the index evaluation criteria.(3)Based on the deep learning model,a modeling method suitable for the prediction of long-term(3h)water logging process is proposed.Based on the sensitivity index combination scheme of ponding depth,the precipitation and ponding process data are split and reorganized by isometric splitting and reorganization method.Combined with GBDT algorithm,the ponding process prediction model of ponding point is constructed.By splitting and reorganizing the data of rainfall and ponding process,the prediction of waterlogging process based on non-time series model is realized,which avoids the error accumulation phenomenon of time series model in multi-step prediction,and can complement the advantages of time series model.It provides a new research idea and modeling method for urban flood and waterlogging process prediction.(4)Based on the rainfall forecast data,the real-time forecast and early warning of water accumulation process in water accumulation point is realized.Taking the rainfall event on August 1,2019 as the sample,based on the constructed water accumulation process prediction model,combined with the rainfall forecast data,the hierarchical early warning of the water accumulation process of the ponding point is realized.Through the real-time correction of the rainfall forecast data,the real-time correction of the classification forecast and early warning of the water accumulation point is realized,which provides a new idea and method for the urban flood forecasting and early warning.