Predicting discharge coefficient of triangular side orifice using ANN and GEP models

Abstract

ABSTRACT This study utilized machine learning models to predict the discharge coefficient for a sharp-crested triangular side orifice (TSO). The chosen models were the Artificial Neural Network (ANN) and Gene Expression Programming (GEP). Development of the models was based on 570 experimental datasets, with 70% allocated for training and the remaining 30% for testing. Five nondimensional parameters were utilized as inputs for the models, including TSO’s crest height to its height (W * =W/H), main channel width to TSO’s base length (L*=B/L), main channel width to TSO’s height (H * =B/H), upstream flow depth to the TSO’s height (Y * =y 1/H), and upstream Froude number of the main channel (F r ). While the discharge coefficient (C d ) was defined as the output. Then, the developed models were evaluated by three performance metrics, violin boxplots, and Taylor diagrams to ensure their reliability and accuracy. Furthermore, a sensitivity analysis was conducted to indicate the most effective parameter affecting the C d value. The findings revealed that both models predicted very well compared to the actual values, with the ANN model emerging as the most reliable predictor. It exhibited the highest determination coefficient (R 2 ), nearing 1, along with the lowest Mean-Square-Error (MSE) and Mean-Absolute-Error (MAE) values, both close to zero. The sensitivity analysis highlighted that the orifice crest height and Froude number significantly impacted the C d value, contributing to more than 36%. In addition, the predicted discharge coefficient stayed within the range of ± 5.0% of the experimental values. Finally, the developed models demonstrated a high level of equivalence compared to previous studies, especially the ANN model. Therefore, these models are recommended as accurate, robust, and rapid tools to predict the TSO’s discharge coefficient.