Estimating Discharge Capacity of Flumes with Converging Triangular Walls Using Machine Learning

Accurate prediction of flow discharge in non-standard and unconventional flume geometries is vital for water resources management and hydraulic design. This study develops a new machine learning framework to estimate the dimensionless discharge coefficient (Q*) in flumes with converging triangular walls. Firstly, we considered six dimensionless geometric and flow parameters. Then, correlation analysis identified four dominant predictors, which were used as model inputs. Two ensemble learning algorithms, Random Forest (RF) and Extreme Gradient Boosting (XGBoost), were applied to a comprehensive experimental dataset. We used multiple performance metrics, including the Pearson correlation coefficient, coefficient of determination (R²), mean squared error (MSE), root mean squared error (RMSE), fourth root of mean quadrupled error (R۴MS۴E), and mean absolute error (MAE), to evaluate both models. Both models achieved high accuracies. In the test phase, XGBoost achieved an RMSE of ۰.۰۱۴۰ and an R² value of ۰.۹۸۷۲. Selecting dimensionless parameters as inputs and the output ensures the generalizability of the models across various flume and flow scales. Results have concluded the high effectiveness of both models for modeling complex hydraulic behavior in water resources engineering.

کلیدواژه ها:

Flume hydraulics ، Correlation analysis ، Random Forest ، XGBoost ، Dimensionless discharge