Abstract
It is important to forecast the water flow into the dam in order to operate gates safely. The water flow in the river is a complex natural phenomenon, therefore it is difficult to make the high performance forecasting system by conventional mathematical methods.
We have developed a high performance water flow forecasting system into the dam using neural networks and fuzzy theory. To improve forecasting performance and to decide the structure of the forecasting system, we propose two new methods. One is to decide the structure of neural networks and fuzzy parameters by the relation to the data. The other is to use two neural networks to improve forecasting performance when the river condition changes. One neural network is for normal river condition, and the other one is for flood condition. Forecasts obtained with two neural networks are combined by fuzzy theory which infers river condition. The forecasting results show the effectiveness of the proposed methods.
