Study On The Method And Application Of Runoff Classification Ensemble Forecasting

With the aggravation of the water resources scarcity and frequently flood disaster, it's urgent to take certain measures for utilizing limited water resources and improving the efficiency of the hydropower with the current hydraulic engineering under the safety of flood control. Hydrological prediction and optimal operation is one of the effective solutions for flood protection and can improve the benefit for hydropower station. Therefore, based on the Huanren hydropower station and Ertan hydropower station, some key problems for the runoff prediction and optimal operation of the hydropower station are studied. In this dissertation, firstly, the ensemble prediction model by stages (EPMS) is developed for medium-long term hydrological prediction, the model developed utilizes the advantages of various traditional and intelligent methods for runoff prediction, and then, based on the different runoff description, a new method for hydropower station optimal operation and risk analysis is presented. As for the flood forecasting, much attention have been put on the uncertainty of the hydrological model, therefore, take the Xinanjiang model as an example, this dissertation studies the uncertainty resulting from the optimal selection, model parameters calibration and model input error. Further, the method for flood ensemble forecasting based on the fuzzy classification is developed. The major research work is outlined as follows:(1) The ensemble prediction model by stages is presented for medium-long term hydrological prediction to overcome the disadvantages of the single method. The model developed using a number of models for streamflow forecasting so as to improving the forecasting accuracy. Considering the differences of the runoff formation, the year is divided into several periods, then, the ensemble model for each periods is established. Due to the uncertainty of the precipitation, the Bayesian Model Averaging (BMA) is employed as the ensemble model for certain period, which can release both deterministic and probabilistic forecast. Take the Ertan hydropower station as the study case, results indicate the model proposed is superior to the single forecasting model and the probabilistic prediction is reliable which can support decision making.(2) To impove the waterpower utilization ratio, EPMS is employed in the medium-long term optimal operation for hydropower station. For quantifing the power generation risk due to the flow prediction error, a noval method called fuzzy risk analysis is developed. The method proposed includes: monthly streamflow recursive prediction using the EPMS; Optimal operation model with the object of maximum the annual power generation, which uses the prediction of different runoff description provided by EPMS; Fuzzy risk analysis method for the power generation. The method is applied to the Ertan hydropower station located in the Yalong river, the operation results for the typical years indicate the effectiveness and efficiency of the model, and the fuzzy risk analysis method can describe the risk more reasonabily than commom method.(3) Aiming at the uncertainty of the Xinanjiang hydrological model parameters selection, a noval two-stage parameters calibration method is presented. In the first stage, in order to analyzed the sensitivity of the objective weights which will lead to bottleneck when employs the stochastic simulation method, the MultiObjective Shuffled Complex Evolution Metropolis (MOSCEM-UA) algorithm is employed for searching the non-dominated solutions, thus the computational complexity can be reduced effectively, further, the satisfactory solution is determined using the sorting characteristic value considering the ranking reliability in the second stage. Furthermore, the influence of using different methods for calculating the mathematical weights and combination weights on the alternatives ranking is investigated. The study case demonstrates the proposed method can provide both satisfactory parameters and ranking reliability.(4) Predictive uncertainty analysis aiming at the parameters and input of the Xinanjiang model is presented. In this study, the predictive uncertainty analysis considering the model parameters and model input using the Shuffled Complex Evolution Metropolis (SCEM-UA) algorithm is proposed. Compared with the GLUE method, SCEM-UA algorithm doesn't require determine the value of the cut-off threshold for dividing the behavior and non-behavior parameter set, it only using the non-inferior parameters for flood simulation based on the appropriate converage of the algorithm. Taking the Huanren reservoir as an example, shows the model parameters and model input have the significant influence on the rainfall-runoff simulation, and the latter is larger than the former. Therefore, the predictive uncertainty should be fully considered during the real time flood forecasting.(5) Aiming at the equifinality of the hydrological model, a noval ensemble forcasting method based on the flood fuzzy classification is introduced for improving the predictive accuracy. Based on the fuzzy classification, parameters of the different flood categories are calibrated individually, then, the rainfall-runoff forecasting can be achieved using the "equifinality" parameters. First, the ensemble prediction is employed for the individual flood category, further, they are combined using the weighed averaging method. The case indicates the method proposed can improve the forecasting accuracy compared with the traditional forecast method which only use one set of parameters, and can increase reservoir discharge in advance which can effectively decrease the highest water level and flood control risk in the big flood regulation.Finally, a summary is given and some problems to be further studied are discussed.