Research On Urban Flooding Depth Prediction Based On Deep Learnin

With the continuous intensification of global climate change and the rapid development of urbanization,the urban waterlogging disaster caused by extreme rainfall process is becoming more and more serious,which has brought huge losses to the people’s life and property safety and social and economic development,and has become a serious challenge for many cities around the world.It is of great significance to predict the depth of urban waterlogging.Traditional physically-driven hydrological and hydrodynamic models have problems such as low computational efficiency and complex modeling due to the need for a large number of high-precision and difficult to obtain drainage network information and elevation data.With the development of machine learning,the deep learning method has been widely used in the research of urban water depth prediction,and has achieved excellent prediction results.Among them,the long and short term memory network LSTM(Long Short Term Memory)has been well applied with its unique performance.The Random forest(RF)and Artificial neural network(ANN)are also popular prediction methods.Therefore,this study takes Zhuji City,Zhejiang Province as the research area,based on the rainfall data of 75 national automatic meteorological observation stations and the water depth data of five typical waterlogging points in the area from May to August 2021,selects the rainfall of rainfall observation stations with strong correlation with the waterlogging depth of each waterlogging point through correlation analysis,and uses the deep learning method,short-term and short-term memory network LSTM,to carry out the waterlogging depth prediction research for each independent waterlogging point in the next 2 hours with an interval of 15 minutes,Random forest RF and artificial neural network ANN are used to compare the prediction results.The main work and achievements of this paper are summarized as follows:(1)Analyze the waterlogging situation of each waterlogging point and the impact of rainfall on the waterlogging,and select the rainfall of the rainfall measuring station related to the waterlogging depth of each waterlogging point through the correlation analysis to carry out the research into the model.The results show that the situation of water accumulation at each water accumulation point is different.Short-term heavy rainfall and continuous rainfall will cause high value water accumulation.Rainfall stations with strong correlation with water accumulation at each water accumulation point are mainly located near the water accumulation point.With the increase of accumulated rainfall,the correlation coefficient between accumulated water and rainfall becomes larger and stronger.(2)The relationship model between rainfall and waterlogging depth is built by using the deep learning model short and long term memory network LSTM to provide the urban waterlogging water level forecast within 2 hours with an interval of 15 minutes in the future.The prediction results show that the optimal data input length of each waterlogging point is different,2h and 4h.Under the optimal data input length,the prediction results of LSTM model for each waterlogging point are RMSE within 5.6cm,CC above 0.93,and NSE above0.86,with good prediction effect.(3)The relationship model between rainfall and waterlogging depth is built using random forest RF and artificial neural network ANN model,and the prediction performance of each model is compared with LSTM prediction results.The results show that the prediction effects of RF and ANN models on the two waterlogging points are not different.RMSE is within 7.6cm,CC is above 0.88,NSE is above 0.27.There is a big gap compared with LSTM,and LSTM has higher prediction skills.Moreover,the weak reporting phenomenon of RF forecast results is relatively serious,which is quite different from the true value,and the ANN forecast effect is relatively unstable.