Performance analysis of different ANN modelling techniques in discharge prediction of circular side orifice

A side orifice is a mechanism integrated into one or both side walls of a canal to redirect or release water from the main channel, and it has numerous applications in environmental engineering and irrigation. This research paper evaluates different artificial neural network (ANN) modeling algorithms for the estimation of discharge of a circular side orifice in open channels under free flow conditions. Four training algorithm were compared, namely, Gradient Descent (ANN-GD), Levenberg–Marquardt (ANN-LM), Gradient-Descent with Momentum (GDM), and Gradient-Descent with Adaptive Learning (GDA). Among all the models developed for discharge prediction through a circular side orifice, the ANN-LM model, which employed the LM algorithm for optimization during the backpropagation process, had the best performance during both training and testing. The AARE, R, E, and RMSE values were 3.13, 0.9994, 0.9987, and 0.0005, respectively, during training and 4.43, 0.9976, 0.9952, and 0.0010, respectively, during testing. The predicted discharge from the ANN-LM model was compared to the discharge equation proposed in the literature, and the comparison revealed that the ANN-LM model reduced the error in predicted discharge by 50%.