Application And Research On Robust Theory In Flood Forecasting

The theory of robust estimation is introduced into the real-time flood forecasting system. The robust system can prevent abnormal factors from entering the flood system, so as to ensure the stability of the system and the accuracy of flood forecast. Firstly, the theory of robust estimation is introduced into parameter estimation of the auto-regressive model (AR model). Also some estimation methods commonly used, including the Huber estimation and IGG estimation, are introduced and compared with the Least Square Method (LSM). Examples show that, with normal observed data, the results obtained by the three methods are satisfactory, however, if the observed data are abnormal, the estimated results by LSM is not satisfactory, while the results from the other two methods are still satisfactory. In addition, the characteristic of outlier in the reservoir inflow is analyzed and the theory of robust estimation is introduced into the forecast of reservoir inflow and real-time error correction of it. The robust system can prevent outlier and extreme error from influencing correction efficiency, so as to improve the stability of correction results. Robust mean square error is put forward to compare validity of the three estimation methods. Results show robustified estimation methods are effective to the prevention of outlier. In the reservoirs with big fluctuation of inflow, the difference between LSM and robustified estimation methods (including Huber estimation and IGG estimation) is large and reduction extent of robust mean square error is large, and vice versa. Furthermore, from the viewpoint of hydrological application, the types and characteristic of rainfall observation error in hydrological remote system are analyzed and a three-stepwise weight function correction method of the error in the system is developed on the basis of theory of robustified Least Square Method. Four robust schemes are proposed by consideration of the problem that if the mean areal precipitation is classified and the different format of descending weighted function. Rainfall data of 28 floods between 1974 and 1999 in Wuxigou Basin of Weishui Reservoir are analyzed and calculated and the weight function of every step are obtained. The method is tested using Xin'anjiang model to calculate the discharge hydrograph. Results show that a three-stepwise weight function correction method is effective and applicable and can be spread in other basins. It can eliminate the outlier of rainfall data and improve the accuracy of flood forecast. Finally, Robustified ridge estimation that has been widely used in surveying and mapping is suggested and applied to the estimation of unit hydrograph. The results show that the minus value and fluctuation of unit hydrograph can be mitigated and the error of observed discharge can be withstood using the robustified ridge estimation.