Utilization of ANN technique to estimate the discharge coefficient for trapezoidal weir-gate

Abstract

Abstract Developing and assessing the performance of water projects and irrigation networks is based on many factors, such as flow measurement. Weir and gate structures have been extensively utilized for flow measurement and to get rid of sediments. The process of modeling and estimating the coefficient of discharge in the weir is an essential part of hydraulic engineering. Recently the application of computer skills was adopted instead of traditional methods. In the present study, the artificial neural network (ANN) was adopted to estimate the coefficient of discharge for a combined weir that consisted of the trapezoidal weir and rectangular gate. For this purpose, the experimental data were collected and analyzed. The dimensional analysis was used to identify the effective dimensionless parameters related to the discharge coefficient. The developed ANN network structure was designed as 6-10-1 and adopted the default scaled conjugate gradient algorithm for training using SPSS V24 software. It was found that the proposed model with ten neurons was highly accurate in predicting the discharge coefficient. The sensitivity analysis was adopted to assess the performance of the ANN using different numbers of effective input parameters. Assessing five models, the ratio of upstream head to gate height (H/d), slope of trapezoidal angle (tan θ), and the ratio of distance between weir and gate and gate height (y/d) parameters are adequate for estimating the discharge coefficient compared to other parameters. ANN model with input parameters of H/d, h/d (h is the flow depth over the trapezoidal weir), b g/d (b g is the gate width), tan θ, b g/b (b is the total width of flume), and y/d shows reasonable accuracy with acceptable statistical indicators, coefficient of determination (R 2 = 0.87), relative error (RE = 0.096), and mean squared error (MSE = 1.86) for the discharge coefficient. The ANN model gave a good idea about which factors are more effective on the discharge coefficient, and the process of training the network is more accessible than the traditional method which represent the discharge coefficient by equation.