Application of artificial neural networks for disaggregation of river seasonal inflow series

Disaggregation models divide annual or seasonal flows among sub-periods and divide aggregate flow acrosssub basins. The objective of this study was to apply of disaggregation model in the temporal domain. Feed forward neural network with back propagation algorithm was used to disaggregate the seasonal inflow data to monthly series for Mahi River. Hyperbolic tangent function and identity function were used as transfer functions in the hidden and output layers, respectively. Learning rate and momentum were taken as 0.1 and 0.9. Seventy percent data was used in training set and the remaining 30% in the validation set. AllANN networks were trained to minimise the Mean Squared Error (MSE). Amongst architectures tried, the optimum architecture was found to be ANN (1-4-5) for inflow series disaggregation. The average relative error for the training set and testing set was 0.707 and 0.697. respectively. Low RMSE, AIC, SIC and R² was obtained for all the months for the selected architecture. The correlation coefficient between the observed and computed series for June, July, August September and October was estimated to be 0.68, 0.68, 0.73, 0.71 and 0.65, respectively.