Comparative modeling of output discharge from rockfill dam by artificial neural network and numerical method

Flood is one of the most important natural disasters that threats human life and wealth. Application of rockfill dams is one of the low-cost methods for flood controlling. The use of these dams is resulted in higher peak discharge and shorter time in output flood hydrograph than those in input hydrograph. Artificial neural network (ANN) is one of the methods which can predict complex and non-linear processes at desirable level of accuracy. However، the accuracy of its prediction depends on the type of used learning algorithm and threshold function. In this study for estimation of flow through rockfill dam based on experimental data، multilayer perceptron model with different learning algorithms and threshold functions was evaluated. Afterwards، the output discharge values predicted by ANN method were compared with the values computed by two-dimensional numerical model. The results showed that the Multilayer perceptron model using Delta-Bar-Delta learning algorithm and Tanh threshold function with mean square error equal to 0. 00011 predicted the output discharge from rockfill dam with high accuracy. The ANN method with a R2 of 0. 962 performed as good as the numerical model (R2 =0. 984) for estimation of the mentioned parameter. Therefore، the drawback of the time-consuming and complex numerical methods analysis in estimating the output discharge of the dams can be overcome by using artificial neural network.