Study On Mid-long Term Hydrological Forecasting Model Of Reservoir

Mid-long term precipitation and runoff forecast is a significant part of hydrological forecast which is vital to flood regulation and effective water resources utilization. However,with long forecast period, the accuracy of mid-long term forecast failed to satisfy the need of production department, restricted by objective cognitive ability, technology development and many uncertain factors. Large numbers of studies have verified that the rational selection of predictors is the key to improve prediction precision of mid-long term precipitation and runoff forecast.Therefore, the mid-long term hydrological forecast research should focus on establishing forecast model by identifying predictors according to the synoptic background and physical mechanism controlling precipitation and runoff.A review of theoretic researches and practice on mid-long term hydrological forecast both domestically and abroad was drawn in this paper. Then, some key problems of mid-long term hydrological forecast, such as mid-term rainfall prediction, long-term runoff prediction, Bayesian probabilistic forecast and monthly water balance runoff model were investigated. The Danjiangkou Reservoir basin,which is the waterhead area of middle of South-to-North Transfer Project, is selected as the study object.The atmospheric circulation systems and sea surface temperature influencing the rainfall in China and the Danjiangkou Reservoir basin were firstly analyzed in this paper, in order to select the key factors of rainfall and runoff forecast. Subsequently, forecast models were established with multiple regression, stepwise regression and artificial neural network respectively.At the same time, the combination prediction model was built by integration of the result of all the models aforementioned. In order to test the applicability of these models, ten-day rainfall and monthly runoff both from 2001 to 2006 in Danjiangkou Reservoir basin were predicted by simulation based on the data of ten-day areal rainfall from 1980 to 2000 and monthly runoff from 1968 to 2000 in the basin. Results showed that all of these models can attain a promising accuracy, in which the combination prediction model is recommended as forecast model with the highest accuracy.To further investigate the uncertainty of mid-long term forecast, a Bayesian probabilistic forecast model was established. The predicated value range with a confidence level is provided.Finally, a three-parameter monthly water balance model for the Danjiangkou Reservoir basin was proposed. This model was based on the hydrothermal equilibrium equation and the relationship among water storage, runoff, rainfall and evaporation. Results showed that the three-parameter monthly balance model is feasible for simulating the runoff processes of the whole basin and different sub-basin.