Investigation and Comparison Performance Three Data Mining Techniques ANN, GEP and RF in Estimating Flow Discharge Coefficients of Trapezoidal Broad-Crested Weirs

Accurate determination of the discharge coefficients plays an important role in estimating the discharge of the weirs. As a result, it is significant to estimate the discharge coefficients correctly. The purpose of this study is simulation and estimation the discharge coefficients of the broad-crested weirs. Hence, numerical simulation of hydraulic characteristics of these weirs were performed by ANSYS FLUENT software and results were obtained. Then, three intelligent models of ANN (MLP), GEP and RF were used to determine the discharge coefficients and the results of these models were compared. Assessment of the results were carried out using the statistical indices as R2 , RMSE, KGE, and graphical diagrams. The amounts of R2 , RMSE and KGE for ANN (MLP) were 0.906, 0.016, 0.927, for GEP 0.790, 0.025, and 0.780, and for RF 0.898, 0.013, and 0.841 were obtained, respectively. Due to the statistical criteria results and the range of RE% (∓5%), the ANN (MLP) was selected as the superior model. Violin plot illustrates a great and close agreement between the probability distribution of estimated data with ANN (MLP) and the obtained data from the numerical simulation using the finite volume method. Furthermore, the probability density function plot of residuals in ANN (MLP) model was closer to the normal distribution function, compared with the other models in this study.