Document Type : Research Paper
Authors
1
M.Sc. Student, Department of Civil Engineering, Ahvaz Branch, Islamic Azad University, Ahvaz, Iran.
2
Associate professor,, Department of Water Sciences and Engineering, Ahvaz Branch, Islamic Azad University, Ahvaz, Iran.
3
Assistant professor, Department of Water Sciences and Engineering, Ahvaz Branch, Islamic Azad University, Ahvaz, Iran.
Abstract
Morning glory spillways play a critical role in water flow management in dams and reservoirs, influenced significantly by the discharge coefficient. This coefficient determines the efficiency and risk of spillway performance under flood conditions. In this study, using 80 experimental datasets collected from two morning glory spillway inlet sections with square and circular zigzag shapes (featuring 4, 8, and 12 zigzags), two machine learning models—Support Vector Machine (SVM) and Gene Expression Programming (GEP)—were applied to simulate the discharge coefficient. Independent variables included the number of zigzags (n), Froude number (Fr), relative water head (H/P), and spillway shape index (R/D). Performance metrics (RMSE, MAE, R²) were employed to evaluate the accuracy of the models. Among various SVM models, the RBF kernel with γ = 0.1 yielded the most optimal results. The training and testing phases for the circular spillway showed (RMSE, MAE, R²) values of (0.9262, 0.0696, 0.0848) and (0.9820, 0.0346, 0.0398), respectively, while for the square spillway, these values were (0.9707, 0.073, 0.0904) and (0.9334, 0.0676, 0.0787). The GEP model demonstrated superior performance, particularly for the circular spillway with three genes, a head size of 9, and 45 chromosomes, yielding (RMSE, MAE, R²) values of (0.9778, 0.0375, 0.0451) and (0.9811, 0.0315, 0.0396) in the training and testing phases, respectively. For the square section, the GEP model with 55 chromosomes achieved (RMSE, MAE, R²) values of (0.9741, 0.0494, 0.0597) and (0.9591, 0.0503, 0.0594) for training and testing, respectively.
Keywords
Main Subjects
)