Machine learning-based modeling of discharge coefficients in labyrinth sluice gates

Abstract

A labyrinth sluice gate is a novel design capable of transferring high flow compared to a conventional one owing to its leaf nonlinearity in plane form. From a novel point of view, this content investigates the impact of leaf configuration on the discharge coefficient of a modern design of labyrinth sluice gates, a parameter that is challenging to estimate precisely due to their complex structure by adopting seven diverse artificial intelligent models comprising gradient Boosting model GBM, KNeighbors Regression(KNN), Huber Regression (HR), Support Vector Regression (SVR) Linear and Radial Basis Functions (RBF), Random Forest (RF), and Linear Regression (LR).Experiments conducted by the literature were employed to extract different dimensionless parameters, including water depth contraction ratio H/G, orientation ratio l/L, cycles ratio 1/N, and Froude Number F_r as independent variables. The results indicated that the gradient Boosting model GB performed the best, with the highest coefficient of determination (R² of 99.74 %) and mean absolute percentage error (MAPE of 0. 3872 %). Consequently, the main contribution of this study is to introduce a robust machine learning tool that can be depended on to estimate the discharge coefficient of labyrinth sluice gates confidently. Furthermore, it not only deduces machine learning as a solution to a persistent hydraulic challenge but also provides a valuable template for integrating data-driven approaches into future gate design and optimization.