A study of five types of ANN-based approaches to predict discharge coefficient of combined weir-gate

Since many years ago, flow measurement has become a fundamental issue in hydraulic engineering. One of the conventional methods of flow measurement is the use of combined structures. In this regard, using a combined structure, including a gate and a weir, is one of the approaches that has attracted the attention of researchers in this field. Therefore, in this research, five different methods based on artificial neural networks were used to predict the discharge coefficient. The networks architecture includes an input layer with four neurons, a hidden layer with seven neurons, and an output layer with one neuron. be mentioned that the number of neurons within the hidden layer is set to 4 only for the recurrent network. For the hidden layer, the logarithmic sigmoid activation function was used. Also, the linear activation function was used for the output layer. Finally, the results showed that the Levenberg-Marquardt (LM) algorithm performs better than the other methods. The convergence speed of this algorithm, which also uses the second derivative, is much higher than others. In this case, the coefficient of determination (R^2) for the training and the test stage was equal to 0.92616 and 0.94079, respectively. In addition to, the first type of rough model with the gradient descent training algorithm also had an acceptable performance and was placed in second place. Also, the sensitivity analysis on the dimensionless parameters affecting this issue showed that the H⁄d, y⁄d, b⁄B, and b⁄d parameters have maximum to minimum effect on the model results, respectively.