Prediction of Suspended Sediment Concentration in River Flow Using Artificial Neural Networks

Building a numerical model in not an easy task. In addition to its need for verification, it requires special skills and experiences in mathematics, computer programming and may take long period to be ready for operation and prediction. On the other hand, artificial neural network ANN) prediction models are more efficient in predictions once they are trained from examples or patterns. These types of ANN models need large amount of data which should be at hand before thinking to develop such models. In this paper, the capability of ANN model to predict suspended sediment in 2-D flow field is investigated. The data used for training the network are generated from a pre-verified 2-D hydrodynamic and a 2-D suspended sediment models which were recently developed by the authors. About two-thirds of the data are used for training the network while the rest of the data are used for validating and testing the developed ANN model. Field data measured by hydraulic research Institute are used to compare the results of the ANN model. A tansh activation function is used at the hidden layer which consisted of 5-10-1. The conjugate gradient learning algorithm is adopted. The results of the developed ANN model proved that the technique is reliable in such field compared to both the results of the previously developed models and the field data provided that the trained network is used to generate prediction within the range of training data.