Comparing Artificial Neural Network and Regression Model to predict Sediment Trap Efficiency of Delayed Dams

Artificial neural network is a new method to estimate phenomena changes that have wide application in various branches of science. Reservoir sediment traps efficiency and sediment volume is an issue that can be investigated with this method. The purpose of this study is the determinate of sediment trapping efficiency in delayed dams and a comparison between neural network methods with regression models. So in this study, was used the physical model of delayed dams in the soil conservation and watershed management research center. In order to simulate a flood hydrograph, was used the idea of a linear reservoir model. Then with the release of flood and sediment trap sediments, examined the function of delayed dam. Then, with identify the affecting parameters on reservoirs sediment trapping; was developed neural network model on based back propagation of error. Also was applied the regression models for investigation the relationship between the parameters and compare the estimation results with observation. Finally, was applied the statistical indexes R2, RMSE and MAPE to assess the accuracy and precision of the model. According to results, the average values of R2, RMSE and MAPE in regression models are equal to 0.456, 26.6 and 62.1 respectively; but are the values in the neural network model 0.982, 4.6 and 6.1. So the artificial neural network has more ability in compared to regression equation to predict trap sediment in delayed dams. Also the results showed estimation of sediment trapping efficiency has depend on the number of parameters in equation and must be determined the optimized equation on based number of parameters.