A prediction of river flow rate into a dam for a hydro power plant by artificial neural network trained with data classified according to total amount of rain

A forecasting system for the time variation of a river flow rate following a spell of rainfall is developed by using a perceptron-type neural network. Rainfalls are classified as light or heavy according to total amount of precipitation ; here 200 mm is adopted as the critical value. As long as the amount of the rainfall is less than 200 mm as measured accumulatively, the dynamic river flow rate is forecasted by using a neural network trained with the data of the rainfall history belonging to the light rainfall group. When the accumulated amount of rainfall exceeds 200 mm, it is proposed that this system should be replaced by a neural network trained with the data of the heavy rainfall group. This dual system brings about an improvement in the forecasting accuracy for the rainfall period.