On the improvement of real-time flood prediction model for dam operation

Abstract

In order to improve the flood forcasting accuracy for dam operation, we developed a real-time prediction system. The system is based on the filtering and prediction theory developed by Kalman and others. A lumped runoff model is transformed into a stochastic model by incorporating noise terms. The system takes account of the fact that the optimum values of the model parameters change from flood to flood. Three kinds of parameter are selected through the sensitivity analysis for many runoff data. Discharges are forcasted by use of prediction systems more than one, with different values of parameter respectively. Hence contributions to runoff is evaluated for each predicted discharge stochastically. Then most suitable discharge for flood runoff specified is determined by a weighting average method.