Study On Mid-long Term Runoff Forecasting Method For Dongjiang Reservoir

The importance of water for human is unquestionable,but with the impacts of climate change and human activities,water is facing unprecedented challenges.In order to deal with water crisis,it is necessary to strengthen the research of hydrological forecasting,not only requiring correct short-term forecasts,but also asking for long-term forecasts.Due to the large number of predictor of mid-long term hydrological forecasting and the complex relationship between influencing factors,the physical mechanism is still not clear.At present,most of the practical applications use statistical models to explore the evolution of the forecast object itself or the quantitative relationships between influencing factors and forecasting objects.Predictor selection and model construction are two key issues in mid-long term runoff forecasting.This paper is also mainly focus on these two aspects.Firstly,the theoretical principle of runoff time series is introduced.Based on the analysis of basic composition of runoff time series,the main methods of runoff time series characteristics analyzing are introduced from three aspects: trend,periodicity and randomness.Meanwhile,the application of stepwise regression method extracting trend function and periodic function is also included.Secondly,this paper analyzes the influencing factors of mid-long term runoff forecasting according to physical mechanism and the statistical analysis method of selecting effective predictor from the predictor set.Then the time series model,the linear regression model,the neural network model,combination forecasting model theory and accuracy evaluation method are introduced.Finally,the model is applied to Dongjiang reservoir.This paper analyzes the periodicity and trend of the monthly runoff series of Dongjiang reservoir and extracts the periodic function and trend function.Through rank correlation analysis and stepwise regression analysis,effective predictor is selected from the runoff and 74 climatic indices;The time series model,BP neural network model,stepwise regression model and combined forecasting model are established,and the accuracy of each model is evaluated by using the deterministic coefficient and the qualified rate.The results show that the time series model with the highest accuracy,the qualified rate of the simulation period is over 95%,and the deterministic coefficient is above 0.9,which meets the accuracy of grade A.Therefore,it is recommended as the long runoff forecast model of Dongjiang reservoir.The stepwise regression model considering the hydro-climatic teleconnection is more accurate than considering the runoff autocorrelation,which shows that the introduction of meteorological factors can improve the prediction accuracy.The BP neural network model is slightly better than the stepwise regression model in consideration of accuracy.The combined model considers both advantages and disadvantages of the stepwise regression model and the neural network model,as a result,it obtains the better forecast results.