An artificial neural network to predict river flow rate into a dam for a hydro-power plant

This paper describes a modified perceptron type of an artificial neural network to predict the river flow rate following a spell of rainfall. The neural network system comprises two subsystems: a linear-type subsystem and a perceptron-type subsystem. The former subsystem has 11 input nodes corresponding to the rainfall amounts and the river flow rates which are directly connected to a single output node. The latter subsystem is a typical perceptron network with three layers. The output layer has a single node which is commonly used as an output node of each subsystem. The output from the system is the predicted river flow rate. A case study is carried out on a dam for a hydro-power plant located on the upper section of the Hida River in Central Japan. It is found that the proposed system saves computation time with no degradation of the prediction accuracy