Mid-Long Term Runoff Forecast Models And Their Applications

Mid-long term runoff forecast has much great significance on obtaining future runoff information and realizing scientific management and optimal scheduling of water resources.Coupling forecast models and comprehensive evaluation of mid-long term runoff forecast models can provide important reference for the applications and improve hydrological stations forecast accuracy.This paper introduced the principles and models of mid-long term runoff forecast,analyzed and summarized the current problems existing in the research progresses.We took the natural annual runoff series of 5 hydrological stations as examples,such as Minhe,Lanzhou,Longmen,Baimasi and Heishiguan station in the Yellow River basin.Mean absolute error(MAE),mean relative error(MRE),root mean square error(RMSE)and mean square percentage error(MSPE)were selected as error evaluation criteria.The information entropy was used to determine the weight of each error index,and the evaluation values of the runoff forecasting model were obtained and used to assess the comprehensive evaluation of mid-long term runoff models.The research content and conclusions mainly includes three aspects:(1)Single runoff forecast models for hydrological stations were constructed in the study area.The periodic analysis and time series forecast models for mid-long term runoff forecast were used to establish single forecast models,such as periodic epitaxial superposition model,autoregressive(AR)model,mean generating function model and Markov model.In single models comprehensive evaluation values set,comprehensive evaluation results show that the maximum evaluation values of the mean generating function model of Minhe and Baimasi hydrological stations are 0.963 and 1.000 respectively.AR model of Lanzhou,Longmen and Heishiguan hydrological stations gets the maximum values 0.966,0.989 and 1.000,respectively.When the Markov model was used for qualitative forecast,the model accuracy and reliability are low.(2)Coupled runoff forecast models for hydrological stations were constructed in the study area.According to the merit of the single forecast model,we employed the coupled models to improve the forecast accuracy.The coupled models include gray-periodic epitaxial superposition model,mean generating function-stepwise regression model,weighted Markov model,grey-stepwise regression period model,EEMD-BP neural network model,PPRGSRP model,PPARWD model and fuzzy comprehensive analysis model.In the coupled model comprehensive evaluation values set,the PPARWD model of Minhe,the EEMD-BP model of Lanzhou and Heishiguan get the evaluation value of 1.000.The PPRGSRP model of Longmen gets evaluation value 0.962,and the fuzzy comprehensive analysis model in Baimasi is 0.967.Applied the weighted Markov model to ovecome the disadvantage that the maximum transition probability of the Markov model unique,and the reliability and accuracy of the model are improved.(3)Optimal models of runoff forecast model for hydrological stations were given in the study area.Combining the evaluation value of the single forecast model and the coupled forecast model,the assessment values of all models at each site were calculated.The results show that the optimal model of Minhe hydrological stations is PPARWD model,Lanzhou and Heishiguan hydrological station is EEMD-BP model,Longmen hydrological station is PPRGSRP model,Baimasi hydrological station is fuzzy comprehensive analysis model.