考虑时空特征的城市内涝智能预报模型研究OA北大核心CSTPCD

Given the problems of the traditional urban waterlogging forecast model,such as being time-consuming,few meas-ured waterlogging samples,and insufficient consideration of waterlogging characteristic factors,an urban waterlogging mechanism model was first built by coupling the SWMM model and the LISFLOOD-FP model.The mechanism model was used to numerical-ly simulate the designed rainstorm in different recurrence periods to generate waterlogging samples.Based on samples and water-logging characteristic factors,a three-dimensional spatio-temporal matrix was constructed to realize the orderly organization of waterlogging characteristic factor data.Based on the above,a convolutional neural network(CNN)was coupled with long short term memory network(LSTM),and an urban waterlogging intelligent forecast model considering multi-temporal characteristics(CNN-LSTM)was constructed.The intelligent model was trained by a three-dimensional space-time matrix using measured samples from Tianhe District,Guangzhou City.The results show that the CNN-LSTM model can quickly predict the inundation depth and inundation range.The Nash coefficient of waterlogging control point water level simulation was above 0.9,and the aver-age matching rate of the inundated area at every moment reached 92.2%.Compared with the mechanism model,the simulation ef-ficiency was improved by nearly 70 times.The intelligent model had good forecasting accuracy and efficiency,and could effectively support the work of urban waterlogging prevention and disaster reduction.