Artificial Neural Network (ANN) has proved to be one of the most successful mathematical training techniques that can find the complicated nonlinear relation between the input and output of the model. As the ANN uses the trial and error method between its neurons to modify the amount of error it may be trapped in local minimums that may differ from global minimums. Combination of ANN and GA has proved to partially solve this problem. In the current research, the daily flow for Bakhtiari River has been estimated using the combination of ANN and GA. Prediction of the daily flow at downstream has been done using the temperature, evaporation and rainfall data at upstream combined with the rainfall and flow data of the previous days at downstream. At the end, results of the combination of the ANN and GA using the training algorithm of levenberg, activate function of sigmoid and activate function of tangent hyperbolic. Results show that the combination of GA and ANN slightly improves the quality of the prediction over the solely use of ANN. Also the levenberg training algorithm can make better predictions comparing to conjugate gradient. Also the first pattern reduces the amount of produced error in comparison to sigmoid and tangent hyperbolic functions.
