Modeling Crump Weir Discharge Coefficient Using Machine Learning Methods

Crump weir is classified as a short-edged weir. The upstream slope of this weir is greater than its downstream slope, which allows easy discharging of the sediments. In this research, the performance of k-nearest neighborhood and support vector regression (SVR) methods were investigated for modelling crump weir discharge coefficient using experimental data. 174 data sets in 9 combinations of the input parameters including the upstream and downstream slopes (Sup, Sdo), Reynolds number (Re) and water head in upstream to weir height ratio (h1/P) were used for modelling the discharge coefficient. The training was done in four stages using 66, 70, 75 and 80 percent of experimental data and the rest of these data at each stage were applied for the test phase. According to the results, the highest accuracy for the both applied models was obtained using 80% of the data in the training and the rest 20% in the test phases. Also, this investigation showed that the nearest neighborhood method presented a more accurate result than SVR method. Furthermore, water head in upstream to weir head height ratio (h1/P) had a significant role in modeling crump weir discharge coefficient. This ratio was the only parameter which could be used for predicting the coefficient accurately. Finally, this work showed that input combination including h1/P, Sup, Sdo parameters gave the best outcome. Both the Nearest Neighborhood and Support Vector Regression methods with coefficient of determination values of0.987 and 0.969 respectively, provided accurate predictions.