Reservoir inflow forecasting using artificial neural network

Hydrologic forecasting plays an ever increasing role in water resource management, as engineers are required to make component forecasts of natural inflows to reservoirs for numerous purposes. Resulting forecast techniques vary with the system purpose, physical characteristics, and availability of data. As most hydrological parameters are subjected to the uncertainty, a proper forecasting method is of interest of experts to overcome the uncertainty. This paper presented an Artificial Neural Network (ANN) approach for forecasting of long term reservoir inflow using monthly inflow available data. A Levenberg-Marquardt Back Propagation (LMBP) algorithm has been used to develop the ANN models. In developing the ANN models, different networks with different numbers of neuron hidden layers were evaluated. A total of 21 years of historical data were used to train and test the networks. The optimum ANN network with 4 inputs, 5 neurons in hidden layer and one output was selected. To evaluate the accuracy of the proposed model, the Mean Squared Error (MSE) and the Correlation Coefficient (CC) were employed. The network was trained and converged at MSE = 0.0188 by using training data subjected to early stopping approach. The network could forecast the testing data set with the accuracy of MSE = 0.0283. Training and testing process showed the correlation coefficient of 0.7282 and 0.7228 respectively.