Study On The Early Warming Of Flash Floods Based On Rainfall Pattern Clustering And Identification

Flash floods has brought great losses to the safety of life and property of the people in China,so the early warning of mountain flash floods is the focus of disaster prevention and mitigation in China.At present,the critical rainfall is the most commonly used index in the forewarning and forecasting of flash floods in China.When determining the critical rainfall,only the design rainfall pattern of the region is used to allocate the rainfall time schedule,which will ignore the uncertainty of the critical rainfall caused by the randomness of rainfall,and then lead to the occurrence of adverse phenomena in the early warning.In order to solve the uncertainty of critical rainfall caused by the randomness of rainfall,a precipitation pattern generation method based on clustering analysis was proposed in this paper through the statistical analysis of historical rainfall characteristics.In addition,combining with meteorological mining technology,the rainfall prediction model based on random forest and BP neural network is proposed,and the flash floods warning model based on rainfall pattern identification is constructed,which provides a new idea for flash floods innovation warning model.The main research contents and results are as follows:(1)Analysis of the influence of rainfall pattern on the uncertainty of critical rainfall.Through peak shaving of the designed rainfall pattern in the study area,different rain pattern scenarios are constructed.The HEC-HMS hydrological model is used to simulate the flood situation and the corresponding critical rainfall under different rain pattern scenarios.The results show that in three typical disaster prevention objects,the deviation of critical rainfall caused by rainfall tpattern is more than 20%.(2)Rainfall pattern determination method based on cluster analysis.Through the statistics of historical rainfall in the small watershed,the rainfall database is constructed and the characteristics of the rainfall process are analyzed.Combining the characteristics of runoff generation and concentration in the basin,determine the duration of rainfall patterns and construct a rainfall information matrix.K-means cluster analysis is used to cluster the rainfall information matrix,and the cluster rain pattern set that appears most frequently in the historical rainfall process in the study area is determined,and the critical rainfall threshold space of three typical disaster prevention objects is calculated.The results show that the number of rainfall patterns in three typical small watersheds is 5,5 and 4,respectively,and the corresponding critical rainfall threshold space is 73～140mm,82～154mm and 56～114mm,respectively.(3)Rainfall prediction based on random forest and BP neural network.Random forest is used to screen meteorological data to remove interference data and retain the most important meteorological elements for rainfall.Taking relative humidity,maximum wind speed,minimum air temperature,minimum air pressure and water vapor pressure as the main meteorological elements of rainfall prediction,the BP neural network was used to construct the qualitative and quantitative rainfall prediction models to predict the future short-term rainfall sequence.The results show that the accuracy of the model is more than 70%.(4)Early warning mode of flash floods based on rainfall pattern recognition.According to the predicted future short-term rainfall sequence,the cluster rainfall pattern with the strongest correlation is identified,and the corresponding critical rainfall is determined.At the same time,the possible total rainfall in the future is determined based on the analysis of real-time rainfall information,and the corresponding early warning information is issued by comparing the critical rainfall of the identified rainfall pattern with the total rainfall.Taking three rainfalls of20160701,20140719 and 20150806 as examples,the results show that this warning model can well improve the forecast period of the early warming of flash floods,and provides a new idea for mountain flood disaster warning and prediction in small watershed of hilly area.