Research On The Data-driven Flood Simulation And System In Weihe River Basin

Flood disasters have a serious impact on the natural environment,social economy and the safety of people’s life and property in China.In 2022,five of top ten natural disaster events in China are directly related to rainfall and flood disasters.Based on hydrological models and machine learning methods to improve the accuracy of river basin flood simulation,the information technology is used to improve simulation efficiency and provide reference for river basin flood control decision.In this paper,both SWAT and LSTM model are used to simulate floods in the Weihe River Basin.A flood simulation system is designed and developed to visualize the simulation results and provide reference for flood management in the river basin.The main findings are as follows:(1)The characteristics of flood variability in the WRB are revealed.Nine flood indicators were selected using the annual maximum method and the super-threshold peak volume model,and the trends and variation characteristics of the flood indicators were analyzed based on the daily flow data of six typical hydrological stations in the WRB from 1990 to 2020.The results show that the WRB experienced fewer floods in the mid to late 1990s.Floods were more frequent during 1990-1993,2003-2005 and 2011-2013,with the highest frequency of flood disasters occurring in the middle and lower reaches.(2)A river basin flood simulation model was established based on SWAT model.The Nash coefficient NSE,deterministic coefficient R2 and relative error Re were selected as evaluation indicators to rate and validate the model parameters.And SWAT model was applied to the WRB.The results show that there are six runoff related parameters,the most sensitive of which is the runoff curve number CN2.R2 is greater than 0.60 and NSE is greater than 0.65 for each station.The model accuracy meets the requirements.The flood pass rates of Xianyang and Huaxian stations are both grade B,and Weijiabao station reaches grade C.The SWAT model has good applicability in the WRB.(3)A machine learning-based flood simulation model for the WRB was developed.A river basin LSTM flood simulation model was constructed using machine learning methods.The model was trained based on a data driven approach.The validated model was also applied to the flood simulation of the WRB.The results show that compared with the SWAT model,the R2 and NSE of the daily runoff simulation based on the LSTM model are greater than 0.70.The flood pass rate of all stations reaches the grade A standard,and the LSTM model has higher simulation accuracy.(4)The WRB Flood Simulation System was designed.A flood simulation system for the WRB has been built based on a comprehensive and integrated platform using technologies such as components and knowledge visualization.The system provides functions such as querying flood information,feature analysis,model simulation and results presentation in the WRB.The results show that the system construction process is easy and flexible.The efficiency of the model simulation has been improved.The simulation results are reliable and can be personalized for the user.