Optimization of Artificial Neural Network Structure in Prediction of Sediment Discharge Using Taguchi Method

In recent decades, artificial intelligence development as a powerful tool has made a tremendous progress in prediction and estimation of hydrological events. Although using artificial neural networks allows estimating rivers' suspended sediment load with appropriate accuracy and speed, predictive accuracy of these models is greatly influenced by the knowledge and understanding of ANN by the user. In natural resources and in especial hydrology and sediment studies, the optimization of ANN structure has not been adequately investigated; and the typical method is test and error. In this study, Taguchi optimization method was used to detect the best ANN structure in prediction of suspended sediment load of Neka River. Four important factors in the structure of artificial neural networks include the number of neurons in the first layer, the number of neurons in the second layer, the training algorithm and the transfer function; which are effective factors on estimation of ANN output and were considered in three different levels in the Taguchi experiments design. The results showed that the structure of 3 neurons in the first layer (level 3), 7 neurons in the second layer (level 2), the Levenberg-Marquarate training algorithm (level 3) and the PURELINE transfer function (level 2) has detected as the optimal ANN structure. The optimal ANN structure is able to estimate the sediment discharge of Nekaroud with high accuracy.