| Hydrologic forecasting is an important means for water conservancy workers to predict natural disasters such as floods and droughts,and take measures in advance.Runoff forecasting can provide detailed information about future runoff,and has an important indicator role.However,most of the current physical hydrological models have poor portability,and the prediction results are often unsatisfactory due to various factors.Data driven model is a hot topic in recent years,but its main logic is to explore the mathematical laws in historical data,often neglecting the hidden physical laws in runoff forecast,resulting in unclear interpretation of the physical mechanism of runoff generation,and poor interpretability of the model.In view of this,this thesis uses the spatial characteristics of the hydrological stations and reservoir stations in the basin,combined with the upstream station influence,water intake,rainfall and other factors that affect the station runoff,to extract the spatial structure characteristics of the basin with GCN,and then uses GRU to learn the spatial characteristics of the station runoff in the basin,to forecast the runoff of Ankang,Huangjiagang and Huangzhuang stations in the Han River basin in different length of the forecast period.On the basis of understanding the runoff characteristics,rainfall,water intake and other characteristics of each station in the Hanjiang River basin,the runoff prediction accuracy under different forecast periods has been improved.The main conclusions are as follows:(1)The modified Wetspa model has good applicability in the Hanjiang River basin.When the Wetspa model is used to model Ankang Reservoir,Danjiangkou Reservoir and Huangzhuang Reservoir,the rate of the three sections and the verified Nash efficiency coefficient are all above 0.7,and the absolute value of the relative error is less than 20%.When the runoff of three typical sections of the Hanjiang River basin in 2020 is simulated with the model with calibrated parameters,the model has a good simulation effect on the flood peaks of Ankang and Huangzhuang stations,with NSE above 0.55.The simulation effect in the overall wet season is better than that in the dry season,the simulation result in the dry season is slightly smaller than the measured value,and the overall simulation result of Huangjiagang station is smaller.(2)When the prediction period of LSTM is one day,the simulation results of LSTM in the three sections of Hanjiang River are all good.The NSE of the three sections is above0.85,and the simulation effect is not different in the wet/dry seasons.When the prediction period is two days,the results will be slightly worse,but the NSE can still reach 0.81,0.96 and 0.90.However,when the prediction period exceeds two days,there will be significant deviation,and the simulation results will lag behind the measured values,And with the extension of the forecast period,the offset will become more and more serious.(3)The overall accuracy of the prediction results of the coupled model in different forecast periods is high.Even in the 7-day forecast period,the NSE value of the three sections can still be more than 0.7,and the CC value can still be more than 0.7,and the MAE,RMSE and RB are less than the LSTM prediction results in the same forecast period. |
