Application of artificial neural network and genetic programming models for estimating the longitudinal velocity field in open channel junctions

Estimating the accurate longitudinal velocity fields in an open channel junction has a great impact on hydraulic structures such as irrigation and drainage channels, river systems and sewer networks. In this study, Genetic Programming (GP) and Multi-Layer Perceptron Artificial Neural Network (MLP-ANN) were modeled and compared to find an analytical formulation that could present a continuous spatial description of velocity in open channel junction by using discrete information of laboratory measurements. Three direction coordinates of each point of the fluid flow and discharge ratio of main to tributary channel were used as inputs to the GP and ANN models. The training and testing of the models were performed according to the published experimental data from the related literature. To find the accurate prediction ability of GP and ANN models in cases with minor training dataset, the models were compared with various percents of allocated data to train dataset. New formulations were obtained from GP and ANN models that can be applied for practical longitudinal velocity field prediction in an open channel junction. The results showed that ANN model by Root Mean Squared Error (RMSE) of 0.068 performs better than GP model by RMSE of 0.162, and that ANN can model the longitudinal velocity field with small population of train dataset with high accuracy.