نوع مقاله : مقاله پژوهشی
نویسندگان
Department of Civil Engineering, Na.C., Islamic Azad University, Najafabad, Iran.
10.22034/nawee.2026.585562.1232
چکیده
Objective: The accurate prediction of the discharge coefficient (e hydraulic design of labyrinth weirs. While existing literature extensively covers conventional vertical-sidewall geometries, there is a critical knowledge gap regarding the synergistic effect of sloped sidewalls and arced planform geometry on discharge capacity. This study addresses this limitation by investigating and predicting Cd for arced labyrinth weirs with varying sidewall slopes.
Methods: Using a comprehensive dataset of 277 experimental runs from a laboratory flume, five dimensionless parameters—namely sional analysis to capture both geometric and hydraulic complexity- were used as inputs for three soft computing approaches: Support Vector Machine (SVM), Gene Expression Programming (GEP), and Artificial Neural Network (ANN).
Results: All three data‑driven approaches produced satisfactory predictions of Cd. However, the ANN model based on a multilayer perceptron (MLP) demonstrated the best performance. The optimal architecture was MLP 5‑9‑1 trained using the BFGS algorithm, with the SOS error function, a logistic activation function in the hidden layer, and an identity activation function in the output layer. During training and validation, the performance metrics (RMSE, MAE, R², Cd(DDR(max))) were (0.022, 0.0145, 0.8402, 11.436) and (0.0199, 0.0145, 0.8961, 12.805), respectively. Graphical analyses also showed strong agreement between observed and predicted Cd values across different dataset quartiles.
Conclusions: The findings confirm that the MLP framework provides a robust and reliable alternative for hydraulic engineers, filling the current modeling void for non-conventional labyrinth weirs. This approach offers a practical tool for spillway optimization where complex geometric interactions preclude the use of simplified linear regressions.
کلیدواژهها
