Study On Real-time Flood Prediction By Classification In Typical Basins Of The Middle Reaches Of The Yellow River

Under the combined effect of climate change and human activities,the water cycle of the Yellow River Basin has changed significantly.Especially in the middle reaches of the Yellow River.The state of production and confluence is complex.Coupled with the uncertainty of the flood itself.The use of traditional single forecasting methods to forecast floods that is difficult to get ideal forecast results.This article is supported by the "Thirteenth Five-Year" National Science and Technology Support Plan Project "Mechanism and Trend Forecast of Water and Sediment Changes in the Yellow River Basin"(2016YFC040240203).Select the representative Luohe Guxian Reservoir and the control basin of Jingle Station on the upper Fenhe River as typical research objects.To Construct F-K ++ flood classification forecast method,Real-time classification of floods in typical watersheds and based on the classification flood.The Excess Storage and Excess Infiltration Simultaneously Model and the LSTM model are used for forecasting respectively.In order to improve the forecast accuracy and extend the forecast period use the full-extra-saturated compatible model for flood forecasting and LSTM neural network model for real-time correction.The main contents and conclusions of the study are as follows:(1)Construction of F-K ++ flood classification forecast method.In order to solve the shortcomings of current flood classification methods,the process of constructing F-K ++ real-time flood classification method mainly includes three modules: Sample processing module,Cluster analysis module,Realtime flood classification module that can effectively solve the problem of redundancy between multi-dimensional values and the determination of the optimal clustering number k of k-means ++ algorithm.The classification method is applied to the Luohe Guxian Reservoir and the Fenhe Upper Jingle Station to control the watershed.(2)Application of the Excess Storage and Excess Infiltration Simultaneously Model in typical watersheds.Based on the Excess Storage and Excess Infiltration Simultaneously Model,the floods of Guxian Reservoir and Jingle Station controlled watershed classification are forecasted and the simulation accuracy was significantly improved.Compared with unclassified floods,the classified floods have stronger regularity,and the corresponding model parameters can be better suited to flood forecasting under complex conditions in typical watersheds.(3)Application of LSTM neural network model in typical watershed.The LSTM neural network model was used to forecast the floods after the control basin classification of Guxian Reservoir and Jingle Station,and the simulation results of the two models were compared.The correlation coefficients of the simulation results of the two types of floods are all above 0.8,the Nash efficiency coefficients are above 0.7,and the absolute value of the relative error is within 20%.The forecast results have reached the level of B forecast.Among them,the prediction result of the LSTM neural network model is better than the simulation result of the Excess Storage and Excess Infiltration simultaneously,but the prediction period of the LSTM neural network model is shorter.(4)Real-time flood correction forecast based on LSTM neural network.In order to achieve a better forecasting effect,we use the Excess Storage and Excess Infiltration Simultaneously Model for flood forecasting and LSTM neural network model to simulate runoff error sequences as input items for real-time correction.Therefore,the accuracy of the flood forecast after the correction of the control basin of Guxian Reservoir and Jingle Station has been significantly improved.