Predicting Discharge Capacity in Semi-Cylindrical Side-Contracted Flumes Using MLPNN and GMDH

نوع مقاله : مقاله پژوهشی

نویسندگان

1 Department of Water Structures, Faculty of Water and Environmental Engineering, Shahid Chamran University of Ahvaz, Ahvaz, Iran

2 Associate Professor, Department of Water Structures, Faculty of Water and Environmental Engineering, Shahid Chamran University of Ahvaz, Ahvaz, Iran

10.22034/nawee.2026.574888.1194
چکیده
Objective: The primary objective of this study is to develop and evaluate two soft‑computing models (MLPNN and GMDH) for accurately predicting the discharge capacity of Semi‑Cylindrical Side-Contracted Flumes (SMBF) under both free‑flow and submerged regimes. The research aims to identify the optimal model structure, determine the most influential dimensionless parameters via dimensional analysis, and perform a comprehensive sensitivity analysis. A systematic comparison between the black‑box MLPNN and the interpretable GMDH model is conducted to recommend the most reliable and accurate predictive tool.
Methods: The study employed two data‑driven soft‑computing models: a Multilayer Perceptron Neural Network (MLPNN) and the Group Method of Data Handling (GMDH). First, dimensional analysis using the Buckingham Π theorem was applied to derive the key dimensionless parameters (e.g., h/w, w/B, ht/hs) as model inputs. Experimental data (450 free‑flow and 170 submerged records) were collected from published sources and randomly split into training and testing sets (70‑30 ratio). Model performance was evaluated using statistical indices (R², RMSE), and sensitivity analysis was conducted to identify the most influential input parameters on discharge prediction.
Results: The MLPNN model achieved outstanding accuracy with R² = 0.987 (free‑flow) and R² = 0.992 (submerged), and RMSE ≈ 0.001 for both regimes, significantly outperforming the GMDH model (R² ≈ 0.88–0.96). Sensitivity analysis revealed that the depth‑to‑throat ratio (h/w) is the most influential parameter under free‑flow conditions, while the contraction ratio (w/B) dominates under submerged flow. The GMDH model, though less accurate, provided superior interpretability due to its transparent, self‑organizing polynomial structure. Overall, MLPNN is recommended for high‑precision applications, whereas GMDH serves as a robust alternative when model transparency is prioritized.
Conclusions: Both MLPNN and GMDH models demonstrated high capability in predicting SMBF flume discharge, with MLPNN showing superior accuracy (R² > 0.98) for both free and submerged flows, while GMDH offered better interpretability. Sensitivity analysis confirmed that h/w governs free‑flow discharge, whereas w/B dominates under submerged conditions due to backwater effects. The study concludes that MLPNN is recommended for high‑precision field applications, while GMDH serves as a transparent alternative when understanding parameter relationships is essential.

کلیدواژه‌ها


عنوان مقاله English

Estimation of Discharge Capacity in Rectangular Flumes Containing Side Semi-Cylinders Using Data-Driven Soft Computing Models

نویسندگان English

Abbas Parsaie 1
Farid Ahmadi 1
Mohammad Bahrami-Yarahmadi 2
1 Department of Water Structures, Faculty of Water and Environmental Engineering, Shahid Chamran University of Ahvaz, Ahvaz, Iran
2 Associate Professor, Department of Water Structures, Faculty of Water and Environmental Engineering, Shahid Chamran University of Ahvaz, Ahvaz, Iran
چکیده English

This study investigates the discharge capacity of Semi-cylindrical contracted flumes installed in rectangular channels under both free-flow and submerged regimes. Two data-driven models, namely the Multilayer Perceptron Neural Network (MLPNN) and the Group Method of Data Handling (GMDH), are developed and compared. After dimensional analysis, influential dimensionless parameters were identified and used as model inputs. The MLPNN model with a single hidden layer of five neurons and tangent sigmoid activation function achieved outstanding accuracy for both free-flow (R² = 0.987, RMSE = 0.001) and submerged conditions (R² = 0.992, RMSE = 0.001). The GMDH model, while slightly less accurate (R² = 0.88 for free-flow and 0.96 for submerged), offered superior interpretability due to its transparent, self-organizing structure. Sensitivity analysis revealed that the depth-to-throat ratio (h/w) is the most influential parameter in free-flow, whereas the contraction ratio (w/B) dominates in submerged conditions. The MLPNN model is recommended for high-precision applications, whereas GMDH serves as a robust alternative when model interpretability is prioritized.

کلیدواژه‌ها English

SMBF flume
Soft Computing
Artificial Neural Network (MLPNN)
Group Method of Data Handling (GMDH)
Free and Submerged Flow
Sensitivity Analysis