Geographia Technica, Vol 15, Issue no.1/2020, pp. 162-172

CALCULATION OF THE RIVER FLOW WITH DIFFERENT PROBABILITIES OF OCCURRENCE USING ARTIFICIAL NEURAL NETWORK

Ioan Florin MOLDOVAN

DOI: 10.21163/GT_2020.151.15

ABSTRACT: The aim of this paper is to calculate the river flow with different necessary probabilities of occurrence, by the Artificial Neural Network (ANN) method. The studied ANN uses a radial basis function (RBF) architecture, with an input layer, a hidden layer and an output layer. Three series of maximum annual flow were used for the input layer. The series of maximum annual flow were collected from gauge stations, situated on three tributaries of the Mures river, Romania. The number of the neurons in the input layer is variable, according to the number of the flow values in the series. It means that the architecture of the ANN is different from one series to another. Consequently, the ANN has to repeat the calculation stages for each series. The results revealed that the ANN interpolates very well in the range of the values of the series. Each time, the stop condition of the target error was met. A comparison was made between the values of the flow calculated by the ANN method and by the statistical (classical) method using the Pearson type 3 distribution. The comparison showed that there are significant differences for the flow values corresponding to the 0.01% and 0.1% probabilities, and less or not significant for the rest.


Keywords: River flow, Probabilities of occurrence, Artificial Neural Network (ANN), Romania.

Full article here